WO2024159986A1 - 一种无线局域网络动态阈值参数生成方法及装置 - Google Patents

一种无线局域网络动态阈值参数生成方法及装置 Download PDF

Info

Publication number
WO2024159986A1
WO2024159986A1 PCT/CN2023/142126 CN2023142126W WO2024159986A1 WO 2024159986 A1 WO2024159986 A1 WO 2024159986A1 CN 2023142126 W CN2023142126 W CN 2023142126W WO 2024159986 A1 WO2024159986 A1 WO 2024159986A1
Authority
WO
WIPO (PCT)
Prior art keywords
threshold
target
evaluation model
parameters
parameter
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/142126
Other languages
English (en)
French (fr)
Inventor
黄启圣
张耀东
王子晟
刘昕颖
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
ZTE Corp
Original Assignee
ZTE Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by ZTE Corp filed Critical ZTE Corp
Priority to EP23919547.2A priority Critical patent/EP4642079A4/en
Publication of WO2024159986A1 publication Critical patent/WO2024159986A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0806—Configuration setting for initial configuration or provisioning, e.g. plug-and-play
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0813—Configuration setting characterised by the conditions triggering a change of settings
    • H04L41/082—Configuration setting characterised by the conditions triggering a change of settings the condition being updates or upgrades of network functionality
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/142—Network analysis or design using statistical or mathematical methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/16—Threshold monitoring
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/06—Testing, supervising or monitoring using simulated traffic
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0823—Errors, e.g. transmission errors
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0823—Errors, e.g. transmission errors
    • H04L43/0829—Packet loss
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level
    • H04L43/0888—Throughput
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W84/00—Network topologies
    • H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
    • H04W84/10—Small scale networks; Flat hierarchical networks
    • H04W84/12—WLAN [Wireless Local Area Networks]

Definitions

  • the embodiments of the present disclosure relate to the field of wireless local area networks, and in particular, to a method and device for generating dynamic threshold parameters of a wireless local area network.
  • WIFI products usually use fixed threshold parameters and adaptive algorithms to implement channel selection or bandwidth expansion functions, such as automatic channel selection algorithms based on threshold parameters, dynamic bandwidth algorithms based on thresholds, dynamic spatial multiplexing threshold algorithms, etc.
  • a set of fixed threshold parameters is difficult to apply to all environments.
  • the embodiments of the present disclosure provide a method and device for generating dynamic threshold parameters in a wireless local area network, so as to at least solve the problem in the related art that fixed threshold parameters are difficult to apply to all network environments.
  • a method for generating dynamic threshold parameters in a wireless local area network comprising:
  • the target threshold parameter configuring the target threshold parameter to an edge computing node connected to the wireless access node, wherein the edge computing node is used to run a preset adaptive algorithm according to the target threshold parameter;
  • the threshold evaluation model is updated using the target label pair, and a new target threshold parameter is determined using the updated threshold evaluation model.
  • a device for generating dynamic threshold parameters of a wireless local area network comprising:
  • a determination module configured to determine a target threshold parameter through a threshold evaluation model pre-trained by a server
  • a configuration module configured to configure the target threshold parameter to an edge computing node connected to the wireless access node, Wherein, the edge computing node is used to run a preset adaptive algorithm according to the target threshold parameter;
  • a collection module configured to collect target statistical parameters corresponding to the target threshold parameters from the edge computing node to obtain a target label pair
  • An updating module is configured to update the threshold evaluation model using the target label pair, and determine a new target threshold parameter through the updated threshold evaluation model.
  • a computer-readable storage medium in which a computer program is stored.
  • the computer program is executed by a processor, the steps in any one of the above method embodiments are executed.
  • an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
  • FIG. 1 is a hardware structure block diagram of a mobile terminal of a method for generating dynamic threshold parameters of a wireless local area network according to an embodiment of the present disclosure
  • FIG2 is a flow chart of a method for generating dynamic threshold parameters in a wireless local area network according to an embodiment of the present disclosure
  • FIG3 is a structural diagram of a dynamic threshold parameter generation system according to an embodiment of the present disclosure.
  • FIG4 is a flow chart of timing training model and updating threshold parameters according to an embodiment of the present disclosure
  • FIG5 is a flow chart of model training and threshold updating according to an embodiment of the present disclosure.
  • FIG6 is a block diagram of a device for generating dynamic threshold parameters for a wireless local area network according to an embodiment of the present disclosure.
  • FIG1 is a hardware structure block diagram of a mobile terminal of the method for generating dynamic threshold parameters of a wireless local area network in the embodiment of the present disclosure.
  • the mobile terminal may include one or more (only one is shown in FIG1) processors 102, and the processor 102 may include but is not limited to a microcontroller unit (MCU) or a programmable logic device (FPGA) and a memory 104 configured to store data, wherein the mobile terminal may also include a transmission device 106 and an input and output device 108 configured to have a communication function.
  • MCU microcontroller unit
  • FPGA programmable logic device
  • FIG1 is only for illustration and does not limit the structure of the mobile terminal.
  • the mobile terminal may also include more or fewer components than those shown in FIG1, or have a configuration different from that shown in FIG1.
  • the memory 104 may be configured to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for generating dynamic threshold parameters of a wireless local area network in the embodiment of the present disclosure.
  • the processor 102 executes various functional applications and service chain address pool slice processing by running the computer program stored in the memory 104, that is, implementing the above method.
  • the memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may be
  • the first step includes a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.
  • the transmission device 106 is configured to receive or send data via a network.
  • Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal.
  • the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet.
  • the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is configured to communicate with the Internet wirelessly.
  • RF Radio Frequency
  • FIG2 is a flow chart of a method for generating dynamic threshold parameters of a wireless local area network according to an embodiment of the present disclosure. As shown in FIG2, the process includes the following steps:
  • Step S202 determining a target threshold parameter through a threshold evaluation model pre-trained by the server
  • Step S204 configuring the target threshold parameter to an edge computing node connected to the wireless access point (AP), wherein the edge computing node is used to run a preset adaptive algorithm according to the target threshold parameter;
  • Step S206 collecting target statistical parameters corresponding to the target threshold parameters from the edge computing node to obtain a target label pair
  • Step S208 Use the target label pair to update the threshold evaluation model, and determine a new target threshold parameter through the updated threshold evaluation model.
  • the method before step S202, further includes: obtaining initial training parameters of the threshold evaluation model obtained through pre-training from a server; and initializing the threshold evaluation model using the initial training parameters to obtain the pre-trained threshold evaluation model.
  • the initial training parameters of the threshold evaluation model are ⁇ (w,b), where w is the weight and b is the offset;
  • the threshold evaluation model is
  • the neural network may include: Deep Neural Networks (DNN), Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN).
  • DNN Deep Neural Networks
  • CNN Convolutional Neural Networks
  • RNN Recurrent Neural Networks
  • Activation functions may include: Linear Rectification (ReLU) activation function, Hyperbolic Tangent (Tanh) activation function or Sigmoid activation function, etc.
  • ReLU Linear Rectification
  • Tih Hyperbolic Tangent
  • Sigmoid activation function etc.
  • step S202 may specifically include the following steps:
  • Step S2022 selecting a plurality of threshold parameters from a threshold parameter space by a search algorithm, wherein the threshold parameter space is composed of a plurality of preset threshold parameters;
  • Step S2024 determining statistical parameters corresponding to the multiple threshold parameters respectively according to the pre-stored multiple label pairs or the threshold evaluation model
  • Step S2026 Determine a threshold parameter with the largest statistical parameter among the multiple threshold parameters as the target threshold parameter.
  • the search algorithm in step S2022 can randomly select a threshold parameter from the threshold parameter space, or can select a threshold parameter from the threshold parameter space according to the rules of the selected search algorithm itself, such as depth-first search, breadth-first search, best-first search, etc.
  • the search algorithm may include a reinforcement learning algorithm or a genetic algorithm.
  • step S2024 may specifically include: determining whether the threshold parameter exists in the multiple label pairs; if the judgment result is yes, determining the statistical parameter corresponding to the threshold parameter from the label pairs; if the judgment result is no, inputting the threshold parameter into the threshold evaluation model, and determining the output result of the threshold evaluation model as the statistical parameter corresponding to the threshold parameter.
  • step S204 may specifically include: sending the target threshold parameter to the edge computing node, wherein the edge computing node adjusts the preset adaptive algorithm according to the target threshold parameter; during the channel access and information transmission process, running the preset adaptive algorithm through the edge computing node.
  • the preset adaptive algorithms may specifically include: automatic channel selection algorithm, dynamic bandwidth algorithm, dynamic spatial multiplexing threshold algorithm, EDCA (Enhanced Distributed Channel Access) parameter adjustment algorithm, etc.
  • the target threshold parameters include at least one of the following: a noise threshold, a channel utilization threshold, a weighting factor for channel scoring, a packet error rate threshold, and a transmission power threshold.
  • the target threshold parameters can also be selected according to the application scenario and the corresponding adaptive algorithm. For example, when selecting a channel, you can also select any of the following threshold parameters: adjacent channel interference threshold, received signal strength indicator (RSSI) threshold of the main channel, RSSI threshold of the extended channel, and the number of APs on the same frequency.
  • RSSI received signal strength indicator
  • step S206 may specifically include the following steps:
  • Step S2062 collecting target statistical parameters generated by the edge computing node in the process of running the adaptive algorithm using the target threshold parameter
  • Step S2064 normalizing the target threshold parameter by the edge computing node to obtain the normalized target threshold parameter
  • Step S2066 combining the normalized target threshold parameter and the target statistical parameter to obtain the target label pair.
  • step S2062 may specifically include: in the process of running the adaptive algorithm using the target threshold parameter, obtaining a normalized throughput and/or a packet sending success rate from the edge computing node, wherein the statistical parameters include the normalized throughput and/or the packet sending success rate, the normalized throughput is the ratio of the number of successfully sent load bits to the total number of sent load bits, and the packet sending success rate is the ratio of the number of successfully sent MAC protocol data units to the total number of sent MAC protocol data units.
  • step S2064 may specifically include:
  • the target threshold parameter is normalized by the following formula:
  • s norm is the target threshold parameter after the normalization process
  • si is the target threshold parameter.
  • step S208 may specifically include: at the wireless access node, training the threshold evaluation model according to the gradient descent method and the target label pair to obtain training parameters, wherein the training parameters include weights and offsets; and using the training parameters to update the threshold evaluation model.
  • the loss function of the threshold evaluation model is:
  • the method may further include the following steps:
  • Step S210 the plurality of wireless access nodes respectively report a plurality of training parameters obtained by training the threshold evaluation model according to the newly collected target tags to the server;
  • Step S212 updating, through the server, initial training parameters of the threshold evaluation model of the server according to the federated learning algorithm and the multiple training parameters.
  • step S212 may specifically include the following steps:
  • Step S2122 Update the total loss value loss_total of the threshold evaluation model of the server according to the loss values of the threshold evaluation models of the multiple wireless access nodes:
  • loss_total ⁇ L loss l , where loss l is the loss value of the threshold evaluation model of the l-th wireless access node;
  • Step S2124 updating a second weight and a second offset value of the threshold evaluation model of the server according to the first weight and the first offset value of the threshold evaluation model of the plurality of wireless access nodes, wherein the initial training parameters include the second weight and the second offset value:
  • wi wi + ⁇ L ⁇ wl
  • bi bi + ⁇ L ⁇ bl
  • wi is the second weight
  • bi is the second offset value
  • ⁇ wl is the first weight of the threshold evaluation model of the lth radio access node
  • ⁇ bl is the first offset value of the threshold evaluation model of the lth radio access node.
  • the evaluation of the threshold parameters can be achieved by collecting specific statistical parameters, and then the threshold parameters can be dynamically adjusted according to the network environment, thereby solving the problem that the fixed threshold parameters used in the related technology are difficult to apply to all network environments, and realizing the determination of the optimal threshold parameters according to the changes in the network environment, thereby improving the system throughput, reducing the transmission delay, and improving the user experience.
  • FIG3 is a structural diagram of a dynamic threshold parameter generation system according to an embodiment of the present disclosure. As shown in FIG3 , the system includes the following structures: a platform server 32 , an AP central controller 34 , and an AP (wireless access node) 36 .
  • a platform server 32 the system includes the following structures: a platform server 32 , an AP central controller 34 , and an AP (wireless access node) 36 .
  • AP wireless access node
  • the dynamic threshold parameter generation system is a networking system, so there is an AP central controller.
  • each structure in the system includes a computing core with computing power.
  • An intelligent module configured to dynamically generate thresholds is deployed on the computing core, and the input of the computing core is a threshold parameter vector and a label pair corresponding to the throughput.
  • the computing core on the AP (wireless access node) 36 can be just a system-on-chip (SoC), and the computing cores in the platform server 32 and the AP central controller 34 with strong computing power can undertake the main training of the threshold evaluation model.
  • SoC system-on-chip
  • the edge AP 36 only needs to undertake the inference work or receive the model parameters and the initialized threshold value for simple training and optimal threshold inference.
  • the platform computing core of the platform server 32 trains the threshold evaluation model based on Federated Learning.
  • the platform computing core receives the back-propagation gradient from the controller computing cores of all AP central controllers 34, thereby training to obtain a parameter (i.e., initial training parameter) that can be initialized with all platform computing cores.
  • the platform computing core provides a generalized template for generating dynamic threshold parameters.
  • the computing core of the AP central controller 34 is configured to receive the tag pairs from the AP, thereby completing the mapping from the threshold parameter to the score. After completing the mapping from the threshold parameter to the score, the computing core of the AP central controller 34 will start the search algorithm to generate the current optimal threshold, wherein the search algorithm can be a genetic algorithm or a reinforcement learning algorithm.
  • the reasoning and search functions can also be combined into one and unified by the Deep Reinforcement Learning (DQN) algorithm.
  • DQN Deep Reinforcement Learning
  • the edge computing core on the AP (wireless access node) 36 mainly works as follows: setting the current threshold and obtaining the statistical parameters currently used to evaluate the pros and cons of the current threshold.
  • This statistical parameter is usually the successful transmission rate or throughput of the MAC protocol data unit (MAC Protocol Data Unit, MPDU for short) over a period of time.
  • model parameters refer to the trainable parameters that define the deep learning network: weights and offsets.
  • the network structure of the deep learning network is pre-set and will not change over time.
  • the statistical parameter refers to a relevant statistical parameter used to evaluate the quality of the current threshold, such as the MPDU transmission success rate, throughput, etc.
  • the simplified dynamic threshold parameter generation system may be associated with only one AP and one station (STA).
  • the AP will integrate the computing core of the AP central controller 34 and the edge core on the AP (wireless access node) 36 at the same time.
  • the simplified dynamic threshold parameter generation system may not deploy a central controller, and the AP directly interacts with the platform server 32.
  • the AP returns the traditional measurement parameters to the platform server 32, and the core computing node of the platform server 32 performs the function of the AP central controller 34.
  • FIG4 is a flow chart of timing training model and updating threshold parameters according to an embodiment of the present disclosure. As shown in FIG4 , the flow includes the following steps:
  • Step S402 determining whether the update time has been reached
  • Step S404 start model training and threshold update.
  • the process can be used to implement the timed training of the threshold evaluation model and the dynamic adjustment of the threshold parameters.
  • FIG5 is a flow chart of model training and threshold updating according to an embodiment of the present disclosure. As shown in FIG5 , the flow includes the following steps:
  • Step S502 determining whether it is the first training
  • Step S504 the platform performs model pre-training and sends model parameters
  • Step S506 the central controller initializes the model and searches for the optimal threshold
  • Step S508 the AP starts the adaptive algorithm according to the optimal threshold parameter
  • Step S510 the AP measures and collects statistical parameters for training the model
  • Step S512 the AP returns a tag pair (threshold, statistical parameters), and the central controller updates the model according to the tag pair;
  • Step S514 the central controller sends the gradient, and the platform updates the training model
  • Step S516 The central controller uses the updated model to search for a new threshold and sends it down.
  • step S504 is executed, and if the judgment result of step S502 is no, the process directly jumps to step S508.
  • step S514 and step S516 may be performed simultaneously.
  • the platform sends the pre-trained model inference parameters ⁇ (w, b) to the core computing nodes of each central controller as model initialization.
  • the core computing node deploys the threshold to the AP for use by the adaptive algorithm, and the statistical parameters (normalized throughput, successful transmission rate of MPDU) used for evaluation are obtained during the operation of the AP.
  • the edge computing node normalizes the threshold parameter, it forms a (threshold, statistical parameter) label pair and feeds it back to the core computing node to train the evaluation network.
  • the central controller updates the threshold evaluation model (deep learning module) using the training label pairs (threshold parameters, statistical parameters) collected in step S512.
  • the key parameters of the threshold evaluation model may specifically include the following:
  • Input training labels (threshold parameters, statistical parameters) sent back by the edge computing core.
  • Output The prediction result of the threshold evaluation model, that is, the evaluation parameter corresponding to the current threshold. The higher the value, the better the effect of the current threshold.
  • the threshold evaluation model is usually a type of parameterized function, expressed as follows:
  • the final parameterized function is formed by cascading the basic network unit functions, and its general expression is as follows:
  • ⁇ represents the network operator.
  • the commonly used networks are fully connected networks (Dense Neural Network, DNN for short), convolutional neural networks (CNN), and recurrent neural networks (RNN).
  • f a (x) represents the activation function.
  • Common activation functions include Relu activation function, Tanh activation function, Sigmoid activation function, etc.
  • Loss function represents the gap between the threshold evaluation model and the actual situation.
  • the loss function uses the mean square error and is defined as follows:
  • cross entropy can also be used as the cost function:
  • Training method The trainer is called at the end of each cycle. At this time, the network will use the new labels collected in real time during this cycle to train the existing network.
  • the training object is the trainable parameters w i and b i.
  • the training method is gradient descent. Commonly used gradient descent methods include RMSProp and Adam algorithm.
  • the search function i.e., the above-mentioned search algorithm
  • the search algorithm for the threshold parameters can be reinforcement learning, genetic algorithm, etc.
  • the deep learning module will be set to evaluate the currently searched threshold and complete the mapping from the threshold to the evaluation value.
  • the search algorithm may be a threshold generation algorithm based on Deep-Q-learning.
  • the search algorithm is: Q-learning algorithm and the Q function is replaced by a deep learning module.
  • the reward function is initialized to 0, and the action is a modification of the threshold.
  • the search algorithm can also be based on a deep learning evaluation module and a threshold generation algorithm.
  • the search algorithm is a genetic algorithm, which is used to search for the optimal threshold.
  • the population individuals are a set of threshold schemes, the gene is a specific threshold parameter, and the evaluation function of the genetic algorithm is a deep learning module.
  • the loss function of the platform side is the sum of the statistical parameters of all core controllers it manages:
  • loss l represents the loss value from the lth central control device. Therefore, the gradient generated in step 3 needs to be sent to the platform to update the trainable parameters of the model stored in the platform.
  • the update method is as follows:
  • ⁇ w l and ⁇ b l are the gradients from the lth core node. This approach will ensure that the platform can provide a model with extremely high generalization when initializing the model.
  • FIG. 6 is a block diagram of a device for generating dynamic threshold parameters of a wireless local area network according to an embodiment of the present disclosure. As shown in FIG. 6, the device includes:
  • a determination module 62 configured to determine a target threshold parameter through a threshold evaluation model pre-trained by a server
  • a configuration module 64 configured to configure the target threshold parameter to an edge computing node connected to the wireless access node, wherein the edge computing node is used to run a preset adaptive algorithm according to the target threshold parameter;
  • a collection module 66 is configured to collect target statistical parameters corresponding to the target threshold parameters from the edge computing node to obtain a target label pair;
  • the updating module 68 is configured to update the threshold evaluation model using the target label pair, and determine a new target threshold parameter through the updated threshold evaluation model.
  • An embodiment of the present disclosure further provides a computer-readable storage medium, in which a computer program is stored.
  • a computer program is stored.
  • the steps of any of the above method embodiments are executed.
  • the above-mentioned computer-readable storage medium may include, but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.
  • An embodiment of the present disclosure further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
  • the electronic device may further include a transmission device and an input/output device, wherein the transmission device is connected to the processor, and the input/output device is connected to the processor.
  • modules or steps of the present disclosure described above can be implemented by general computing.
  • the present invention may be implemented by a computer system, they may be concentrated on a single computing device, or distributed on a network of multiple computing devices, they may be implemented by program codes executable by the computing device, so that they may be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described may be performed in a different order than that shown here, or they may be made into individual integrated circuit modules, or multiple modules or steps thereof may be made into a single integrated circuit module.
  • the present disclosure is not limited to any particular combination of hardware and software.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开实施例提供了一种无线局域网络动态阈值参数生成方法及装置，该方法包括：通过服务器预先训练好的阈值评估模型确定目标阈值参数，将目标阈值参数配置到与无线接入节点连接的边缘运算节点，从边缘运算节点采集与目标阈值参数对应的目标统计参数，得到目标标签对，使用目标标签对更新阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。

Description

一种无线局域网络动态阈值参数生成方法及装置
相关申请的交叉引用
本公开基于2023年2月1日提交的发明名称为“一种无线局域网络动态阈值参数生成方法及装置”的中国专利申请2023101459860，并且要求该专利申请的优先权，通过引用将其所公开的内容全部并入本公开。
技术领域
本公开实施例涉及无线局域网络领域，具体而言，涉及一种无线局域网络动态阈值参数生成方法及装置。
背景技术
现有的WIFI产品通常是采用固定阈值参数和自适应算法来实现信道选择或带宽扩展等功能。比如基于阈值参数的自动信道选择算法，基于阈值的动态带宽算法，动态空间复用阈值算法等等。但是，一套固定的阈值参数难以适用于所有环境中。
因此，需要设计一套可以根据环境动态调节阈值参数的系统从而能够最大程度的发挥自适应算法的功能。该问题的难点在于完成从阈值参数到评价值的映射，由于实际环境复杂且多变，很难直接抽象出一套泛化性很好的数学模型来完成此映射。
综上，针对相关技术中采用固定阈值参数难以适用于所有网络环境的问题，尚未提出解决方案。
发明内容
本公开实施例提供了一种无线局域网络动态阈值参数生成方法及装置，以至少解决相关技术中采用固定阈值参数难以适用于所有网络环境的问题。
根据本公开的一个实施例，提供了一种无线局域网络动态阈值参数生成方法，所述方法包括：
通过服务器预先训练好的阈值评估模型确定目标阈值参数；
将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点，其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
根据本公开的又一个实施例，提供了一种无线局域网络动态阈值参数生成装置，所述装置包括：
确定模块，设置为通过服务器预先训练好的阈值评估模型确定目标阈值参数；
配置模块，设置为将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点， 其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
采集模块，设置为从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
更新模块，设置为使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
根据本公开的又一个实施例，还提供了一种计算机可读的存储介质，所述存储介质中存储有计算机程序，其中，所述计算机程序被处理器运行时执行上述任一项方法实施例中的步骤。
根据本公开的又一个实施例，还提供了一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行上述任一项方法实施例中的步骤。
附图说明
图1是本公开实施例的无线局域网络动态阈值参数生成方法的移动终端的硬件结构框图；
图2是本公开实施例的一种无线局域网络动态阈值参数生成方法的流程图；
图3是根据本公开实施例的动态阈值参数生成系统的结构图；
图4是根据本公开实施例的定时训练模型和更新阈值参数的流程图；
图5是根据本公开实施例的模型训练和阈值更新的流程图；
图6是根据本公开实施例的无线局域网络动态阈值参数生成装置的框图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本公开的实施例。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
本公开实施例中所提供的方法实施例可以在移动终端、计算机终端或者类似的运算装置中执行。以运行在移动终端上为例，图1是本公开实施例的无线局域网络动态阈值参数生成方法的移动终端的硬件结构框图，如图1所示，移动终端可以包括一个或多个(图1中仅示出一个)处理器102，处理器102可以包括但不限于微控制单元(Microcontroller Unit，简称为MCU)或可编程逻辑器件(Field Programmable Gate Array，简称为FPGA)等处理装置和设置为存储数据的存储器104，其中，上述移动终端还可以包括设置为通信功能的传输设备106以及输入输出设备108。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述移动终端的结构造成限定。例如，移动终端还可包括比图1中所示更多或者更少的组件，或者具有与图1所示不同的配置。
存储器104可设置为存储计算机程序，例如，应用软件的软件程序以及模块，如本公开实施例中的无线局域网络动态阈值参数生成方法对应的计算机程序，处理器102通过运行存储在存储器104内的计算机程序，从而执行各种功能应用以及业务链地址池切片处理，即实现上述的方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进 一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至移动终端。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
传输装置106设置为经由一个网络接收或者发送数据。上述的网络具体实例可包括移动终端的通信供应商提供的无线网络。在一个实例中，传输装置106包括一个网络适配器(Network Interface Controller，简称为NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输装置106可以为射频(Radio Frequency，简称为RF)模块，其设置为通过无线方式与互联网进行通讯。
在本实施例中提供了一种运行于上述移动终端或网络架构的无线局域网络动态阈值参数生成方法，图2是本公开实施例的一种无线局域网络动态阈值参数生成方法的流程图，如图2所示，该流程包括如下步骤：
步骤S202，通过服务器预先训练好的阈值评估模型确定目标阈值参数；
步骤S204，将所述目标阈值参数配置到与所述无线接入节点(Access Point,简称AP)连接的边缘运算节点，其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
步骤S206，从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
步骤S208，使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
在本实施例中，在步骤S202之前，该方法还包括：从服务器获取所述阈值评估模型通过预训练得到的初始训练参数；使用所述初始训练参数对所述阈值评估模型进行初始化处理，得到所述预先训练好的阈值评估模型。
具体的，阈值评估模型的初始训练参数为θ(w,b)，其中，w为权重，b为偏移量；
阈值评估模型为
预先训练好的阈值评估模型为fi(x)＝fa(wi⊙x+bi)，其中，⊙为与神经网络对应的网络算子，fa(x)为激活函数。
进一步的，神经网络可以包括：深度神经网络(Deep Neural Networks，简称DNN)、卷积神经网络(Convolutional Neural Networks，简称CNN)或循环神经网络(Recurrent Neural Network，简称RNN)；
激活函数可以包括：线性整流(Linear Rectification，简称ReLU)激活函数、双曲正切(Hyperbolic Tangent，简称Tanh)激活函数或S型(Sigmoid)激活函数等。
在本实施例中，步骤S202具体可以包括如下步骤：
步骤S2022，通过搜索算法从阈值参数空间中选择多个阈值参数，其中，所述阈值参数空间由多个预设的阈值参数组成；
步骤S2024，根据预先存储的多个标签对或所述阈值评估模型分别确定所述多个阈值参数对应的统计参数；
步骤S2026，将所述多个阈值参数中统计参数最大的一个阈值参数确定为目标阈值参数。
在本实施例中，步骤S2022中的搜索算法可以从阈值参数空间中随机选择阈值参数，或者，可以根据所选择的搜索算法自身的规则从阈值参数空间中选择阈值参数，例如深度优先搜索、广度优先搜索、最佳优先搜索等。
在本实施例中，所述搜索算法可以包括强化学习算法或遗传算法。
在本实施例中，步骤S2024具体可以包括：判断所述多个标签对中是否存在所述阈值参数；在判断结果为是的情况下，从所述标签对中确定与所述阈值参数对应的统计参数；在判断结果为否的情况下，将所述阈值参数输入所述阈值评估模型，并将所述阈值评估模型的输出结果确定为与所述阈值参数对应的统计参数。
在本实施例中，步骤S204具体可以包括：将所述目标阈值参数下发给所述边缘运算节点，其中，所述边缘运算节点根据所述目标阈值参数调整所述预设的自适应算法；在信道接入和信息传输的过程中，通过所述边缘运算节点运行所述预设的自适应算法。
进一步的，预设的自适应算法具体可以包括：自动信道选择算法，动态带宽算法，动态空间复用阈值算法，EDCA(Enhanced Distributed Channel Access，增强的分布式信道访问)参数调整算法等等。
在本实施例中，在所述预设的自适应算法用于信道选择的情况下，所述目标阈值参数至少包括以下之一：噪声阈值、信道利用率阈值、信道评分用加权因子、错包率阈值、发送功率阈值。
进一步的，除上述典型的阈值参数以外，目标阈值参数也可以根据应用场景和对应的自适应算法进行选择，例如在信道选择时还可以选择以下任意的阈值参数：邻频干扰阈值、主信道的接收信号强度指示(Received Signal Strength Indicator，简称RSSI)阈值、扩展信道的RSSI阈值、同频AP个数阈值等。
在本实施例中，步骤S206具体可以包括以下步骤：
步骤S2062，采集所述边缘运算节点在使用所述目标阈值参数运行所述自适应算法过程中产生的目标统计参数；
步骤S2064，通过所述边缘运算节点对所述目标阈值参数进行归一化处理，得到归一化处理后的目标阈值参数；
步骤S2066，组合所述归一化处理后的目标阈值参数和所述目标统计参数得到所述目标标签对。
在本实施例中，步骤S2062具体可以包括：在使用所述目标阈值参数运行所述自适应算法的过程中，从所述边缘运算节点中获取归一化吞吐量和/或发包成功率，其中，所述统计参数包括所述归一化吞吐量和/或所述发包成功率，所述归一化吞吐量为成功发送的负载比特数和总发送的负载比特数的比值，所述发包成功率为MAC协议数据单元的成功发送数量和MAC协议数据单元的总发送数量的比值。
在本实施例中，步骤S2064具体可以包括：
通过以下公式对所述目标阈值参数进行归一化处理：
其中，snorm为所述归一化处理后的目标阈值参数，s＝[s1,s2,...,sn]为多个预设的阈值参数，si为所述目标阈值参数。
在本实施例中，步骤S208具体可以包括：在所述无线接入节点，根据梯度下降法和所述目标标签对对所述阈值评估模型进行训练，得到训练参数，其中，所述训练参数包括权重和偏移量；使用所述训练参数更新所述阈值评估模型。
具体的，所述阈值评估模型的损失函数loss为：
或者，
在本实施例中，该方法还可以包括以下步骤：
步骤S210，多个所述无线接入节点分别将根据新采集的目标标签对对所述阈值评估模型进行训练得到的多个训练参数上报给所述服务器；
步骤S212，通过所述服务器，根据联邦学习算法和所述多个训练参数更新所述服务器的阈值评估模型的初始训练参数。
在本实施例中，步骤S212具体可以包括以下步骤：
步骤S2122，根据所述多个无线接入节点的阈值评估模型的损失值更新所述服务器的阈值评估模型的总损失值loss_total：
loss_total＝∑Llossl，其中，lossl为第l个无线接入节点的阈值评估模型的损失值；
步骤S2124，根据所述多个无线接入节点的阈值评估模型的第一权重和第一偏移值更新所述服务器的阈值评估模型的第二权重和第二偏移值，其中，所述初始训练参数包括所述第二权重和所述第二偏移值：
wi＝wi+∑LΔwl，bi＝bi+∑LΔbl，其中，wi为所述第二权重，bi为所述第二偏移值，Δwl为第l个无线接入节点的阈值评估模型的第一权重，Δbl为第l个无线接入节点的阈值评估模型的第一偏移值。
在本公开实施例中，通过上述步骤S202至S208中的方法，通过采集特定的统计参数可以实现对阈值参数的评价评估，进而实现根据网络环境动态调整阈值参数，解决相关技术中采用固定阈值参数难以适用于所有网络环境的问题，实现了根据网络环境变化确定最佳的阈值参数，进而提升系统的吞吐量，降低传输时延，提升用户的体验。
图3是根据本公开实施例的动态阈值参数生成系统的结构图，如图3所示，该系统包括以下结构：平台服务器32，AP中央控制器34，AP(无线接入节点)36。
在本公开实施例中，动态阈值参数生成系统是一个组网系统，因此存在一个AP中央控制器。
在本实施例中，系统中的每个结构都包含一个具备运算能力的运算核心，针对运算能力来说，平台服务器32的运算能力>AP中央控制器34的运算能力>AP(无线接入节点)36的运算能力。运算核心上部署了设置为动态生成阈值的智能模块，运算核心的输入为阈值参数矢量和对应吞吐量的标签对。
具体的，AP(无线接入节点)36上的运算核心可以仅仅是一个系统级芯片(System on Chip，简称SoC)，运算能力较强的平台服务器32和AP中央控制器34中的运算核心可以承担主要的阈值评估模型的训练工作。而边缘的AP36仅需要承担推断工作或是接收模型参数和初始化的阈值进行简单的训练和最佳阈值推断。
在本实施例中，平台服务器32的平台运算核心是基于联邦学习(Federated Learning)对阈值评估模型进行训练的，平台运算核心接收来自所有AP中央控制器34的控制器运算核心的反向传播梯度，从而训练得到一个可用与所有平台运算核心进行初始化的参数(即初始训练参数)。简言之，平台运算核心提供泛化的动态阈值参数的生成模板。
在本实施例中，AP中央控制器34的运算核心设置为接收来自AP的标签对，从而完成由阈值参数到评分的映射。在完成了阈值参数到评分的映射以后，AP中央控制器34的运算核心就将启动搜索算法从而产生当前的最佳阈值，其中搜索算法可以是遗传算法或是强化学习算法。推理与搜索功能也可以合二为一由深度强化学习(Deep Reinforcement Learning，简称DQN)算法统一完成。
在本实施例中，AP(无线接入节点)36上的边缘运算核心主要工作如下：设置当前阈值并获取当前用于评估当前阈值优劣的统计参数，这个统计参数通常为一段时间内的MAC协议数据单元(MAC Protocol Data Unit，简称MPDU)的成功发送率或是吞吐量。
在本实施例中，模型参数指定义深度学习网络的可训练参数：权重与偏移量。深度学习网络的网络结构是预先设定好的不会随时间的变化而改变。
在本实施例中，统计参数指用于评价当前阈值的优劣的相关统计参数如MPDU的发送成功率，吞吐量等。
在另一实施例中，简化的动态阈值参数生成系统可以只有一个AP和一个站点(Station，简称STA)进行关联。此时AP中将同时集成AP中央控制器34的运算核心和AP(无线接入节点)36上的边缘核心。
在另一实施例中，简化的动态阈值参数生成系统也可以不部署中央控制器，AP与平台服务器32直接交互。AP回传统计参数至平台服务器32，由平台服务器32的核心计算节点代行AP中央控制器34的功能。
图4是根据本公开实施例的定时训练模型和更新阈值参数的流程图，如图4所示，该流程包括以下步骤：
步骤S402，判断是否达到更新时间；
步骤S404，启动模型训练和阈值更新。
在本实施例中，通过该流程可以实现对阈值评估模型的定时训练以及对阈值参数的动态调整。
图5是根据本公开实施例的模型训练和阈值更新的流程图，如图5所示，流程包括以下步骤：
步骤S502，判断是否为第一次训练；
步骤S504，平台进行模型预训练并下发模型参数；
步骤S506，中央控制器初始化模型并搜索最佳阈值；
步骤S508，AP根据最佳阈值参数启动自适应算法；
步骤S510，AP测量并收集用于训练模型的统计参数；
步骤S512，AP回传标签对(阈值，统计参数)，中央控制器根据标签对更新模型；
步骤S514，中央控制器上送梯度，平台更新训练模型；
步骤S516，中央控制器利用更新的模型搜索新的阈值并下发。
在本实施例中，在步骤S502判断结果为是的情况下执行步骤S504，在步骤S502判断结果为否的情况下直接跳转步骤S508。
在本实施例中，步骤S514可以和步骤S516同步进行。
在本实施例中，平台会发送预训练完成的模型推理参数θ(w,b)至各个中央控制器的的核心运算节点中作为模型的初始化。
在本实施例中，核心运算节点部署阈值至AP中供自适应算法使用，AP运行过程中获取用于评价的统计参数(归一化吞吐量，MPDU的成功发送率)。边缘运算节点将阈值参数归一化以后，形成(阈值，统计参数)标签对反馈至核心运算节点以训练评价网络。
具体的，如果设阈值参数为s＝[s1,s2,...,sn]，则归一化过程如下所示：
进一步的，归一化吞吐量的计算公式如下：
MPDU的成功发送率的计算公式如下：
在本实施例中，中央控制器利用步骤S512中搜集的训练标签对(阈值参数，统计参数)更新阈值评估模型(深度学习模块)。
在本实施例中，阈值评估模型的关键参数具体可以包括以下内容：
输入：边缘运算核心回传的训练标签(阈值参数，统计参数)。
输出：阈值评估模型的预测结果，即对当前阈值对应的评价参数，值越高说明当前阈值的效果越好。
模型与可训练参数：阈值评估模型通常是一类参数化函数，表达式如下：
最终的参数化函数由基本网络单元函数级联形成，其一般表达式如下：
fi(x)＝fa(wi⊙x+bi)
其中⊙表示网络算子由神经网络类型定义常用的网络有全连接网络(Dense Neural Network,简称DNN)，卷积神经网络CNN，循环神经网络RNN。fa(x)表示激活函数，常用的激活函数有Relu激活函数，Tanh激活函数，Sigmoid激活函数等。
损失函数：表征阈值评估模型和实际的差距。损失函数采用均方误差其定义如下：
或者，由于网络器使用的是归一化的评价参数所以也可以用交叉熵作为代价函数：
训练方法：训练器在每个周期的结束时被调用，此时网络会利用本周期内实时搜集到的新的标签对已有的网络进行训练，训练对象为可训练参数wi，bi训练方法为梯度下降，常用的梯度下降方法有RMSProp，Adam算法等。
在本实施例中，搜索函数(即上述的搜索算法)将在阈值参数构成的空间中随机搜索直到找到一组最合适当前自适应算法的阈值参数，阈值参数的搜索算法可以为强化学习，遗传算法等。在这个过程中，深度学习模块将被设置为对当前搜索到的阈值进行评价，完成阈值到评价值的映射。
在本实施例中，搜索算法可以是基于Deep-Q-learning的阈值生成算法，搜索算法为 Q-learnning算法且Q函数由深度学习模块代替。奖励(Reward)函数初始化为0，动作action为对阈值的修改。
在本实施例中，搜索算法也可以是基于深度学习评价模块和阈值生成算法，搜索算法为遗传算法，遗传算法用于搜索最佳阈值，种群个体为一套阈值方案，基因为具体的某个阈值参数，遗传算法的评价函数为深度学习模块。
在本实施例中，平台侧的损失函数是其管理的所有核心控制器的统计参数之和：
具体的，lossl表示来自第l个中央控制设备的损失值，因此，步骤3中生成的梯度需要上送至平台，以更新存储于平台的模型的可训练参数，更新方式如下：

进一步的，Δwl和Δbl为来自第l个核心节点的梯度，该做法将保证平台提供模型初始化时，能够提供一个泛化度极高的模型。
根据本公开实施例的另一方面，还提供了一种无线局域网络动态阈值参数生成装置，图6是根据本公开实施例的无线局域网络动态阈值参数生成装置的框图，如图6所示，所述装置包括：
确定模块62，设置为通过服务器预先训练好的阈值评估模型确定目标阈值参数；
配置模块64，设置为将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点，其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
采集模块66，设置为从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
更新模块68，设置为使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
本公开的实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序，其中，该计算机程序被处理器运行时执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述计算机可读存储介质可以包括但不限于：U盘、只读存储器(Read-Only Memory，简称为ROM)、随机存取存储器(Random Access Memory，简称为RAM)、移动硬盘、磁碟或者光盘等各种可以存储计算机程序的介质。
本公开的实施例还提供了一种电子装置，包括存储器和处理器，该存储器中存储有计算机程序，该处理器被设置为运行计算机程序以执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述电子装置还可以包括传输设备以及输入输出设备，其中，该传输设备和上述处理器连接，该输入输出设备和上述处理器连接。
本实施例中的具体示例可以参考上述实施例及示例性实施方式中所描述的示例，本实施例在此不再赘述。
显然，本领域的技术人员应该明白，上述的本公开的各模块或各步骤可以用通用的计算 装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本公开不限制于任何特定的硬件和软件结合。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (18)

  1. 一种无线局域网络动态阈值参数生成方法，所述方法包括：
    通过服务器预先训练好的阈值评估模型确定目标阈值参数；
    将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点，其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
    从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
    使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
  2. 根据权利要求1所述的方法，其中，在通过服务器预先训练好的阈值评估模型确定目标阈值参数之前，所述方法还包括：
    从服务器获取所述阈值评估模型通过预训练得到的初始训练参数；
    使用所述初始训练参数对所述阈值评估模型进行初始化处理，得到所述预先训练好的阈值评估模型。
  3. 根据权利要求2所述的方法，其中，
    所述初始训练参数为θ(w,b)，其中，w为权重，b为偏移量；
    所述阈值评估模型为
    所述预先训练好的阈值评估模型为fi(x)＝fa(wi⊙x+bi)，其中，⊙为与神经网络对应的网络算子，fa(x)为激活函数。
  4. 根据权利要求3所述的方法，其中，
    所述神经网络包括：深度神经网络DNN、卷积神经网络CNN或循环神经网络RNN；
    所述激活函数包括：线性整流ReLU激活函数、双曲正切Tanh激活函数或S型Sigmoid激活函数等。
  5. 根据权利要求1所述的方法，其中，通过服务器预先训练好的阈值评估模型确定目标阈值参数，包括：
    通过搜索算法从阈值参数空间中选择多个阈值参数，其中，所述阈值参数空间由多个预设的阈值参数组成；
    根据预先存储的多个标签对或所述阈值评估模型分别确定所述多个阈值参数对应的统计参数；
    将所述多个阈值参数中统计参数最大的一个阈值参数确定为目标阈值参数。
  6. 根据权利要求5所述的方法，其中，根据预先存储的多个标签对或所述阈值评估模型分别确定所述多个阈值参数对应的统计参数，包括：
    判断所述多个标签对中是否存在所述阈值参数；
    在判断结果为是的情况下，从所述标签对中确定与所述阈值参数对应的统计参数；
    在判断结果为否的情况下，将所述阈值参数输入所述阈值评估模型，并将所述阈值评估模型的输出结果确定为与所述阈值参数对应的统计参数。
  7. 根据权利要求5所述的方法，其中，所述搜索算法包括强化学习算法或遗传算法。
  8. 根据权利要求1所述的方法，其中，将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点，包括：
    将所述目标阈值参数下发给所述边缘运算节点，其中，所述边缘运算节点根据所述目标阈值参数调整所述预设的自适应算法；
    在信道接入和信息传输的过程中，通过所述边缘运算节点运行所述预设的自适应算法。
  9. 根据权利要求8所述的方法，其中，从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对，包括：
    采集所述边缘运算节点在使用所述目标阈值参数运行所述自适应算法过程中产生的目标统计参数；
    通过所述边缘运算节点对所述目标阈值参数进行归一化处理，得到归一化处理后的目标阈值参数；
    组合所述归一化处理后的目标阈值参数和所述目标统计参数得到所述目标标签对。
  10. 根据权利要求9所述的方法，其中，采集所述边缘运算节点在使用所述目标阈值参数运行所述自适应算法过程中产生的目标统计参数，包括：
    在使用所述目标阈值参数运行所述自适应算法的过程中，从所述边缘运算节点中获取归一化吞吐量和/或发包成功率，其中，所述统计参数包括所述归一化吞吐量和/或所述发包成功率，所述归一化吞吐量为成功发送的负载比特数和总发送的负载比特数的比值，所述发包成功率为MAC协议数据单元的成功发送数量和MAC协议数据单元的总发送数量的比值。
  11. 根据权利要求9所述的方法，其中，通过所述边缘运算节点对所述目标阈值参数进行归一化处理，得到归一化处理后的目标阈值参数，包括：
    通过以下公式对所述目标阈值参数进行归一化处理：
    其中，snorm为所述归一化处理后的目标阈值参数，s＝[s1,s2,...,sn]为多个预设的阈值参数，si为所述目标阈值参数。
  12. 根据权利要求1所述的方法，其中，使用所述目标标签对更新所述阈值评估模型，包括：
    在所述无线接入节点，根据梯度下降法和所述目标标签对对所述阈值评估模型进行训练，得到训练参数，其中，所述训练参数包括权重和偏移量；
    使用所述训练参数更新所述阈值评估模型；
    其中，所述阈值评估模型的损失函数loss为：
    或者，
  13. 根据权利要求12所述的方法，其中，所述方法还包括：
    多个所述无线接入节点分别将根据新采集的目标标签对对所述阈值评估模型进行训练得到的多个训练参数上报给所述服务器；
    通过所述服务器，根据联邦学习算法和所述多个训练参数更新所述服务器的阈值评估模型的初始训练参数。
  14. 根据权利要求13所述的方法，其中，根据联邦学习算法和所述多个训练参数更新所述服务器的阈值评估模型的初始训练参数，包括：
    根据所述多个无线接入节点的阈值评估模型的损失值更新所述服务器的阈值评估模型的总损失值loss_total：
    loss_total＝∑Llossl，其中，lossl为第l个无线接入节点的阈值评估模型的损失值；
    根据所述多个无线接入节点的阈值评估模型的第一权重和第一偏移值更新所述服务器的阈值评估模型的第二权重和第二偏移值，其中，所述初始训练参数包括所述第二权重和所述第二偏移值：
    wi＝wi+∑LΔwl，bi＝bi+∑LΔbl，其中，wi为所述第二权重，bi为所述第二偏移值，Δwl为第l个无线接入节点的阈值评估模型的第一权重，Δbl为第l个无线接入节点的阈值评估模型的第一偏移值。
  15. 根据权利要求1所述的方法，其中，
    在所述预设的自适应算法用于信道选择的情况下，所述目标阈值参数至少包括以下之一：噪声阈值、信道利用率阈值、信道评分用加权因子、错包率阈值、发送功率阈值。
  16. 一种无线局域网络动态阈值参数生成装置，所述装置包括：
    确定模块，设置为通过服务器预先训练好的阈值评估模型确定目标阈值参数；
    配置模块，设置为将所述目标阈值参数配置到与所述无线接入节点连接的边缘运算节点，其中，所述边缘运算节点用于根据所述目标阈值参数运行预设的自适应算法；
    采集模块，设置为从所述边缘运算节点采集与所述目标阈值参数对应的目标统计参数，得到目标标签对；
    更新模块，设置为使用所述目标标签对更新所述阈值评估模型，并通过更新后的阈值评估模型确定新的目标阈值参数。
  17. 一种计算机可读的存储介质，所述存储介质中存储有计算机程序，其中，所述计算机程序被处理器运行时执行所述权利要求1至15任一项中所述的方法。
  18. 一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行所述权利要求1至15任一项中所述的方法。
PCT/CN2023/142126 2023-02-01 2023-12-26 一种无线局域网络动态阈值参数生成方法及装置 Ceased WO2024159986A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP23919547.2A EP4642079A4 (en) 2023-02-01 2023-12-26 METHOD AND APPARATUS FOR GENERATION OF A DYNAMIC THRESHOLD PARAMETER FOR A WIRELESS LOCAL AREA NETWORK

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202310145986.0A CN118433738A (zh) 2023-02-01 2023-02-01 一种无线局域网络动态阈值参数生成方法及装置
CN202310145986.0 2023-02-01

Publications (1)

Publication Number Publication Date
WO2024159986A1 true WO2024159986A1 (zh) 2024-08-08

Family

ID=92033487

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/142126 Ceased WO2024159986A1 (zh) 2023-02-01 2023-12-26 一种无线局域网络动态阈值参数生成方法及装置

Country Status (3)

Country Link
EP (1) EP4642079A4 (zh)
CN (1) CN118433738A (zh)
WO (1) WO2024159986A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119996270A (zh) * 2025-04-02 2025-05-13 中国民航大学 一种网络性能基线调整方法、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012004689A1 (en) * 2010-05-24 2012-01-12 Selex Communications S.P.A. A method and system of bandwidth control
CN112995343A (zh) * 2021-04-22 2021-06-18 华南理工大学 一种具有性能与需求匹配能力的边缘节点计算卸载方法
CN112990018A (zh) * 2021-03-18 2021-06-18 江苏边智科技有限公司 一种动态变化网络环境下深度学习模型的加速执行方法
US20220036123A1 (en) * 2021-10-20 2022-02-03 Intel Corporation Machine learning model scaling system with energy efficient network data transfer for power aware hardware
CN115190489A (zh) * 2022-07-07 2022-10-14 内蒙古大学 基于深度强化学习的认知无线网络动态频谱接入方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113467949B (zh) * 2021-07-07 2022-06-28 河海大学 边缘计算环境下用于分布式dnn训练的梯度压缩方法
CN115017541B (zh) * 2022-06-06 2025-12-02 电子科技大学 一种云边端协同的泛在智能联邦学习隐私保护系统及方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012004689A1 (en) * 2010-05-24 2012-01-12 Selex Communications S.P.A. A method and system of bandwidth control
CN112990018A (zh) * 2021-03-18 2021-06-18 江苏边智科技有限公司 一种动态变化网络环境下深度学习模型的加速执行方法
CN112995343A (zh) * 2021-04-22 2021-06-18 华南理工大学 一种具有性能与需求匹配能力的边缘节点计算卸载方法
US20220036123A1 (en) * 2021-10-20 2022-02-03 Intel Corporation Machine learning model scaling system with energy efficient network data transfer for power aware hardware
CN115190489A (zh) * 2022-07-07 2022-10-14 内蒙古大学 基于深度强化学习的认知无线网络动态频谱接入方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4642079A4

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119996270A (zh) * 2025-04-02 2025-05-13 中国民航大学 一种网络性能基线调整方法、电子设备及存储介质
CN119996270B (zh) * 2025-04-02 2025-08-08 中国民航大学 一种网络性能基线调整方法、电子设备及存储介质

Also Published As

Publication number Publication date
EP4642079A1 (en) 2025-10-29
EP4642079A4 (en) 2026-04-08
CN118433738A (zh) 2024-08-02

Similar Documents

Publication Publication Date Title
Sandoval et al. Optimizing and updating lora communication parameters: A machine learning approach
CN110324170B (zh) 数据分析设备、多模型共决策系统及方法
CN111867139A (zh) 基于q学习的深度神经网络自适应退避策略实现方法及系统
CN110770761A (zh) 深度学习系统和方法以及使用深度学习的无线网络优化
CN109769280B (zh) 一种基于机器学习的wifi智能预测切换方法
CN117667749B (zh) 一种模糊测试用例优化方法及系统
CN118612754B (zh) 可智能组网的三合一终端控制系统及方法
Al-Kaseem et al. A new intelligent approach for optimizing 6lowpan mac layer parameters
CN107786989B (zh) 一种Lora智能水表网络网关部署方法及装置
WO2023230818A1 (zh) 一种波束管理方法及装置、用户设备、网络设备
WO2024159986A1 (zh) 一种无线局域网络动态阈值参数生成方法及装置
CN114492849A (zh) 一种基于联邦学习的模型更新方法及装置
Yadawad et al. Energy-efficient data aggregation and cluster-based routing in wireless sensor networks using tasmanian fully recurrent deep learning network with pelican variable marine predators algorithm
Chi et al. Enhancing adaptability and efficiency of task offloading by broad learning in industrial iot
CN113573342B (zh) 一种基于工业物联网的节能计算卸载方法
Cengiz et al. SOHCL-RDT: A self-organized hybrid cross-layer design for reliable data transmission in wireless network
CN108770010B (zh) 一种面向服务的无线网络组网模式智能重构方法
US20250005369A1 (en) Value-based action selection algorithm in reinforcement learning
CN112804074B (zh) 网络参数配置方法及装置
CN116737260A (zh) 基于人工蜂-鱼群算法的计算卸载方法、装置及系统
WO2024031535A1 (zh) 无线通信方法、终端设备和网络设备
CN116562363A (zh) 一种模型训练方法、装置及通信设备
CN114020473A (zh) 数据传输与处理方法、装置及系统
CN114125945A (zh) 5g网络切换模板的选取方法及装置
Balbi et al. Ultra-Low Power DNN-based TSCH Scheduling at the Edge using the MAX78000

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23919547

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2023919547

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2023919547

Country of ref document: EP

Effective date: 20250722

REG Reference to national code

Ref country code: BR

Ref legal event code: B01A

Ref document number: 112025015542

Country of ref document: BR

ENP Entry into the national phase

Ref document number: 2023919547

Country of ref document: EP

Effective date: 20250722

NENP Non-entry into the national phase

Ref country code: DE

WWP Wipo information: published in national office

Ref document number: 2023919547

Country of ref document: EP