LU601247B1 - An industrial big data processing method and system based on artificial intelligence algorithms - Google Patents

Abstract

The present invention provides an industrial big data processing method and system based on artificial intelligence algorithms and relates to the field of industrial intelligence technology. The industrial big data processing method and system based on artificial intelligence algorithms comprises the following steps: S1: Based on heterogeneous data from multiple sources of industrial plants, vibration signals are processed to generate spectrograms, extract image features, semantically encode text protocols, align various modal feature distributions, and generate multimodal common feature vectors. The vibration signal is converted into a spectrogram by a multimodal adverse domain matching algorithm, and image features are extracted by combining a neural convolutional network, while the text protocols are semantically encoded using a bidirectional long- and short-term memory network to eliminate interference from device differences, achieve effective alignment of cross-modal features, solve the problem of heterogeneity of industrial multi-source data, and improve the quality of data fusion.

Description

An industrial big data processing method and system based on LU601247 artificial intelligence algorithms

Technical area

The present invention relates to the field of industrial

Intelligence technology, in particular an industrial big data processing method and system based on artificial intelligence algorithms.

Technology in the background

Industrial intelligence is a cross-technology field that encompasses artificial intelligence.

Intelligence, big data analytics and industrial mechanics are integrated and focus on optimizing manufacturing processes, predicting equipment failures and improving the

Resource utilization through data-driven intelligent technologies is the focus. Its core objective is the development of an industrial system with autonomous perception, real-time analysis, and dynamic decision-making. Key technologies include machine learning,

Deep learning, industrial knowledge graphs, and edge computing, as well as application scenarios such as plant health management, production process optimization, and supply chain collaboration. The artificial intelligence-based method and system for industrial big data processing relate to the processing of heterogeneous industrial data from multiple sources using machine learning models and the establishment of a continuous data analysis and

Decision chain. The goal is to monitor and predict the operational status of

Real-time optimization of industrial plant configurations

Production process parameters, the identification of the causes of quality defects and the dynamic adjustment of the production strategy to improve manufacturing efficiency and resource utilization, in order to support the demand for precise decisions, cost reductions and efficiency improvements in the intelligent transformation of companies.

When dealing with heterogeneous data from multiple sources in industry, multimodal feature fusion typically relies on simple splicing or weighted analysis.

Averaging, which accounts for the difference caused by the different types of equipment

Distribution bias is difficult to eliminate, leading to lower accuracy in cross-device state prediction. Most methods for evaluating the

Device status is based on an isolated model analysis, in which the topological

Relationships between the devices and the fault propagation paths are not explicitly modeled, and the dynamic correlation between the nodes in a complex

The industrial system is not captured, leading to a distorted localization of the cause of the error.

Content of the invention

In view of the shortcomings of the prior art, the present invention provides an industrial big data processing method and system based on artificial intelligence algorithms, which solves the problem that the prior art lacks in the

Handling heterogeneous industrial data from multiple sources, the multimodal

Feature fusion is usually based on simple splicing or weighted averaging, which makes it difficult to effectively eliminate distribution errors caused by differences in device types, leading to a decrease in the accuracy of cross-device state prediction.

To achieve the above-mentioned purpose, the present invention is achieved by the following technical solution: an industrial big data processing method based on the

The basis of an artificial intelligence algorithm, which includes the following steps:

S1: Based on heterogeneous data from multiple sources of industrial plants, vibration signals are processed to generate spectrograms, extract image features, semantically encode text protocols, and various modal...

to align feature distributions and generate multimodal common feature vectors;

S2: Based on the multimodal common feature vector, the

Topology map of the equipment constructed, the law on the development of the

Equipment status extracted, data on working status integrated to the

To calculate the weighting of the node's health status and the assessment of the

to create an assessment of the equipment's health condition;

S3: Combine the equipment condition assessment and the process parameters, train the intelligent body to control the process variables, simulate the extreme working condition data to adapt the strategy network, and generate the dynamically optimized process parameter set;

S4: Construction of the causal diagram of process parameters and quality indicators,

Simulation of the effects of parameter adjustments, iterative search for the optimal

Value range of key parameters and creation of causally controlled

Process improvement plans.

Preferably, the generation of a multimodal common feature vector based on S1 comprises the following steps:

S101: Processing the vibration signal to generate a spectrogram, extracting local features, eliminating sensor noise, and generating the common vibration and image feature vector;

S102: Semantic coding of the operating and maintenance protocols, combination of the common vibration and image feature vectors to align the distribution, and

Generation of the multimodal common feature vector.

Preferably, generating the assessment of the plant condition on the

The basis of S2 is the following steps:

S201: Constructing device-node relationships, calculating

Error propagation weights and generating a topological

Device relationship feature map;

S202: Segmentation of equipment operating data, merging of historical data

Operating status data, dynamic weighting of time step characteristics and generation of the

Assessment of the equipment's health condition.

Preferably, the generation of a dynamically optimized set of

Process parameters based on dynamic optimization in S3 include the following:

Steps:

S301: Assignment of process parameter intelligences, initialization of the

Strategy network and creation of a cooperative

Multi-parameter control strategy network;

S302: Simulating extreme operating condition data, adjusting the boundary conditions of the

Strategy network and generation of the dynamically optimized process parameter set.

Preferably, the generation of a causally controlled+/601 247

Process improvement schemes based on S4 include the following steps:

S401: Creating a causal map of process parameters and quality indicators,

Eliminating pseudocorrelated variables and generating a causal process parameter and

Quality network:;

S402: Simulation of the effect of parameter changes, search for the optimal

Combination of parameters that satisfy the constraints and generation of a causality-driven process improvement scheme.

Fin's industrial big data processing system is based on artificial intelligence algorithms, and the industrial big data processing system comprises the following modules: a module for processing multimodal features, a module for

Device condition assessment, a module for decision-making in the

Process optimization and a module for improving causal analysis;

The module for processing multimodal features processes vibration signals to

Generating spectrograms, extracting image features, encoding text protocols, matching multimodal distributions, and generating multimodal common feature vectors;

The module for processing multimodal features includes a submodule for

Processing of vibration images, a sub-module for semantic text coding, and a

Submodule for feature alignment;

Module for evaluating the device's condition, which uses topological relationship graphs of the

Device constructed, dispersion weights calculated, time series operating condition data fused, and device condition assessments generated;

The module for assessing the condition of the plant includes a sub-module for creating topological relationships and a sub-module for convolution of time series graphs;

Process optimization decision-making module that

assigns process parameter control strategies, trains intelligent bodies to respond to

Working together to achieve optimization goals, extreme working conditions were simulated to adjust parameters, and dynamically optimized process parameter sets were generated;

The process optimization decision-making module includes a sub-module for the

Collaboration of multiple intelligent bodies and a sub-module for simulating extreme conditions

Working conditions;

The causal analysis and improvement module constructs a parameter and

Quality causal network, simulates the effect of parameter interventions, iteratively searches for the optimal solution and generates a causally controlled process improvement scheme;

The module for improving causal analysis includes a sub-module for

Causality detection and a sub-module for counterfactual optimization.

Preferably a sub-module for processing vibration patterns, which includes a

A spectrogram was generated, frequency domain features were extracted, noise was eliminated, and common vibration and image feature vectors were created;

Submodule for semantic text encoding that encodes contextual semantics,

Error description time correlation features extracted and textual feature vectors generated;

Sub-module for feature alignment, which measures the distributions of vibration, image and

Aligns text features, generates multimodal common feature vectors, generates multimodal common feature vectors.

Preferably, the submodule for constructing the topological relationship /601 247 constructs the relationship between the device nodes, defines the weights of the fault propagation paths and generates the feature map for the topological relationship of the devices;

A time-step graph convolution submodule that merges historical operational state data, dynamically weights time-step features, and generates equipment condition assessments.

Preferably a multi-intelligence synergy sub-module that assigns independent intelligences to the process parameters and generates a multi-parameter synergy control strategy network;

Submodule for simulating extreme operating conditions, which stores data on extreme

Operating conditions are created, the strategy network is optimized, and a dynamically optimized set of process parameters is generated.

Preferably a sub-module for causality detection, which includes parameter and

Quality causal diagrams created, pseudo-correlated variables eliminated, and causal relationships determined.

Process parameter and quality networks were created;

Submodule for counterfactual optimization that simulates parameter interventions, searches for optimal parameter combinations that satisfy constraints, and causally driven

Process improvement schemes were generated.

The present invention presents an industrial big data processing method and system based on artificial intelligence algorithms. The following advantageous effects are provided:

The present invention converts vibration signals into spectrograms using a multimodal adverse domain matching algorithm and extracts image features into

Combined with a neural convolutional network, while employing a bidirectional long- and short-term memory network for the semantic coding of text protocols,

Interference caused by device differences is eliminated, enabling effective cross-modal matching.

Features are realized, the problem of heterogeneity in industrial multi-source data is solved, and the quality of data fusion is improved. Based on the graph attention network extended by the knowledge graph, the node connection relationship is defined by the

The combination of the topological relationship mapping of the devices defines the law on the evolution of the device state, and the law on the evolution of the device state is extracted by the temporal sliding window mechanism to improve the spatiotemporal modeling capability of the device state evaluation and to overcome the shortcomings of traditional methods that are present in the

Modeling of device correlation is insufficient. The assumption of multi-intelligence

Body deep reinforcement learning framework, combined with the generative contradictory

Network to simulate extreme working conditions, dynamically balance exploration and use of the strategy network, achieving the dynamic optimum

Configuration of process parameters and avoidance of the local optimality problem caused by human experience and intervention. The structural causal model is introduced to construct the causal diagram of process parameters and quality indicators, illustrating the effects of parameter adjustment based on counterfactual data.

Considerations to simulate and iteratively search for the optimal range of values

to find key parameters by combining it with Bayesian optimization to eliminate the error accumulation in traditional statistical correlation analysis and improve the interpretability and reliability of decision-making.

to improve the process improvement plan. The aforementioned technical means constitute a 601247 closed process of multidimensional synergistic optimization, which is driven by the

Data fusion through condition assessment and parameter optimization to causal analysis

Decision-making is sufficient and ultimately improves precision, dynamic adaptability, and resource utilization optimization throughout.

Life cycle management of industrial plants is enabled.

Description of the attached drawings

Figure 1 shows a block diagram of the system of the present invention;

Figure 2 shows a schematic diagram of the main steps of the present

Invention;

Figure 3 shows a schematic diagram of the S1 refinement of the present invention;

Figure 4 shows a schematic diagram of the S2 refining process of the present invention;

Figure 5 shows a schematic diagram of the S3 refinement of the present invention;

Figure 6 shows a schematic diagram of the S4 refinement of the present

Invention.

Detailed description

The technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings, and it is obvious that the described embodiments represent only a part of the embodiments of the present invention.

This does not represent the invention, but rather all its embodiments. Based on the

In the present invention, all other embodiments that can be achieved by a person skilled in the art without creative work fall within the scope of protection of the present invention.

embodiment

As shown in Figures 1-6, one embodiment of the present invention provides an industrial big data processing method based on an artificial intelligence algorithm, comprising the following steps:

S1: Based on heterogeneous data from multiple sources of industrial plants, the

Vibration signal processed to generate a spectrogram, extract image features, semantically encode the text protocol, various modal

to compare feature distributions and generate a common multimodal feature vector;

The vibration signal is converted into a short-time Fourier transform.

Time-frequency spectrogram is converted, and frequency domain features are extracted using a neural convolutional network; the semantics of the text protocol are encoded using a bidirectional long- and short-term memory network, and the multimodal

Distributions are matched using an adverse range fitting algorithm to generate multimodal common feature vectors.

S2: Based on the multimodal common feature vector, construct the topological relationship map of the equipment, extract the law for the evolution of the equipment state, and fuse the working state data to determine the weight of the

to calculate the node's health status and generate the rating of the

The equipment's condition;

Application of the graph attention network to create the adjacency matrix of the plant topology, combination of the long-term and short-term memory networks to calculate the time sequence data of the V601 247

to process the working condition and to create an assessment of the plant's health status by merging features via the self-awareness mechanism.

S3: Combine the equipment health values with the

Process parameters, train intelligent bodies to control the process variables, simulate extreme operating conditions data to adapt the strategy network, and generate dynamically optimized process parameter sets;

Construction of a deep deterministic strategy gradient framework with multiple intelligent bodies, simulation of data on extreme working conditions through an adversarial conditional generation network, and application of the proximal

Strategy optimization for updating the control strategy.

S4: Construction of the causal diagram of process parameters and quality indicators,

Simulation of the effects of parameter adjustment, iterative search for the optimal

Value range of the key parameters and creation of the causally controlled

Process improvement plan.

Construct a causal diagram based on a Bayesian network, use counterfactual considerations to simulate parameter interventions, and combine these with a multi-objective optimization algorithm to create a

To create a process improvement program.

The generation of a multimodal common feature vector based on

S1 includes the following steps:

S101: Processing the vibration signal to generate a spectrogram, extracting local features, eliminating sensor noise, and generating a common

Vibration image feature vector;

Performing a wavelet packet decomposition of the vibration signal, extracting from

Meier frequency cepstrum features and application of a non-local

Mean-value denoising to generate vibration and image connection feature vectors.

S102: Semantic coding of operational and maintenance logs, combined with

Vibration and common feature vectors aligned to the distribution to generate multimodal common feature vectors.

Based on the comparative learning system, the vibration and

Image features were compared with the text features encoded in the transformer to generate a common multimodal feature vector.

Generating a device health assessment based on S2 involves the following steps:

S201: Constructing device-node relationships, calculating

Error propagation weights and generating a topological

Device relationship feature graphs;

Constructing the topological graph structure, calculating the error propagation influence factor embedded in the nodes, and extracting critical path features to generate the topological relationship feature map.

Symbols to generate a topological relationship feature map.

S202: Segment the plant's operational data, merge the historical data

Dynamically analyze operating status data and weight the time step characteristics to determine the

To generate an assessment of the plant's condition. LU601247

The topological feature map is fed into a spatiotemporal graph convolution mesh that fuses and dynamically weights the time step segments of the operational data to generate a state assessment.

Generating a dynamically optimized set of process parameters on the

The basis of dynamic optimization in S3 includes the following steps:

S301: Assignment of process parameter intelligences, initialization of the strategy network and generation of a cooperative multi-parameter control strategy network;

Assignment of independent value networks for process parameters and construction of a

Co-optimization strategy network through a parameter-sharing mechanism.

S302: Simulation of extreme working condition data, adaptation of the boundary conditions of the

Strategy network and generation of a dynamically optimized process parameter set.

The strategy network is invoked to impose extreme disturbances in the operating conditions, and confidence interval optimization is applied to the

to adjust parameter restrictions.

The generation of a causality-driven process improvement scheme on the

The foundation of S4 comprises the following steps:

S401: Construction of a causal graph of process parameters and

Quality indicators, elimination of pseudocorrelated variables and generation of causal relationships

Networks of process parameters and quality indicators;

Application of a causal discovery algorithm to create a directed acyclic

to construct graphs, eliminating pseudocorrelated variables to establish a causal relationship

To create a network.

S402: Simulation of the effects of parameter changes, search for the optimal parameters

Parameter combinations that meet the boundary conditions, and creation of a causally driven process improvement plan.

Implementation of measures based on structural equation models and

Search for Pareto-optimal solutions through restricted Bayesian optimization.

An industrial big data processing system based on artificial intelligence algorithms, comprising the following modules: a module for processing multimodal features, a module for assessing plant condition, a module for

Decision-making in process optimization and a module for improving the

Causal analysis;

The module for processing multimodal features processes vibration signals to

Generating spectrograms, extracting image features, encoding text protocols, matching multimodal distributions, and generating multimodal common feature vectors;

The module for processing multimodal features includes a submodule for

Processing of vibration images, a sub-module for semantic text coding, and a

Submodule for feature alignment;

Module for evaluating the device's condition, which uses topological relationship graphs of the

Device constructed, dispersion weights calculated, time series operating condition data fused, and device condition assessments generated;

The module for assessing the condition of the plant includes a sub-module for creating topological relationships and a sub-module for convolution of time series graphs;

The process optimization decision-making module, which

assigns process parameter control strategies, intelligent bodies are trained to at 601247

Working together to achieve optimization goals, extreme working conditions were simulated to adjust parameters, and dynamically optimized process parameter sets were generated;

The process optimization decision-making module includes a sub-module for the

Collaboration of multiple intelligent bodies and a sub-module for simulating extreme conditions

Working conditions;

The causal analysis and improvement module constructs a parameter and

Quality causal network, simulates the effect of parameter interventions, iteratively searches for the optimal solution and generates a causally controlled process improvement scheme;

The module for improving causal analysis includes a sub-module for

Causality detection and a sub-module for counterfactual optimization.

Sub-module for processing vibration images, which generates spectrograms,

Frequency range characteristics extracted, noise eliminated, and common vibrations and

Image feature vectors generated;

Generation of a time-frequency spectrogram of the vibration signal by

Short-term Fourier transform, extraction of frequency domain features by a neural convolutional network, combination with a noise reduction algorithm (non-local mean denoising) and output of a common feature vector of vibration and image.

Submodule for semantic coding of text, coding of context semantics,

Extraction of fault description time correlation features, generation of a

Text feature vector;

Based on the bidirectional network of long-term and short-term memory,

Operational and maintenance logs are encoded, temporal association semantics are created by the

Attention mechanism extracted and text feature vectors generated.

Submodule for feature matching, matching of vibration, image and

Text feature distributions, generation of multimodal common feature vectors,

Generation of multimodal common feature vectors.

Adoption of the algorithm for multimodal adverse domain matching,

Matching the cross-modal distribution by Wasserstein distance optimization in the common hidden space, outputting a 256-dimensional multimodal common feature vector.

Submodule for constructing the topological relationship, constructs the relationship between the device nodes, defines the weights of the fault propagation paths and generates a feature map for the topological relationship of the devices;

Construction of the topological graph of the equipment, based on the

Graph Attention Network = Calculation of error propagation weights between nodes,

Generation of feature maps of the topological relationship of the equipment.

Submodule for convolution of the time diagram, which fuses historical operating condition data, dynamically weights time step characteristics and generates assessments of the plant condition.

Input of the topology map into the spatiotemporal convolution network, merging of the historical time series data of working conditions, and output of the 0-1 interval health status assessment through dynamic weighting of the

Time step characteristics via the gating cycle unit.

Multi-Intelligence Body Synergy Submodule, assigns process parameters independently &-V601 247

intelligences and generates a synergistic multi-parameter control strategy network;

Assign deep deterministic strategy gradient intelligences to process parameters to create a collaborative multi-parameter control strategy network through the

To construct a parameter-sharing mechanism.

Submodule for the simulation of extreme conditions, which provides data for

Extreme conditions are created, the strategy network is optimized, and dynamically optimized

Process parameter sets generated.

Generation of data on extreme working conditions based on an adverse conditional generation network, using a proximal algorithm

Strategy optimization to adapt to the constraints of the strategic network, and

Output of dynamically optimized process parameter sets.

Causal discovery submodule, construction of parameter and quality causal graphs,

Elimination of pseudocorrelation variables and generation of process parameter and

Quality causal networks:

The PC causal recognition algorithm is used to perform a directed acyclic

Constructing graphs of parameters and quality, eliminating pseudocorrelated edges through a conditional independence test, and establishing a causal process parameter and

To create a quality network.

Submodule for counterfactual optimization for simulating parameter interventions, for finding optimal parameter combinations that satisfy constraints, and for

Creation of causally driven process improvement plans.

Based on the causal network, a structural equation model is created which

The intervention effect is simulated by do-acting, and a Bayesian

The optimization algorithm is combined to find a Pareto-optimal solution and output a causally driven process improvement scheme.

Although embodiments of the present invention have been shown and described, it is understandable to a person with ordinary technical knowledge that a multitude of changes, modifications, substitutions and variations to these are possible.

embodiments can be made without deviating from the principles and spirit of the present invention, and the scope of the present invention is limited by the attached claims and their equivalents.

Claims (10)

Claims LU601247 1. An industrial big data processing method based on artificial intelligence algorithms, characterized by the following steps: S1: Based on heterogeneous data from multiple sources of industrial equipment, processing vibration signals to generate spectrograms, extracting image features, semantic coding of text protocols, matching different modal feature distributions, and generating multimodal common feature vectors; S2: Based on the multimodal common feature vector, the topological relationship map of the equipment is constructed, the law on the evolution of the equipment condition is extracted, the data on the operating condition are integrated to calculate the weighting of the node's health condition, and the equipment health condition assessment is generated; S3: The equipment condition assessment and the process parameters are combined, the intelligent body is trained to control the process variables, extreme operating condition data are simulated to adapt the strategy network, and the dynamically optimized process parameter set is generated. S4: Construction of the causal diagram of process parameters and quality indicators, simulation of the effects of parameter adjustments, iterative search for the optimal range of values of key parameters and creation of causally controlled process improvement plans. 2. An industrial big data processing method based on artificial intelligence algorithms according to claim 1, characterized in that: the generation of a multimodal common feature vector based on S1 comprises the following steps: S101: processing the vibration signal to generate a spectrogram, extracting local features, eliminating sensor noise, and generating a common feature vector of vibration and image; S102: semantic coding of the operating and maintenance logs, combining the common feature vectors of vibration and image to match the distribution, and generating the multimodal common feature vector. 3. An industrial big data processing method based on artificial intelligence algorithms according to claim 1, characterized in that: generating an equipment health assessment based on the equipment in S2 comprises the following steps: S201: constructing outage node relationships, calculating fault propagation weights, and generating a topological equipment relationship feature graph; S202: segmenting the plant's operational data, merging historical operational state data, dynamically weighting time step features, and generating a plant health assessment. 4. An industrial big data processing method based on artificial intelligence algorithms according to claim 1, characterized in that: the generation of a dynamically optimized set of process parameters based on the dynamic Optimization in S3 includes the following steps: LUB01247 S301: Assigning process parameter intelligences, initializing a strategy network and generating a multi-parameter cooperative control strategy network; S302: Simulating extreme operating condition data, adjusting the boundary conditions of the strategy network and generating the dynamically optimized process parameter set. 5. An industrial big data processing method based on artificial intelligence algorithms according to claim 1, characterized in that: the generation of a causality-driven process improvement scheme based on S4 comprises the following steps: S401: constructing a causal map of process parameters and quality indicators, eliminating pseudocorrelated variables and generating a causal process parameter and quality network; S402: simulating the effect of parameter interventions, searching for the optimal parameter combinations that satisfy the constraints, and generating a causality-driven process improvement scheme. 6. An industrial big data processing system based on artificial intelligence algorithms, characterized in that the industrial big data processing system comprises the following modules: a multimodal feature processing module, a plant condition assessment module, a process optimization decision-making module, and a causal analysis improvement module; The multimodal feature processing module processes vibration signals to generate spectrograms, extracts image features, encodes text logs, matches multimodal distributions, and generates multimodal common feature vectors; The multimodal feature processing module includes a vibration image processing submodule, a semantic text encoding submodule, and a feature alignment submodule; The plant condition assessment module constructs topological relationship graphs of the plant, calculates propagation weights, fuses time-series operating condition data, and generates plant condition values; The device condition assessment module includes a submodule for constructing topological relationships and a submodule for convolution of time series graphs; the process optimization decision-making module assigns process parameter control strategies, trains intelligences to collaborate on optimization goals, simulates extreme operating conditions to adjust parameters, and generates dynamically optimized process parameter sets; the process optimization decision-making module includes a submodule for multi-intelligent body collaboration and a submodule for simulating extreme operating conditions; the causal analysis and improvement module constructs a causal parameter and quality network, simulates the effects of parameter interventions, iteratively searches for the optimal solution, and generates a causally driven process improvement scheme; the causal analysis improvement module includes a submodule for causal detection and a submodule for counterfactual optimization. 7. An industrial big data processing system based on algorithms of artificial intelligence according to claim 6, characterized in that: a sub-module for processing vibration images generates a spectrogram, extracts frequency domain features, eliminates noise, and generates a common vibration and image feature vector; a sub-module for semantic coding of text that encodes contextual semantics, extracts temporal association features of the fault description, and generates text feature vectors; a sub-module for feature alignment that aligns vibration, image, and text feature distributions and generates multimodal common feature vectors. 8. An industrial big data processing system based on artificial intelligence algorithms according to claim 6, characterized in that: a submodule for constructing topological relationships that constructs connection relationships between device nodes, defines weights for fault propagation paths and generates feature maps for topological relationships between devices; a time sequence graph convolution submodule that fuses historical operating state data, dynamically weights time step features and generates an equipment condition assessment. 9. An industrial big data processing system based on artificial intelligence algorithms according to claim 6, characterized in that: a synergistic sub-module with multiple intelligences that assigns independent intelligences to the control process parameters and generates a synergistic multi-parameter control strategy network; a sub-module for simulating extreme operating conditions that generates data on extreme operating conditions, optimizes the strategy network, and generates a dynamically optimized set of process parameters. 10. An industrial big data processing system based on artificial intelligence algorithms according to claim 6, characterized in that: a sub-module for causal discovery that constructs causal graphs for parameters and quality, eliminates pseudocorrelated variables, and generates causal networks for process parameters and quality; a sub-module for counterfactual optimization that simulates parameter interventions, searches for optimal parameter combinations that satisfy constraints, and generates causally controlled process improvement schemes.