CN121745577A - Resource scheduling method and apparatus based on industrial topology network - Google Patents

Abstract

The embodiment of the disclosure discloses a resource scheduling method and device based on an industrial topology network. The method comprises the steps of generating a corresponding operation information set, generating a corresponding abnormal score sequence set based on the operation information set, calling a parallel computing node cluster of a topology construction server to generate an industrial topology network representing a product circulation relation among production platforms, synchronously storing the abnormal score sequence set and the industrial topology network into a shared storage unit, executing the following generation steps through a parallel computing pipeline of a hardware accelerator, namely generating comprehensive abnormal information of a target production platform, generating industrial chain breakage probability information, generating a corresponding resource scheduling instruction, and controlling a resource control terminal to execute scheduling operation on resources. The implementation mode can realize the accurate scheduling of the industrial chain resources, save the time resources and reduce the production loss caused by the interruption of the supply chain.

Description

Resource scheduling method and device based on industrial topology network

Technical Field

The embodiment of the disclosure relates to the technical field of computers, in particular to a resource scheduling method and device based on an industrial topology network.

Background

Currently, in the field of industrial chain risk early warning and emergency resource scheduling, the prior art mainly relies on an isolated monitoring system based on a single data source and a scheduling decision mechanism relying on manual experience.

However, when the resource scheduling is performed by adopting the method, the technical problems that the multi-source data fusion is difficult, the industrial topology network reflecting the full appearance of a plurality of production platforms cannot be quickly constructed, the early warning and the resource scheduling are further disjointed, the automatic closed loop from risk perception to intervention execution cannot be quickly realized, the response time of the resource scheduling is prolonged, and the production loss is caused are often caused.

The above information disclosed in this background section is only for enhancement of understanding of the background of the inventive concept and, therefore, may contain information that does not form the prior art that is already known to those of ordinary skill in the art in this country.

Disclosure of Invention

The disclosure is in part intended to introduce concepts in a simplified form that are further described below in the detailed description. The disclosure is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

Some embodiments of the present disclosure propose resource scheduling methods, apparatuses, electronic devices, and computer-readable media based on an industrial topology network to solve one or more of the technical problems mentioned in the background section above.

In a first aspect, some embodiments of the present disclosure provide resource scheduling based on an industrial topology network, including generating a corresponding set of operation information based on multisource heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to a target product in response to detecting that inventory of the target product is below a target threshold, generating a corresponding set of anomaly score sequences based on the set of operation information using an anomaly detection engine, calling a cluster of parallel computing nodes of a topology construction server based on the set of operation information to generate an industrial topology network characterizing a product circulation relationship between the production platforms, synchronously storing the set of anomaly score sequences and the industrial topology network to a shared storage unit for access by a hardware accelerator, wherein the shared storage unit is used for parallel access of zero copy data, generating comprehensive anomaly information of the target production platform based on the anomaly score sequences and the industrial topology network, generating industrial chain fracture probability information based on the comprehensive anomaly information to map to early warning level information, generating a corresponding scheduling command based on the early warning level information, and executing a scheduling command corresponding to a scheduling resource, wherein the scheduling command is received by a parallel computing pipeline of the hardware accelerator, and the scheduling command is used for controlling a resource, and the resource is controlled by a resource.

In a second aspect, some embodiments of the present disclosure provide a resource scheduling apparatus based on an industrial topology network, including a first generating unit configured to generate a corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to a target product in response to detecting that inventory of the target product is lower than a target threshold, a second generating unit configured to generate a corresponding abnormal score sequence set based on the operation information set using an anomaly detection engine, a calling unit configured to call a cluster of parallel computing nodes of a topology construction server based on the operation information set to generate an industrial topology network characterizing a product flow relationship between the production platforms, a storage unit configured to store the abnormal score sequence set and the industrial topology network synchronously to a shared storage unit for access by a hardware accelerator, wherein the shared storage unit is configured to be used for parallel access of zero copy data, an executing unit configured to generate a corresponding abnormal score sequence set based on the abnormal score sequence and the topology network, a calling unit configured to call a cluster of parallel computing nodes of a topology construction server based on the operation information set, a storage unit configured to store the abnormal score sequence set and the industrial topology network synchronously to a shared storage unit for access by the hardware accelerator, and an executing unit configured to generate a warning response to the integrated resource corresponding to a control order based on the abnormal control order of the industrial chain, wherein the integrated resource is configured to be stored as a relative to the control order.

In a third aspect, some embodiments of the present disclosure provide an electronic device comprising one or more processors, a storage device having one or more programs stored thereon, which when executed by the one or more processors, cause the one or more processors to implement a method as described in any of the implementations of the first aspect.

In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having a computer program stored thereon, wherein the program when executed by a processor implements a method as described in any of the implementations of the first aspect.

The method for scheduling the resources based on the industrial topology network has the advantages that automatic intervention (resource scheduling) of industrial chain breakage of the production platform is achieved through the method for scheduling the resources based on the industrial topology network, time resources are saved, and supply chain interruption loss is reduced. Specifically, the supply chain interruption loss is caused by serious data island, difficulty in correlating multisource heterogeneous original data and lack of risk propagation analysis from the view angle of an industrial chain, so that risk identification lag and insufficient breakpoint prediction are caused, and intervention measures cannot be timely adopted to cause supply chain interruption. Based on this, in some embodiments of the present disclosure, a resource scheduling method based on an industrial topology network first generates a corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to a target product in response to detecting that the inventory of the target product is lower than a target threshold. The multi-source heterogeneous original data is uniformly encoded into a standardized time sequence vector to form a continuous and measurable operation information set, and uniform and comparable basic data representation is provided for subsequent analysis. Then, using an anomaly detection engine, a corresponding anomaly score sequence set is generated based on the running information set. By generating the anomaly score sequence set, the state deviation degree of the production platform is accurately quantized, the early weak anomaly signal recognition sensitivity is improved, and visual quantization basis is provided for risk assessment. And then, based on the operation information set, calling a parallel computing node cluster of the topology construction server to generate an industrial topology network representing the product circulation relation among the production platforms. By constructing the industrial topology network, supply association among production platforms can be efficiently excavated, real-time monitoring of large-scale production platforms is supported, and an industrial chain structure is clearly presented. And secondly, synchronously storing the abnormal score sequence set and the industrial topology network to a shared storage unit for access by a hardware accelerator, wherein the shared storage unit is used for zero-copy data parallel access. The data copying cost between the CPU and the hardware accelerator is eliminated, the data consistency is ensured, and a low-delay and high-bandwidth data supply channel is provided for subsequent hardware acceleration calculation. And generating comprehensive anomaly information of the target production platform based on the anomaly score sequence set and the industrial topology network through the parallel computing pipeline of the hardware accelerator. The parallel computing assembly line improves the multi-source data fusion efficiency, integrates abnormal scores and topological association, realizes comprehensive evaluation of production platform risks, and avoids one-sided performance of isolated analysis. And generating the industrial chain breakage probability information based on the comprehensive abnormal information so as to map the industrial chain breakage probability information into early warning grade information. And the risk of industrial chain breakage is quantified, the risk degree is visually presented through the early warning level, so that quick response is facilitated, and the problem of response hysteresis caused by fuzzy risk level is solved. And then, based on the early warning level information, generating a corresponding resource scheduling instruction. The accurate matching of the abnormality and the scheduling strategy is realized, and the pertinence and the effectiveness of the scheduling instruction are ensured. And finally, controlling the resource control terminal to execute the scheduling operation of the resource in response to receiving the resource scheduling instruction, wherein the scheduling operation comprises the storage of the parts related to the target product and the transportation of the parts. And the storage and transportation scheduling of the quick floor parts are realized, the production resources are timely supplemented, the chain breakage probability of the industrial chain is reduced, the production continuity of the target product is ensured, the risk of quick response is ensured, and the chain breakage loss of the industrial chain is reduced.

Drawings

The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by reference to the following detailed description when taken in conjunction with the accompanying drawings. The same or similar reference numbers will be used throughout the drawings to refer to the same or like elements. It should be understood that the figures are schematic and that elements and components are not necessarily drawn to scale.

FIG. 1 is a flow chart of some embodiments of an industrial topology network-based resource scheduling method according to the present disclosure;

FIG. 2 is a schematic diagram of the architecture of some embodiments of an industrial topology network-based resource scheduling apparatus according to the present disclosure;

fig. 3 is a schematic structural diagram of an electronic device suitable for use in implementing some embodiments of the present disclosure.

Detailed Description

Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete. It should be understood that the drawings and embodiments of the present disclosure are for illustration purposes only and are not intended to limit the scope of the present disclosure.

It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings. Embodiments of the present disclosure and features of embodiments may be combined with each other without conflict.

It should be noted that the terms "first," "second," and the like in this disclosure are merely used to distinguish between different devices, modules, or units and are not used to define an order or interdependence of functions performed by the devices, modules, or units.

It should be noted that references to "one", "a plurality" and "a plurality" in this disclosure are intended to be illustrative rather than limiting, and those of ordinary skill in the art will appreciate that "one or more" is intended to be understood as "one or more" unless the context clearly indicates otherwise.

The names of messages or information interacted between the various devices in the embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

Referring to fig. 1, a flow 100 of some embodiments of an industrial topology network-based resource scheduling method according to the present disclosure is shown. The resource scheduling method based on the industrial topology network comprises the following steps:

Step 101, in response to detecting that the inventory of the target product is lower than the target threshold, generating a corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to the target product.

In some embodiments, the execution body (for example, the electronic device) of the resource scheduling method based on the industrial topology network may be hardware or software. When the computing device is hardware, the computing device may be implemented as a distributed cluster formed by a plurality of servers or terminal devices, or may be implemented as a single server or a single terminal device. When the computing device is embodied as software, it may be installed in the hardware devices listed above. It may be implemented as a plurality of software or software modules for providing distributed services, or as a single software or software module. The present invention is not particularly limited herein.

In other embodiments, the executing entity may generate the corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to the target product in response to detecting that the inventory of the target product is lower than a target threshold. The target product may be a specific product (including a plurality of parts and a plurality of production platforms for supply), for example, a new energy automobile, a mobile phone. The target threshold may be a preset threshold value of the target product inventory (for example, the target threshold of the new energy automobile may be 200). The multi-source heterogeneous raw data can be production platform operation data from different channels and different formats. For example, the multi-source heterogeneous raw data may include business change records (text), recruitment frequency (numerical value), public opinion comments (text). The production platform behavior track vector sequence set may be a set of quantized multidimensional state vectors in which the multisource heterogeneous raw data are arranged in time sequence. The operation information may be a vector formed by a plurality of standardized operation dimension values at a certain time point, and the vector is used for representing the comprehensive state of the production platform at a specific time point. For example, the operation information may be [ Industrial and commercial Change: 0, recruitment index: 0.7, logistic index: 0.5, invoice intensity: 0.8, public opinion risk: 0.2]. For example, the set of operational information may include { operational information for production platform A, operational information for production platform B, operational information for production platform C }, each sequence being a multi-dimensional vector arranged by date.

In some optional implementations of some embodiments, in response to detecting that the inventory of the target product is below the target threshold, the executing entity may generate the corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industry chain corresponding to the target product, and may include the following steps:

the first step, the following steps are executed for each production platform multi-source heterogeneous raw data in the multi-source heterogeneous raw data of the plurality of production platforms:

And step one, carrying out heterogeneous data time sequence isomorphism processing on the multi-source heterogeneous original data of the production platform to obtain isomorphic time sequence data. The heterogeneous data timing isomorphism may be a sequence format that uniformly processes different types of data into alignment and comparison on a time axis. The isomorphic time series data may be time series data obtained by converting the multi-source heterogeneous original data of the production platform through uniform time granularity and format, for example, converting the industrial and commercial change (event) into a 0/1 pulse sequence (daily dimension). In practice, first, event type data (e.g., business changes) and continuous type data (e.g., invoice amounts) are distinguished according to the type of multi-source heterogeneous raw data of the production platform. Then, for event type data, when the occurrence date is marked 1 and the no event date is marked 0, a pulse sequence is generated. For continuous data, the numerical sequence is generated by daily aggregation (e.g., summing, counting). Finally, the sequences are aligned with uniform time stamps (e.g., daily) and the missing days are marked to obtain the structured time series data.

And secondly, constructing a standard multidimensional track vector based on the isomorphic time sequence data. The standard multidimensional track vector may be a vector of a production platform after each dimension is normalized at a plurality of time points. In practice, first, for each point in time (e.g., daily), the values of the respective dimensions for that point in time are extracted from the respective isomorphic sequences of the respective production platforms. The values for each successive dimension are then normalized using the historical extremum of the production platform in the near future (e.g., the past 90 days) and scaled to the [0,1] interval. Finally, the dimensions of each production platform at each time point (event type is 0/1, continuous type is normalized value) are combined into a vector with fixed dimensions.

And thirdly, carrying out continuity maintaining processing on the standard multidimensional track vector so as to generate corresponding running information. In practice, first, it is determined whether there is a missing day for each daily vector sequence of production platforms. Then, for the missing day vector of each production platform, linear interpolation filling is performed by using the vectors of the adjacent days before and after the missing day vector. Finally, it is determined whether the vector sequence of each production platform after filling is continuous in time (without break points) and a first order difference quotient (approximately differentiable) can be calculated, thereby generating a complete production platform behavior trace vector sequence set.

And secondly, integrating each operation information into an operation information set.

Step 102, generating a corresponding anomaly score sequence set based on the running information set by utilizing an anomaly detection engine.

In some embodiments, the execution body may generate the corresponding anomaly score sequence set based on the running information set using an anomaly detection engine. Wherein the anomaly score sequence characterizes the degree of deviation of the state of the production platform. The anomaly detection engine may be a server for automatically analyzing the behavior track of the production platform and identifying an anomaly state. The anomaly detection engine may determine the degree of anomaly in production platform behavior by determining a set of operational information (e.g., quantitative information for production, energy consumption, orders) corresponding to a plurality of production platforms, and generate a continuous sequence of values. The anomaly score sequence set may be an anomaly quantified sequence combination of a plurality of production platforms. For example, the anomaly score sequence described above may be a set of daily anomaly scores for approximately 7 days for the A production platform. For example, the anomaly score sequence described above may be [0.1,0.15,0.3,0.5,0.8,0.9,0.95]. The deviation degree of the state of the production platform can be the difference between the current operation state of the production platform and a normal standard.

In some optional implementations of some embodiments, the executing entity may generate, using an anomaly detection engine, a corresponding anomaly score sequence set based on the running information set, and may include the steps of:

First, for each operation information of the operation information set, the following steps are performed:

And step one, generating a track curvature change sequence corresponding to the running information by using a first node of the abnormality detection engine. The first node may be a node in the anomaly detection engine that is specifically responsible for calculating the trajectory curvature (change acceleration). The track curvature change sequence may be a sequence for characterizing the severity (acceleration) of the change in the production platform behavior track at each point in time. For example, the track curvature change sequence may be a production platform track curvature sequence [0.01,0.02,0.05,0.12,0.3,0.6]. In practice, based on the running information, the first node of the anomaly detection engine is called, and the second derivative of the track corresponding to the running information is determined. Finally, integrating the derivative results to generate a track curvature change sequence.

And secondly, executing double-reference dynamic discrimination operation by using a second node of the anomaly detection engine based on the track curvature change sequence, a preset production platform self-history normal track reference and an industry synchronous average track reference so as to obtain double-reference deviation degree data. The second node may be a processing unit in the anomaly detection engine that exclusively performs the double-reference dynamic discrimination. The preset standard of the historical normal track of the production platform can refer to the behavior track of the production platform in the past normal operation period as a comparison standard. For example, the preset production platform own history normal track standard may be an average behavior track of the production platform B after removing the abnormal month in the past 12 months. The industry contemporaneous average track reference may refer to an average behavior track of the same industry production platform in the same period. The double-reference deviation degree data can refer to deviation values (self deviation degree and industry deviation degree) obtained by respectively comparing the current track of the production platform with the self history reference and the industry reference. In practice, firstly, the track curvature change sequence is compared with the historical normal curvature standard of the production platform by DTW (dynamic time warping) to determine the shape difference distance. Then, EMD (earth moving distance) comparison is carried out with an industry contemporaneous average curvature reference, and the distribution difference is determined. Finally, the two distance values are normalized to a deviation score to obtain double-reference deviation data.

And thirdly, generating an anomaly score sequence based on the double-reference deviation data and the self-adaptive weight parameters, wherein the self-adaptive weight parameters are dynamically determined by the anomaly detection engine according to the historical data integrity and the historical early warning accuracy of the production platform. The adaptive weight parameter may be a coefficient for dynamically adjusting the importance of deviation from industry. The above-mentioned production platform history data integrity may refer to the proportion of valid data in the production platform history data. The historical early warning accuracy may be a proportion of past early warning results that are verified to be correct. In practice, first, adaptive weight parameters (e.g., 0.7 (self-bias) and 0.3 (industry bias)) are determined by the anomaly detection engine based on the production platform historical data integrity (e.g., 92%) and the historical pre-warning accuracy (e.g., 88%). Then, an anomaly score sequence (for example, "anomaly score=0.7×self deviation degree+0.3×industry deviation degree") is generated.

Step 103, based on the operation information set, calling a parallel computing node cluster of the topology construction server to generate an industrial topology network representing the product circulation relation among the production platforms.

In some embodiments, the executing entity may invoke the parallel computing node cluster of the topology building server to generate an industry topology network that characterizes the product flow relationship between the production platforms based on the set of operational information. The topology construction server may be provided with a plurality of parallel nodes, and the topology construction server is used for mining a server of a production platform for providing association and constructing a network. The product circulation relationship between the production platforms can be an upstream and downstream supply and demand relationship between the production platforms, for example, a part manufacturer provides a directional cooperation relationship of accessories for a whole vehicle factory. The industrial topology network may be a network structure in which nodes represent production platforms and directed edges represent supply relationships.

And the first step is to call the first parallel computing node of the parallel computing node cluster to determine the event type track similarity of each two production platforms in the running information set so as to generate an event synchronization matrix. The first parallel computing node may be a node in the topology building server that determines an event type trajectory similarity. The event type may be a discrete state in which data is expressed as occurring or not occurring at a specific point of time, and is generally represented by 0 (not occurring) or 1 (occurring). The event type track similarity may be a degree of synchronization of the event type track of the production platform, for example, a degree of synchronization match between a commercial increase of the production platform and a recruitment peak of the production platform B. The event synchronization matrix may be a square matrix storing the event synchronization between every two of all production platforms. For example, the value of row i and column j in the event synchrony matrix represents the event synchrony of production platform i and production platform j.

As an example, first, a first parallel computing node of a topology building server determines event-type trajectories (e.g., business, public opinion pulse sequences) for each production platform. Then, a dynamic time warping distance of its event sequence under a certain allowed time offset (e.g., 0 to 7 days) is determined for each pair of production platforms, and the distances are converted into similarities. And finally, filling the similarity results of the production platform pairs into a matrix to generate an event synchronization matrix.

And secondly, calling a second parallel computing node of the parallel computing node cluster to determine the numerical correlation of continuous tracks of each two production platforms in the running information set so as to generate a numerical correlation matrix. The second parallel computing node may be a node in the topology construction server that determines a continuous track correlation. The numerical correlation of the continuous track may be a statistical correlation of the trend of continuous indicators (such as the material flow) of two production platforms. For example, the degree of positive correlation of daily logistics frequency sequences for two production platforms. The numerical correlation matrix may be a square matrix storing numerical correlations between every two of all production platforms. For example, each element in the numerical correlation matrix is a correlation coefficient (e.g., 0.9 represents a strong positive correlation). In practice, first, the continuous track (logistics frequency, recruitment index) of the production platform is extracted. The second parallel computing node then determines Pearson correlation coefficients for each two production platform sequences. Finally, a numerical correlation matrix is generated.

And thirdly, calling a third parallel computing node of the parallel computing node cluster to determine the time lag causal relationship of the operation information of each two production platforms in the operation information set so as to generate an optimal time lag matrix. The third parallel computing node may be a node dedicated to parallel computing time lag causal relationship in the topology building server. The time-lapse causal relationship may be the number of days and the degree of correlation that one production platform changes before another production platform changes. For example, the time lag causal relationship may be that the invoice amount of the production platform a changes, and the invoice amount of the production platform B changes similarly after 3 days, and the production platform a leads the production platform B for 3 days. The optimal time-lag matrix may be a square matrix storing time-lag days corresponding to the maximum correlation between the production platforms. The optimal time lag may refer to the number of days corresponding to the time when the behavior change of one production platform leads the change of another production platform and the correlation between the two is the maximum. In practice, first, the third parallel computing node acquires continuous trajectories of the respective production platforms. Then, for each pair of production platforms, the sequence of one production platform is shifted relative to the other sequence for 1 to 7 days, the correlation coefficients under different time lags are respectively determined, and the time lag days with the maximum correlation are recorded. And finally, filling the optimal time lags of the production platform pairs into the matrix to obtain the optimal time lags matrix.

And step four, carrying out weighted fusion on the event synchronicity matrix, the numerical correlation matrix and the optimal time-lag matrix to generate a comprehensive similarity matrix among production platforms. The integrated similarity matrix between the production platforms can be an overall correlation matrix fusing the event synchronization matrix, the numerical correlation matrix and the optimal time-lag matrix. In practice, first, weights (event synchronicity matrix 0.3, numerical correlation matrix 0.4, and optimal time lag matrix 0.3) of three types of matrices are set. And then, carrying out weighted summation on the event synchronicity, the numerical correlation and the optimal time lag matrix according to weights, and finally, obtaining a comprehensive similarity matrix among the production platforms.

And fifthly, carrying out weighted correction on the comprehensive similarity matrix among the production platforms by utilizing the pre-acquired importance scores of the production platforms so as to generate a coupling strength matrix among the production platforms. The importance score of the production platform can be a node importance index calculated in advance through the scale of the production platform, market share, network centrality and the like. For example, production platform A has a high GDP-to-area ratio and a tax rate of 0.9 (full 1). The coupling strength matrix between the production platforms can be a relation strength matrix subjected to weight correction of importance scores of the production platforms, and the side weights are more reasonable. In practice, first, each production platform importance score (e.g., production platform A score 0.9, production platform B score 0.8) is obtained. And then, correcting the corresponding comprehensive similarity by using the importance product of the production platform. And finally, generating a coupling strength matrix between the production platforms.

And sixthly, screening the coupling strength matrix among the production platforms to generate an initial industry chain directional network. The initial industry chain directed network may be a base network formed after screening the coupling strength threshold, for example, only the pair of directed edges of the production platform with the coupling strength >0.5 are reserved. In practice, first, a coupling strength threshold (e.g., 0.5) is set. Then, pairs of production platforms in the matrix with coupling strengths greater than the coupling strength threshold are screened. And finally, constructing an initial industrial chain directed network.

And seventh, noise filtering is carried out on the initial industrial chain directional network so as to generate an industrial topology network. In practice, first, a Louvain algorithm is performed on an initial industry-chain directed network to divide communities. Noise edges (e.g., pseudo-associated edges of production platform a and production platform B, which may be false network connections based on data coincidences rather than true business connections (e.g., business independent enterprises at the same industry park at both sites, resulting in "event synchronization" due to simultaneous response to regional power limit policies, misjudged by algorithms to be edges with supply chain associations)) are then deleted to obtain an industry topology network.

The above operation steps, as an invention point of the present disclosure, solve the technical problem that "the actual supply relationship cannot be efficiently and accurately automatically identified from the massive and dynamic behavior data of the production platform, resulting in the waste of computing resources and time resources. The technical problems are caused by the fact that the prior art cannot automatically and efficiently mine an accurate supply relation network with time sequence cause and effect from the behavior data of the multi-source production platform, relies on manual experience or static association, and is difficult to support large-scale and dynamic industrial chain risk analysis. The method and the system realize the automatic and efficient construction of the high-quality industrial topological network from the operation information set by utilizing the topological construction server to calculate the similarity of the operation information sets of a plurality of dimensions in parallel and fuse the similarity into the high-precision industrial chain network, thereby saving a great amount of time resources and calculation resources.

Step 104, synchronously storing the abnormal score sequence set and the industrial topology network to a shared storage unit for access by the hardware accelerator.

In some embodiments, the execution body may synchronize the set of anomaly score sequences and the industry topology network to a shared memory unit for access by a hardware accelerator, wherein the shared memory unit is used for zero-copy data parallel access. The shared storage unit can support the hardware/software architecture of the CPU and the hardware accelerator (such as GPU) to directly and seamlessly access the same physical or virtual memory, so that data copying is eliminated, and efficient data sharing is realized. The hardware accelerator described above may be a special purpose processor, e.g., GPU, FPGA, with massively parallel computing capabilities, dedicated to accelerating specific computing tasks (e.g., graph computation, matrix operations).

In the process of solving the technical problems in the background technology by adopting the technical scheme, aiming at the scene to be applied, namely the automatic production process of placing a large-scale automobile order (an order with strict time delivery requirement), the assembly height of each part is coordinated, the required parts relate to a plurality of production platforms, the production platforms with high chain breakage risk are required to be positioned quickly, mass production platform abnormal data and topology related data are required to be synchronized to a hardware accelerator within minutes, and further the emergency scheduling of the required parts is carried out in advance, the technical problems of low heterogeneous data transmission efficiency and high hardware access addressing delay are often accompanied, the risk early warning response is delayed, the resource scheduling response is delayed, and the chain breakage loss cannot be avoided timely. Aiming at the following requirements and characteristics of the application scene, namely low delay of data synchronization, zero redundancy of hardware access, heterogeneous data classified transmission and high concurrency of task execution, the following solution is adopted:

In some optional implementations of some embodiments, the executing entity may store the anomaly score sequence set and the industry topology network synchronization to a shared storage unit for access by a hardware accelerator, and may include the steps of:

First, according to the abnormal score sequence set and the data size descriptor of the industrial topology network, corresponding target logic address space is allocated in the shared storage unit. The target logical address space may be a continuous virtual address range predefined for a specific data set in the shared memory unit, for writing by the CPU and reading by the hardware accelerator. In practice, firstly, determining an abnormal score sequence set and a data size descriptor of an industrial topological network, and obtaining the size of a memory space required by each abnormal score sequence set and the data size descriptor of the industrial topological network. Then, two sections of continuous and aligned logic address space are applied to the memory manager of the shared memory unit to serve as target logic address space.

And secondly, establishing a direct memory access channel according to the target logic address space. The memory access channel may be a hardware channel that allows the external device to bypass the CPU and directly perform high-speed data transmission with the main memory, so as to release the CPU load. In practice, first, a physical address or an I/O virtual address corresponding to a target logical address space is acquired. The DMA channel parameters between the CPU and the shared memory unit controller are then configured to establish a direct memory access channel.

And thirdly, performing data slicing processing on the abnormal fractional sequence set to generate a time sequence data slicing set. The time sequence data slicing set can be a smaller and independent data block set which is obtained by slicing the abnormal fraction sequence set according to the parallelism of the computing units, and is convenient for parallel processing. In practice, first, the optimal fragmentation granularity is determined according to the core number and the cache size of the hardware accelerator. Then, the anomaly score sequence set is cut in a time window. Finally, a plurality of data fragments with balanced sizes and continuous inside are generated to be used as time sequence data fragment sets.

And fourthly, performing structural coding processing on the industrial topology network to generate a topology data block set. The topology data block set may be a logical data block set (topology data block set) formed by encoding the graph structure data of the industrial topology network into a format (e.g., CSR) suitable for hardware parallel access.

And fifthly, generating a transmission task stream based on the association relation and the transmission priority of the time sequence data slicing set and the topology data block set. The association relationship may be a calculation dependency relationship between the time-series data fragments and the topology data blocks, and determines the sequence of data transmission and calculation. For example, the risk of computing production platform a depends on the anomaly score of its upstream provider B, so B fragments need to be transmitted preferentially. The transmission priority may be a priority level set for the data transmission task, so as to ensure that the key data is prioritized to be in place, so as to start calculation as early as possible. For example, the topology blocks of the network hub and the abnormal slices of the core production platform are given the highest priority. The concurrent transmission task stream may be an ordered set of data transmission tasks. First, the dependencies between the shards and the data blocks are analyzed (e.g., computation requires edge and node attributes). The tasks are then marked with priorities (e.g., topology block priorities, critical path production platform shards). Finally, a task list is generated, wherein no-or low-dependency tasks may be marked for concurrent execution to generate a transport task stream.

And sixthly, executing the transmission task of the transmission task stream through the direct memory access channel so as to write the anomaly score sequence set and the industrial topology network into the target logic address space to obtain the target logic address space after data writing. The target logical address space after the data writing may be a logical address space with finished data writing and ready content. In practice, first, the source address (CPU memory) and destination address (target logical address space in the shared memory unit) of each task are acquired from the transport task stream. The DMA engine then initiates multiple transfer tasks in parallel, bypassing the CPU, reading data from the source address and writing to the destination address through the established direct memory access channel. And finally, after all transmission tasks are completed, the complete and organized abnormal score fragment set and topology data block set are stored in the target logic address space.

And seventh, generating a global memory mapping table based on the target logic address space and the data type identifier after the data writing. The global memory mapping table may be a global directory table for recording the positions, sizes and types of all data blocks in the shared memory, and is a "map" of the hardware accelerator accessing data. The data type identifier can be used to distinguish tags of different data structures (e.g., vectors, matrices, graphs) in the shared memory, facilitating proper parsing by the hardware accelerator. In practice, first, after each data transfer is completed, the target address space of the written data is scanned. Then, the start address, size, and type of each data block are recorded according to a predefined "data type identification". Finally, a global memory map is generated.

Eighth, by configuring the global memory mapping table to the hardware accelerator, a zero copy access path of the hardware accelerator to the target logical address space after the data writing is established. The zero copy access path may be a path where the hardware accelerator directly accesses the data in the shared memory through the pointer without copying through the CPU, and the delay is extremely low. In practice, first, the generated global memory map is sent to the hardware accelerator through a driver or a dedicated instruction. Then, the hardware accelerator kernel driver establishes an internal pointer according to the global memory mapping table, so that the computing unit can directly 'see' the corresponding address in the shared memory. And finally, the calculation thread of the accelerator can directly read and write data through the pointer, so that the establishment of a zero copy path is completed.

The above operation steps, as an invention point of the present disclosure, solve the technical problems mentioned in the background art, such as low heterogeneous data transmission efficiency, high hardware access addressing delay, and delayed risk early warning response, and further delayed resource scheduling response, which cannot avoid link failure in time. The technical problems are caused by the fact that a classified transmission strategy is not designed for time sequence/topological heterogeneous data, a direct addressing mechanism of hardware and storage is lacked, and the redundancy of data copying is high. The invention realizes low-delay synchronous storage and hardware zero-copy access of the abnormal score sequence set and the industrial topology network through heterogeneous data classification processing, concurrent transmission and global memory mapping table configuration, saves the time cost of data transmission and hardware addressing, and reduces the production loss caused by industrial chain breakage in the production process.

Step 105, through the parallel computing pipeline of the hardware accelerator, performing the following generating steps:

step 1051, generating comprehensive anomaly information of the target production platform based on the anomaly score sequence set and the industrial topology network.

In some embodiments, the execution body may generate the integrated anomaly information of the target production platform based on the anomaly score sequence set and the industrial topology network. The target production platform may be a specific production platform requiring risk analysis and intervention scheduling. The comprehensive anomaly information may be a time-series quantized sequence that fuses itself with risks associated with the industry chain. The parallel pipeline can be an efficient execution model in a hardware accelerator for decomposing a computing task into a plurality of independent phases and enabling different data to flow and process at the same time in different phases so as to maximize the hardware utilization rate. For example, comprehensive anomaly information for 1000 production platforms is determined. The pipeline can be designed as a stage 1 (parallel) of extracting the upstream neighbor set and the downstream neighbor set of each production platform in parallel, a stage 2 (parallel) of calculating the upstream/downstream aggregate value of each production platform in parallel, and a stage 3 (parallel) of merging the three types of scores of each production platform in parallel. Thus, when the data of the first batch production platform enters the stage 2, the data of the second batch production platform can immediately start the calculation of the stage 1, so that the pipelined parallel throughput is realized.

In some optional implementations of some embodiments, the executing entity generates the comprehensive anomaly information of the target production platform based on the anomaly score sequence set and the industrial topology network, and may include the steps of:

The first step is to extract the upstream neighbor set and the downstream neighbor set of the target production platform based on the industrial topology network. The upstream neighbor set may be a set of production platforms in the industrial topology network that provide a supply for a target production platform. The downstream neighbor set may be a set of production platforms in an industrial topology network that receive a target production platform product. In practice, first, a target production platform (e.g., an electronic equipment vendor) is located according to an industry topology network. Then traversing the industrial topology network to determine an upstream supply production platform pointing to the target production platform and a downstream production platform pointed to by the target production platform.

And secondly, determining the abnormal score information corresponding to the target production platform, the upstream neighbor set and the downstream neighbor set in a preset time window based on the abnormal score sequence set so as to obtain the abnormal score information, the upstream influence aggregation value information and the downstream influence aggregation value information of the target production platform. The preset time window may be a set abnormal statistical time range. The anomaly score information may be anomaly score related data (e.g., daily anomaly score, score mean) for the production platform over a time window. The above-mentioned target production platform anomaly score information may be anomaly quantified data of the target production platform itself. The upstream impact aggregate value information may refer to a sequence of comprehensive abnormal impact values of the upstream neighbor set on the target production platform calculated by an aggregate function (e.g., weighted average). The downstream impact aggregation value information may be a comprehensive abnormal impact value sequence of the downstream neighbor set on the target production platform, which is calculated through an aggregation function.

As an example, first, from the anomaly score sequence set, the anomaly scores of the target production platform and the upstream and downstream neighbor production platforms are extracted daily within a preset time window (for example, the last 7 days). Then, for the upstream neighbor set, the anomaly scores of the upstream neighbor set are weighted and summed daily according to the coupling strength (edge weight) of the upstream neighbor set and the target production platform, so that an upstream influence aggregation value is obtained. Similarly, a downstream impact aggregation value is determined. And finally, obtaining the abnormal score information, the upstream influence aggregation value information and the downstream influence aggregation value information of the target production platform.

And thirdly, fusing the anomaly score information of the target production platform, the upstream influence aggregation value information and the downstream influence aggregation value information to obtain comprehensive anomaly information. In practice, first, preset weight coefficients α, β, γ (for example, α=0.6, β=0.3, γ=0.1) are assigned to the target production platform anomaly score information, the upstream influence aggregation value information, and the downstream influence aggregation value information. Then, for each day in the time window, after determining the comprehensive abnormality index of the day, the results of each day are arranged in time sequence to form comprehensive abnormality information.

Step 1052, based on the comprehensive anomaly information, generating the industrial chain breakage probability information to map to the early warning level information.

In some embodiments, the executing entity may generate the industrial chain breakage probability information based on the comprehensive anomaly information to map the industrial chain breakage probability information to the early warning level information. The above-mentioned industrial chain breakage probability information may be a probability value that quantitatively indicates that the target production platform has a problem with a key provider and thus has a production break. The early warning level information may refer to discrete level identifiers for identifying the severity of an event, which are classified according to the risk level. For example, the early warning level information may include blue, yellow, orange, and red four-level early warning.

In some optional implementations of some embodiments, the executing entity may determine, based on the comprehensive risk index sequence, a future production platform risk evolution sequence of the target production platform using a pre-constructed time-series prediction model, and may include the following steps:

First, constructing corresponding time sequence abnormal characteristic vectors based on the comprehensive abnormal information. The time sequence risk feature vector can be a feature representation which is extracted from a historical comprehensive risk index sequence of the target production platform and is used for inputting a time sequence prediction model. In practice, first, a comprehensive risk index sequence of a target production platform is obtained. And then, extracting trend slope, fluctuation variance, risk acceleration, sliding mean value and extremum duty ratio of the comprehensive risk index sequence as five types of characteristics. And finally, quantitatively integrating the five types of features to construct a time sequence risk feature vector.

And secondly, constructing a neighbor anomaly aggregation feature vector based on the industrial topology network and the anomaly score sequence. The neighbor risk aggregation feature vector may be a vector fusing upstream and downstream neighbor risks, for example, an upstream weighted risk vector and a downstream weighted risk vector. In practice, first, an upstream neighbor set and the downstream neighbor set of the target production platform are determined based on the industrial topology network. And then, the neighbor anomaly score sequence is called, and the upstream aggregation risk, the downstream aggregation risk and the neighbor risk fluctuation are calculated according to the coupling strength weighting. And finally, integrating to obtain the neighbor risk aggregation feature vector.

And thirdly, inputting the time sequence abnormal feature vector and the neighbor abnormal aggregation feature vector into the pre-constructed time sequence prediction model to obtain an abnormal evolution sequence of the target production platform in the future. The time series prediction model may be a model for predicting a future trend based on historical time series data. The above-described temporal prediction model may be a bi-directional LSTM model that fuses topological attention mechanisms. The time sequence prediction model can adopt a three-layer structure of encoding- > fusing- > decoding. The coding layer inputs a splicing matrix of a target production platform self time sequence risk feature vector and a neighbor risk aggregation feature vector, and outputs a hidden state sequence. The fusion layer inputs a hidden state sequence, and outputs a comprehensive time sequence feature vector with a fixed length through attention weighting or pooling operation. The decoding layer inputs the comprehensive time sequence feature vector, and outputs the comprehensive time sequence feature vector as a production platform risk evolution sequence through full-connection network mapping. In practice, first, the timing risk feature vector and the neighbor risk aggregation feature vector are input into a pre-trained bi-directional LSTM timing prediction model (including a topological attention mechanism). Then, a future risk evolution sequence of the target production platform is obtained through a time sequence prediction model output layer

Fourth, a key object set is determined based on the industrial topology network. The set of key objects may be a few upstream suppliers screened based on an industry topology network that are critical to the proper operation of the target production platform.

Fifth, for each key object in the set of key objects, performing the following operations:

And step one, determining an aggregation anomaly representative value corresponding to the key object based on the anomaly evolution sequence. The aggregate anomaly representative value may be an integrated quantitative value of future risk of the key provider, for example, a risk average value of 0.78 for 15 days in the future of a key object. In practice, first, from the sequence of abnormal evolution, a sequence of future abnormal predictions of the key object is extracted (e.g., 14 days in the future). Then, according to a preset aggregation strategy (for example, taking the maximum value and the average value), the anomaly prediction sequence is compressed into a single aggregation anomaly representative value.

And sixthly, converting each aggregation anomaly representative value into a provider outage probability set. The provider outage probability set may be a set of probability values of each provider in the key provider set for occurrence of a outage. In practice, first, the aggregate anomaly representative value of each key provider is converted into a corresponding probability value directly or through a preset mapping function. The probability values for each key provider are then formed into a set of probabilities.

And seventh, determining the chain breakage probability based on the provider outage probability set so as to generate industrial chain breakage probability information. The above-mentioned chain breakage probability may be an overall probability that the industry chain breaks due to breakage of the key provider. In practice, first, outage probabilities for each key provider in a provider outage probability set are obtained. Then, based on probability theory, assuming that outage events for these suppliers are nearly independent, the probability that the entire link is not broken (i.e., the probability that all suppliers are constantly supplying) is determined. Finally, subtracting the non-breaking probability from 1 to obtain the total probability of breaking the industrial chain due to the outage of at least one key provider, wherein the total probability is used as the industrial chain breaking probability information.

And eighth step, matching the industrial chain breakage probability information with a plurality of preset probability threshold intervals to generate an initial early warning level. The preset probability threshold intervals may be a predefined numerical range mapping the continuous probability value to different early warning levels. For example, the threshold interval is set to [0, 0.3) - > green, [0.3, 0.5) - > yellow, [0.5,0.7) - > orange, [0.7,1.0] - > red. The initial early warning level may be a preliminary level classification result obtained directly after matching according to a threshold interval. For example, the probability of chain breakage is 0.65, and the initial warning level is "orange warning" when the probability falls into an orange zone.

And ninth, generating early warning grade information based on the initial early warning grade and a predefined early warning grade mapping rule base. The predefined alert level mapping rule base may be a knowledge base storing the correspondence between the alert level and more detailed alert parameters (e.g., signal strength, suggested points of interest). For example, the "orange warning" in the rule base maps to signal strength=0.7.

And 106, generating a corresponding resource scheduling instruction based on the early warning level information.

In some embodiments, the executing entity may generate the corresponding resource scheduling instruction based on the early warning level information. The resource scheduling instruction may be an executable command for driving the scheduling intervention resource.

In some optional implementations of some embodiments, the executing body may generate the corresponding resource scheduling instruction based on the early warning level information, and may include the following steps:

First, generating an initial resource scheduling policy information set based on the early warning level information and a preset resource scheduling policy knowledge base. The preset resource scheduling policy knowledge base may be a rule or case base for storing different recommended resource scheduling schemes. For example, the knowledge base of the preset resource scheduling policy may include "start backup vendor and increase safety stock" when the pre-alarm signal strength is { highest, global }. In practice, firstly, a preset resource scheduling strategy knowledge base is searched according to the early warning signal intensity, and the resource scheduling strategy meeting the condition is matched. And finally, combining the resource scheduling strategies meeting the conditions into an initial resource scheduling strategy information set.

And secondly, encoding the initial resource scheduling policy information set into a standard instruction format to generate a resource scheduling instruction. In practice, first, the system obtains an initial set of resource scheduling policy information. The policy information is then converted into instruction code or data messages that the resource control system can directly recognize and execute according to a predetermined standard instruction format (e.g., a specific JSON structure or API call specification). Finally, the generated standardized resource scheduling instruction.

And step 107, in response to receiving the resource scheduling instruction, controlling the resource control terminal to execute the scheduling operation on the resource.

In some embodiments, the executing entity may control the resource control terminal to perform the scheduling operation on the resource in response to receiving the resource scheduling instruction, where the resource control terminal may be a hardware or software terminal that performs the resource scheduling operation, for example, a vendor cargo source allocation platform.

In the process of solving the technical problems of the background technology by adopting the technical scheme, aiming at the application scene that an automobile manufacturer faces the resource scheduling when the production stopping risk of a key part supplier, particularly under the extreme conditions of single goods source dependence, low safety stock and extremely short reproduction window, the technical problems that the resource scheduling instruction is difficult to reliably execute under the real conditions of heterogeneous resources, complex networks and dynamic state changes, the instruction failure or execution risk is easily caused by the unavailable resources and the inconsistent states, the response time of the resource scheduling is wasted, and the interruption loss caused by the supply chain outage is often accompanied. Aiming at the following requirements and characteristics of the application scene, namely instruction robustness, execution certainty, resource suitability and process controllability, the following solution is adopted:

In some optional implementations of some embodiments, the executing body may control the resource control terminal to execute the scheduling operation for the intervention resource in response to receiving the resource scheduling instruction, and may include the following steps:

First, based on the resource scheduling instruction, a corresponding resource scheduling operation type and a target resource identifier are generated. The type of the resource scheduling operation may be a specific action category of resource scheduling, such as resource scheduling and alternative supplier docking. The target resource identification may be unique identification information of the resource to be scheduled, for example, "alternative supplier- > B production platform". In practice, the operation command (i.e., the resource scheduling operation type) and the operation object (i.e., the target resource identification) of the core may be extracted from the standardized field of the resource scheduling instruction.

And secondly, calling a corresponding resource control interface according to the resource scheduling operation type so as to establish communication connection with the target resource control terminal. The resource control interface may be a standardized programming interface provided by the resource control terminal for receiving instructions and interaction data. The communication connection may be a secure, stable data channel established with the target resource control terminal. In practice, firstly, according to the type of the resource scheduling operation, a target system to be called and a corresponding resource control interface thereof are determined. And then, carrying out handshake authentication with the target resource control terminal through the interface. Finally, a secure communication connection is established

And thirdly, sending a resource state query request to the target resource control terminal through the communication connection so as to acquire real-time resource state data. The resource status query request may be an instruction sent to the resource control terminal to obtain current status information of the resource. The real-time resource status data may be the latest status information (e.g., "Material A, current stock: 1500 pieces") of the resource returned by the resource control terminal at the time of the query. In practice, first, a resource status query request conforming to the target terminal specification is constructed through the established communication connection. The request is then sent to the target resource control terminal. And finally, receiving a response returned by the terminal, and extracting the required real-time resource state data from the response.

And step four, verifying the current executable of the resource scheduling instruction based on the real-time resource state data and the target resource identifier to generate an instruction verification result. The instruction verification result may be boolean judgment that is obtained by comparing the instruction requirement with the real-time resource state and whether the instruction is executable. For example, the instruction needs to dial 2000 pieces, but only 1500 pieces are stocked, and the verification result is "non-executable".

And fifthly, generating a resource control command sequence in response to the command verification result being executable. The above-mentioned resource control command sequence may be a low-level operation instruction which is ordered and can be directly understood and executed by the resource control terminal. In practice, first, according to the detailed requirements of the resource scheduling instruction, it is decomposed and translated into a series of low-level, atomized operation commands that can be directly understood by the resource control terminal. And finally, arranging the operation commands according to the execution sequence to form a resource control command sequence.

And sixthly, responding to the target resource control terminal to receive the resource control command sequence, and executing scheduling operation of the intervention resource. In practice, first, the system sequentially transmits each command in the sequence of resource control commands to the target resource control terminal via the communication connection. Then, after receiving the command, the resource control terminal drives the physical device (e.g., robot arm, conveyor belt, vehicle) controlled by the resource control terminal to perform the corresponding actual operation. Finally, the resource (e.g., cargo) in the physical world begins to undergo the expected displacement or change of state, determining that the scheduling operation is to be performed.

The above operation steps, as an invention point of the present disclosure, solve the technical problem mentioned in the background art, that "under the realistic conditions of heterogeneous resources, complex networks and dynamic state changes, it is difficult to reliably execute the resource scheduling instruction, and the instruction failure or execution risk is easy to be caused by unavailable resources and inconsistent states, which wastes valuable emergency response time and causes interruption loss caused by supply chain outage. The technical problems are caused by the fact that most of scheduling instructions generated by the existing system are human-oriented task sheets, mechanisms for interacting with an automatic terminal, checking the real-time state of the terminal and generating bottom control instructions are lacking, sudden conditions (such as equipment faults and network interruption) of an execution site cannot be dealt with, and the 'instructions are suspended' or the execution effect is uncontrollable. The invention constructs the closed loop control link generated and executed by the instruction analysis, connection establishment and state check received instruction, realizes the reliable and intelligent driving of the abstract business instruction into the concrete resource control action, saves the reworking cost and time delay caused by the incapability of executing the instruction, execution error or resource conflict, and improves the precision and success rate of resource scheduling.

In the process of solving the technical problems of the background technology by adopting the technical scheme, aiming at the scene that key parts are subjected to nationwide outage risk in the automobile production process, the material allocation of the trans-provincial alternative suppliers is required to be completed in a very short time, the technical problems that the resource scheduling instruction is fuzzy in semantics and disordered in operation units, logic conflict easily occurs in execution, emergency response failure is caused, scheduling response time is increased, a large amount of scheduling resources are wasted, production is incoherence, and production loss is caused are often accompanied. Aiming at the following requirements and characteristics of the application scene, which are required to be provided, the instruction analysis is accurate, the operation unit is atomized, the execution flow is ordered, the fault processing is automatic, and the following solution is adopted:

in some optional implementations of some embodiments, the executing entity may generate the resource control command sequence in response to the instruction verification result being executable, and may include the steps of:

and firstly, carrying out semantic analysis on the resource scheduling instruction to extract a corresponding operation target and an operation constraint condition. The operation target may be a final state that the scheduling instruction is expected to achieve. For example, the operating objective may be "1000 cell modules are supplied from the wuhan silo to the Shanghai factory within 8 hours". The above-described operational constraints may be limitations that must be followed when executing instructions. For example, the operating constraints may include "transport process humidity <10%", "antistatic packaging needed", "global GPS tracking".

And secondly, decomposing the resource scheduling instruction into a basic operation unit set according to the operation target and the operation constraint condition. Wherein the basic operation unit set may be a minimum executable action set constituting a complete instruction. For example, the base set of operating units may include querying inventory, generating a dial sheet, scheduling a thermostat, and scheduling reception of four units.

And thirdly, adding a pre-condition identifier and a subsequent effect identifier to each basic operation unit in the basic operation unit set based on the operation constraint conditions so as to generate a plurality of atomic operation tasks. The precondition identifier may be a state that must be satisfied before an operation unit starts. For example, the precondition identification may be that the precondition for dispatching a thermostat is "the pick up has been approved and there is a free vehicle". The subsequent effect identification may be a status change after an operation unit is successfully identified. For example, the subsequent effect identification may be that the subsequent effect of the "out-of-stock scan" is "inventory reduction, and the cargo state becomes" on the way ". The atomic operation tasks may be basic operation units with precondition identifiers and subsequent effect identifiers and independently schedulable. In practice, first, the preconditions for each basic operation unit execution are determined. Then, a state change of the system after its execution is determined. Finally, a prefix and a follow-up identifier are added for each basic operation unit to form a plurality of atomic operation tasks.

And fourthly, constructing a task execution dependency graph based on the plurality of atomic operation tasks. The task execution dependency relationship may be a directed acyclic graph, the node is an atomic task, and the edge represents a sequential dependency relationship between tasks. In practice, first, according to traversing multiple atomic operation tasks, if the precondition of one task is a subsequent effect of another task, a directed edge is established from the preceding task to the task. A directed acyclic graph is then formed for defining the order of execution of the various tasks.

And fifthly, generating a standardized operation instruction template based on the task execution dependency graph and a pre-constructed instruction template library. The standardized operation instruction template can be an instruction framework conforming to the current task and equipment interface. In practice, first, each task of the task execution dependency graph is traversed, and the instruction templates library is searched for matching templates according to its operation type (e.g., "vehicle schedule", "temperature control on"). Then, the most appropriate template is selected for each task. And finally, outputting a standardized operation instruction template which is arranged according to the task sequence.

And sixthly, filling the real-time resource state data, the target resource identification and the corresponding scheduling parameters into the standardized operation instruction template to generate an equipment control instruction. The device control instruction may be a specific instruction that is generated after filling a specific parameter into a standardized template and may be directly executed by a device.

And seventhly, sequencing and synchronizing the equipment control instructions based on the time sequence and the logic relation of the task execution dependency graph to obtain a preliminary command sequence. The preliminary command sequence may be a device control command list which is obtained by sequencing according to a task execution dependency graph and can be sequentially executed logically. In practice, firstly, according to a task execution dependency graph, a device control instruction set is subjected to topological ordering to obtain a basic execution sequence. Then, task branches which can be executed concurrently in the task execution dependency graph are identified, and a concurrency start marker is added to the task branches in the sequence. Finally, a preliminary command sequence is generated.

And eighth step, injecting fault-tolerant processing logic into the preliminary command sequence to generate a resource control command sequence. The fault-tolerant processing logic may be checking and recovering logic pre-embedded with instruction sequences for coping with execution exceptions. For example, if the "injection after scheduling the vehicle" command is not confirmed for 10 minutes, the scheduling of the spare vehicle 002 "is attempted. In practice, first, each critical instruction in the preliminary sequence is analyzed for potential failure points (e.g., dispatch timeout, equipment failure). Then, a condition judgment, retry, or standby scheme instruction is inserted after the corresponding instruction. Finally, a resource control command sequence with robust error handling logic is generated.

The above operation steps, as an invention point of the present disclosure, solve the technical problems mentioned in the background art, namely, the resource scheduling instruction is ambiguous in semantics and the operation units are disordered, and logic conflict is easy to occur during execution, so that emergency response failure is caused, scheduling response time is increased, a large amount of scheduling resources are wasted, production is incoherence, and production loss is caused. The technical problems are caused by the fact that semantic analysis and atomization splitting of the dispatching instruction are not carried out, task dependency relationship carding is lacked, and fault-tolerant processing logic support is not available. The method realizes the standardized conversion from the resource scheduling instruction to the executable command sequence through the instruction semantic analysis- > atomic task splitting- > dependency relation modeling- > fault-tolerant logic injection, saves the instruction debugging and fault repairing time and improves the emergency response efficiency.

The method for scheduling the resources based on the industrial topology network has the advantages that the method for scheduling the resources based on the industrial topology network of some embodiments of the disclosure achieves automatic intervention of industrial chain breakage of the production platform, saves time resources and reduces supply chain interruption loss. Specifically, the supply chain interruption loss is caused by serious data island, difficulty in correlating multisource heterogeneous original data and lack of risk propagation analysis from the view angle of an industrial chain, so that risk identification lag and insufficient breakpoint prediction are caused, and intervention measures cannot be timely adopted to cause supply chain interruption. Based on this, in some embodiments of the present disclosure, a resource scheduling method based on an industrial topology network first generates a corresponding operation information set based on multi-source heterogeneous raw data of a plurality of production platforms in an industrial chain corresponding to a target product in response to detecting that the inventory of the target product is lower than a target threshold. The multi-source heterogeneous original data is uniformly encoded into a standardized time sequence vector to form a continuous and measurable operation information set, and uniform and comparable basic data representation is provided for subsequent analysis. Then, using an anomaly detection engine, a corresponding anomaly score sequence set is generated based on the running information set. By generating the anomaly score sequence set, the state deviation degree of the production platform is accurately quantized, the early weak anomaly signal recognition sensitivity is improved, and visual quantization basis is provided for risk assessment. And then, based on the operation information set, calling a parallel computing node cluster of the topology construction server to generate an industrial topology network representing the product circulation relation among the production platforms. By constructing the industrial topology network, supply association among production platforms can be efficiently excavated, real-time monitoring of large-scale production platforms is supported, and an industrial chain structure is clearly presented. And secondly, synchronously storing the abnormal score sequence set and the industrial topology network to a shared storage unit for access by a hardware accelerator, wherein the shared storage unit is used for zero-copy data parallel access. The data copying cost between the CPU and the hardware accelerator is eliminated, the data consistency is ensured, and a low-delay and high-bandwidth data supply channel is provided for subsequent hardware acceleration calculation. And generating comprehensive anomaly information of the target production platform based on the anomaly score sequence set and the industrial topology network through the parallel computing pipeline of the hardware accelerator. The parallel computing assembly line improves the multi-source data fusion efficiency, integrates abnormal scores and topological association, realizes comprehensive evaluation of production platform risks, and avoids one-sided performance of isolated analysis. And generating the industrial chain breakage probability information based on the comprehensive abnormal information so as to map the industrial chain breakage probability information into early warning grade information. And the risk of industrial chain breakage is quantified, the risk degree is visually presented through the early warning level, so that quick response is facilitated, and the problem of response hysteresis caused by fuzzy risk level is solved. And then, based on the early warning level information, generating a corresponding resource scheduling instruction. The accurate matching of the abnormality and the scheduling strategy is realized, and the pertinence and the effectiveness of the scheduling instruction are ensured. And finally, controlling the resource control terminal to execute the scheduling operation of the resource in response to the received resource scheduling instruction, wherein the scheduling operation comprises the storage of the parts related to the target product and the transportation of the parts. And the storage and transportation scheduling of the quick floor parts are realized, the production resources are timely supplemented, the chain breakage probability of the industrial chain is reduced, the production continuity of the target product is ensured, the risk of quick response is ensured, and the chain breakage loss of the industrial chain is reduced.

With further reference to fig. 2, as an implementation of the method shown in the foregoing figures, the present disclosure provides some embodiments of an industrial topology network-based resource scheduling apparatus, which corresponds to those method embodiments shown in fig. 1, and which is particularly applicable to various electronic devices.

As shown in fig. 2, a resource scheduling apparatus 200 based on an industrial topology network includes a first generation unit 201, a second generation unit 202, a scheduling unit 203, a storage unit 204, a calling unit 205, a third generation unit 206, and a control unit 207. The first generating unit 201 is configured to generate, in response to detecting that the inventory of the target product is below the target threshold, a corresponding set of operation information based on the multi-source heterogeneous raw data of the plurality of production platforms in the industry chain corresponding to the target product. The second generation unit 202 is configured to generate a corresponding anomaly score sequence set based on the above-described running information set using the anomaly detection engine. The invoking unit 203 is configured to invoke the parallel computing node clusters of the topology construction server to generate an industrial topology network characterizing the product flow relationship between the production platforms based on the set of operational information. The storage unit 204 is configured to store the anomaly score sequence set and the industry topology network synchronously to a shared storage unit for hardware accelerator access, wherein the shared storage unit is used for zero copy data parallel access. The execution unit 205 is configured to execute, through the parallel computing pipeline of the hardware accelerator, a generation step of generating comprehensive anomaly information of the target production platform based on the anomaly score sequence and the industrial topology network. And generating the industrial chain breakage probability information based on the comprehensive abnormal information so as to map the industrial chain breakage probability information into early warning grade information. The third generating unit 206 is configured to generate a corresponding resource scheduling instruction based on the above-mentioned early warning level information. The control unit 207 is configured to control the resource control terminal to perform a scheduling operation of the resource in response to receiving the above-mentioned resource scheduling instruction, wherein the above-mentioned scheduling operation includes storage of the target product-related parts, and transportation of the above-mentioned parts.

It will be appreciated that the elements recited in the industrial topology network-based resource scheduling apparatus 200 correspond to the various steps in the method described with reference to fig. 1. Thus, the operations, features and advantages described above for the method are equally applicable to the resource scheduling device 200 based on the industrial topology network and the units contained therein, and are not described herein again.

Referring now to fig. 3, a schematic diagram of an electronic device (e.g., electronic device) 300 suitable for use in implementing some embodiments of the present disclosure is shown. The electronic device shown in fig. 3 is merely an example and should not impose any limitations on the functionality and scope of use of embodiments of the present disclosure.

As shown in fig. 3, the electronic device 300 may include a processing means (e.g., a central processing unit, a graphics processor, etc.) 301 that may perform various suitable actions and processes in accordance with a program stored in a Read Only Memory (ROM) 302 or a program loaded from a storage means 308 into a Random Access Memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic apparatus 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input/output (I/O) interface 305 is also connected to bus 304.

In general, devices may be connected to I/O interface 305 including input devices 306 such as a touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc., output devices 307 including a Liquid Crystal Display (LCD), speaker, vibrator, etc., storage devices 308 including, for example, magnetic tape, hard disk, etc., and communication devices 309. The communication means 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. While fig. 3 shows an electronic device 300 having various means, it is to be understood that not all of the illustrated means are required to be implemented or provided. More or fewer devices may be implemented or provided instead. Each block shown in fig. 3 may represent one device or a plurality of devices as needed.

In particular, according to some embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In such embodiments, the computer program may be downloaded and installed from a network via communications device 309, or from storage device 308, or from ROM 302. The above-described functions defined in the methods of some embodiments of the present disclosure are performed when the computer program is executed by the processing means 301.

It should be noted that, in some embodiments of the present disclosure, the computer readable medium may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of a computer-readable storage medium may include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In some embodiments of the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, however, the computer-readable signal medium may comprise a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to electrical wiring, fiber optic cable, RF (radio frequency), and the like, or any suitable combination of the foregoing.

In some embodiments, the clients, servers may communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol ), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), the internet (e.g., the internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future developed networks.

The computer readable medium may be included in the electronic device or may exist alone without being incorporated into the electronic device. The computer readable medium carries one or more programs, when the one or more programs are executed by the electronic device, the electronic device is caused to respond to detection that the stock of a target product is lower than a target threshold value, a corresponding operation information set is generated based on multi-source heterogeneous original data of a plurality of production platforms in an industrial chain corresponding to the target product, a corresponding abnormal score sequence set is generated based on the operation information set by utilizing an abnormal detection engine, a parallel computing node cluster of a topology construction server is called based on the operation information set to generate an industrial topological network representing the product circulation relation among the production platforms, the abnormal score sequence set and the industrial topological network are synchronously stored in a shared storage unit for being accessed by a hardware accelerator, wherein the shared storage unit is used for parallel access of zero copy data, the following generation steps are executed through a parallel computing pipeline of the hardware accelerator, the comprehensive abnormal information of the target production platform is generated based on the abnormal score sequence and the industrial topological network, the comprehensive abnormal information is used for generating industrial chain fracture probability information based on the comprehensive abnormal information, the early warning information is mapped into a dispatching command, the dispatching command is received based on the corresponding dispatching resource, and the dispatching command is received, and the dispatching command is controlled by a relative resource.

Computer program code for carrying out operations for some embodiments of the present disclosure may be written in one or more programming languages, including an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units described in some embodiments of the present disclosure may be implemented by means of software, or may be implemented by means of hardware. The described units may also be provided in a processor, for example as a processor comprising a first generation unit, a second generation unit, a scheduling unit, a storage unit, a calling unit, a third generation unit and a control unit. The names of these units do not limit the unit itself in some cases, for example, the first generating unit may also be described as "in response to detecting that the inventory of the target product is lower than the target threshold value, a unit that generates a corresponding running information set based on the multi-source heterogeneous raw data of multiple production platforms in the industrial chain corresponding to the target product" as described above ".

The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic that may be used include Field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems-on-a-chip (SOCs), complex Programmable Logic Devices (CPLDs), and the like.

The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above technical features, but encompasses other technical features formed by any combination of the above technical features or their equivalents without departing from the spirit of the invention. Such as the above-described features, are mutually substituted with (but not limited to) the features having similar functions disclosed in the embodiments of the present disclosure.

Claims (10)

1. A resource scheduling method based on an industrial topology network, comprising: In response to detecting that the inventory of a target product is lower than a target threshold, a corresponding set of operational information is generated based on multi-source heterogeneous raw data from multiple production platforms in the industrial chain corresponding to the target product. Using the anomaly detection engine, a corresponding set of anomaly score sequences is generated based on the aforementioned set of operational information; Based on the aforementioned set of operational information, the parallel computing node cluster of the topology construction server is invoked to generate an industrial topology network that characterizes the product flow relationship between production platforms. The abnormal score sequence set and the industry topology network are synchronously stored in a shared storage unit for access by the hardware accelerator, wherein the shared storage unit is used for zero-copy data parallel access; The following generation steps are performed through the parallel computing pipeline of the hardware accelerator: Based on the set of abnormal score sequences and the industry topology network, comprehensive abnormal information of the target production platform is generated; Based on the comprehensive anomaly information, supply chain disruption probability information is generated and mapped to early warning level information; Based on the aforementioned warning level information, a corresponding resource scheduling instruction is generated; In response to receiving the resource scheduling instruction, the resource control terminal is controlled to perform a resource scheduling operation, wherein the scheduling operation includes: storage of components related to the target product and transportation of the components. 2. The method according to claim 1, wherein, in response to detecting that the inventory of the target product is lower than a target threshold, generating a corresponding operational information set based on multi-source heterogeneous raw data from multiple production platforms in the industrial chain corresponding to the target product includes: Perform the following steps on each of the multi-source heterogeneous raw data from the multiple production platforms: The heterogeneous raw data from multiple sources on the production platform are subjected to heterogeneous data time series isomorphism processing to obtain isomorphic time series data; Based on the homogenized time series data, a standard multidimensional trajectory vector is constructed; The standard multidimensional trajectory vector is subjected to continuity preservation processing to generate corresponding running information; All operational information is integrated into an operational information set. 3. The method according to claim 1, wherein generating a corresponding anomaly score sequence set based on the runtime information set using the anomaly detection engine includes: For each piece of operational information in the aforementioned operational information set, the following steps are performed: Using the first node of the anomaly detection engine, a trajectory curvature change sequence corresponding to the running information is generated; Using the second node of the anomaly detection engine, based on the trajectory curvature change sequence, the preset historical normal trajectory benchmark of the production platform itself, and the industry average trajectory benchmark of the same period, a dual-benchmark dynamic discrimination operation is performed to obtain dual-benchmark deviation data. Based on the dual-benchmark deviation data and adaptive weight parameters, an anomaly score sequence is generated, wherein the adaptive weight parameters are dynamically determined by the anomaly detection engine based on the completeness of historical data and the accuracy of historical warnings on the production platform. 4. The method according to claim 1, wherein the step of invoking the parallel computing node cluster of the topology construction server based on the runtime information set to generate an industry topology network characterizing the product flow relationship between production platforms includes: The first parallel computing node of the parallel computing node cluster is invoked to determine the event-type trajectory similarity between every two production platforms in the running information set, so as to generate an event synchronization matrix. The second parallel computing node of the parallel computing node cluster is invoked to determine the numerical correlation of the continuous trajectories of every two production platforms in the running information set, so as to generate a numerical correlation matrix. The third parallel computing node of the parallel computing node cluster is invoked to determine the time-delay causal relationship between the running information of every two production platforms in the running information set, so as to generate the optimal time-delay matrix; The event synchronization matrix, the numerical correlation matrix, and the optimal time delay matrix are weighted and fused to generate a comprehensive similarity matrix between production platforms. Using the pre-acquired importance scores of production platforms, the comprehensive similarity matrix between production platforms is weighted and corrected to generate a coupling strength matrix between production platforms; The coupling strength matrix between the production platforms is filtered to generate an initial directed network of the industrial chain; The initial directed network of the industrial chain is subjected to noise filtering to generate an industrial topology network. 5. The method according to claim 1, wherein generating comprehensive anomaly information of the target production platform based on the anomaly score sequence set and the industry topology network includes: Based on the industry topology network, extract the upstream neighbor set and downstream neighbor set of the target production platform; Based on the abnormal score sequence set, the abnormal score information corresponding to the target production platform, the upstream neighbor set, and the downstream neighbor set within a preset time window is determined, so as to obtain the abnormal score information of the target production platform, the upstream influence aggregation value information, and the downstream influence aggregation value information. The abnormal score information of the target production platform, the aggregated value information of the upstream impact, and the aggregated value information of the downstream impact are fused to obtain comprehensive abnormal information. 6. The method according to claim 1, wherein generating supply chain disruption probability information based on the comprehensive anomaly information, and mapping it to early warning level information, includes: Based on the comprehensive anomaly information, a corresponding time-series anomaly feature vector is constructed; Based on the industry topology network and the anomaly score sequence, a neighbor anomaly aggregation feature vector is constructed; The time-series anomaly feature vector and the neighbor anomaly aggregation feature vector are input into the pre-built time-series prediction model to obtain the future anomaly evolution sequence of the target production platform; Based on the aforementioned industry topology network, a set of key objects is determined; For each key object in the key object set, perform the following operations: Based on the abnormal evolution sequence, determine the aggregated abnormal representative value corresponding to the key object; Each aggregated anomaly representative value is transformed into a supplier disruption probability set; Based on the supplier supply disruption probability set, the supply chain disruption probability is determined to generate supply chain disruption probability information. The probability information of supply chain disruption is matched with multiple preset probability threshold ranges to generate an initial warning level; Based on the initial warning level and the predefined warning level mapping rule base, warning level information is generated. 7. The method according to claim 1, wherein generating a corresponding resource scheduling instruction based on the early warning level information includes: Based on the aforementioned warning level information and the preset resource scheduling strategy knowledge base, an initial resource scheduling strategy information set is generated; The initial resource scheduling strategy information set is encoded into a standard instruction format to generate resource scheduling instructions. 8. A resource scheduling device based on an industrial topology network, comprising: The first generation unit is configured to generate a corresponding set of operational information based on multi-source heterogeneous raw data from multiple production platforms in the industrial chain corresponding to the target product in response to detecting that the inventory of the target product is lower than the target threshold. The second generation unit is configured to use an anomaly detection engine to generate a corresponding set of anomaly score sequences based on the set of runtime information. The calling unit is configured to call the parallel computing node cluster of the topology construction server based on the running information set to generate an industry topology network that represents the product flow relationship between production platforms. A storage unit is configured to synchronously store the abnormal score sequence set and the industry topology network to a shared storage unit for access by a hardware accelerator, wherein the shared storage unit is used for zero-copy data parallel access; The execution unit is configured to perform the following generation steps via the parallel computing pipeline of the hardware accelerator: Based on the set of abnormal score sequences and the industry topology network, comprehensive abnormal information of the target production platform is generated; Based on the comprehensive anomaly information, supply chain disruption probability information is generated and mapped to early warning level information; The third generation unit is configured to generate corresponding resource scheduling instructions based on the warning level information; The control unit is configured to control the resource control terminal to perform resource scheduling operations in response to receiving the resource scheduling instruction, wherein the scheduling operations include: storage of components related to the target product and transportation of the components. 9. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7. 10. A computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any one of claims 1-7.