Method and system for grid management of energy consumption of industrial control link of multi-agent system

Abstract

The invention discloses a method and a system for grid management of energy consumption of an industrial control link of a multi-agent system. The method comprises the following steps: extracting multi-agent system data, and based on agent feature classification, generating a training matrix by adopting a Pareto optimal solution algorithm; calculating a grid segmentation index of the industrial control link to carry out adaptive grid segmentation; a Minimax-Q algorithm is adopted to obtain an optimal matrix training result set; and acquiring memory and CPU resource data of each sub-link through a PLC program, comparing the memory and CPU resource data with the optimal set, if a condition is met, mapping the data as a Socket service request and sending a control signal to execute optimization, otherwise, continuously monitoring and recording. The system comprises a data aggregation and training matrix generation module, an intelligent industrial control grid segmentation module, a reinforcement learning optimization module and an energy consumption monitoring execution module. The method can achieve the optimal energy consumption of the system, optimizes the resource distribution through the self-adaptive grid segmentation, and improves the quality of decision data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent industrial control, and particularly relates to a method and system for grid management of energy consumption of an industrial control link by a multi-agent system. BACKGROUND

[0002] The Foundation for Intelligent Physical Agent (FIPA) of the IEEE Computer Society standardization organization of agent technology defines an agent as "an agent is an entity residing in an environment, which can interpret data obtained from the environment reflecting events occurring in the environment, and perform actions that have an impact on the environment." According to this, an agent can be divided into the following functional modules: an input module for obtaining information from the environment; a target module for judging the task target to provide a standard for decision-making; a learning module for processing and interpreting environmental information, making decisions, and recording results to promote future decision-making efficiency; an output module for performing operations according to the decision-making results to change the environmental information; and a communication module for interacting with other agents in the system to improve decision-making quality. SUMMARY

[0003] The purpose of the present application is to provide a method and system for grid management of energy consumption of an industrial control link by a multi-agent system, to solve the problem of inevitable game in the interaction mechanism due to weak correlation and strong autonomy of each agent in the multi-agent system.

[0004] To achieve the above purpose, the technical solution provided by the present application is a method for grid management of energy consumption of an industrial control link by a multi-agent system, comprising the following steps: S1, extracting data of the multi-agent system, classifying data features according to structural features and behavioral characteristics of the agent, and aggregating the classified data by using a Pareto optimal solution algorithm; S2, calculating grid partition indexes of each industrial control link according to the classified data of the multi-agent system, and performing adaptive grid partition on data of different industrial control links to obtain a grid-labeled training matrix; S3, solving the grid-labeled training matrix by using a Minimax-Q algorithm to obtain an optimal matrix training result set; S4, collecting memory and CPU resource data of the grid-labeled training matrix corresponding to each sub-industrial control link by using a PLC control program, comparing the data with the optimal matrix training result set, and if a preset control starting condition is met, mapping the CPU resource data to a Socket service request, sending a control signal through an Ethernet unit, and performing an optimization action specified by the optimal matrix training result set; otherwise, maintaining monitoring and recording data.

[0005] To optimize the above technical solutions, the specific measures taken also include: In step S1, the classified data is aggregated using the Pareto optimal solution algorithm, and the specific steps are as follows: Define decision vectors x and y and objective function f. If f(x) is not greater than any target and is less than f(y) at least on one target, then x dominates y; Screen all vectors in the decision space that are not dominated by any other decision vector to form a Pareto optimal solution set; Using linear weighting method, set the weight according to the importance of each industrial control link, and determine the final data aggregation strategy from the Pareto optimal solution set; The row vector of the training matrix is the state of the agent and the action of the agent corresponding to the Pareto optimal solution, and the column vector is the identification of the industrial control link and the target parameter; The agent state includes real-time data volume and device load, the agent action includes aggregation frequency and data filtering rule, and the target parameter includes coverage and redundancy.

[0006] Further, in step S1, the data of the multi-agent system is extracted, and the feature classification is performed according to the structural characteristics and behavioral characteristics of the agent, specifically including: According to the physical distribution and logical architecture of the agent in the system, the structural characteristics of the agent are divided into centralized, distributed or hybrid; and according to the decision mode and interaction ability of the agent, the behavioral characteristics of the agent are divided into autonomous or active adaptive type.

[0007] In step S2, the grid partition index of each industrial control link is calculated according to the classified data of the multi-agent system, including: The comprehensive industrial control network usage density is calculated using the industrial control link number density, the industrial control command density and the connection number density, and the calculation formula is:

[0008] Wherein, is the comprehensive industrial control network usage density, is the industrial control link number density, is the industrial control command density, is the connection number density, , and are the weights for the number of industrial control links, industrial control commands and connection numbers respectively.

[0009] Further, in step S2, the grid size of each industrial control link is adjusted according to the comprehensive industrial control network usage density, and the calculation formula is:

[0010] Wherein, This is the adjusted grid size. This is the initial grid size. It refers to the network usage density of each sub-industrial control link. It represents the average network usage density across the entire industrial control system.

[0011] Step S3 involves receiving the grid-marked training matrix and solving it using the Minimax-Q algorithm, specifically including: Define the state-value function of agent i in state s. for:

[0012] Where -i represents the opponent of agent i, and represents the opponent's action. For the combined action state value function; The Q-learning algorithm is used to approximate the true Q value, and linear programming or quadratic programming is used to solve the Nash equilibrium strategy.

[0013] Step S4 includes: The PLC control program receives memory and CPU resource data from each sub-industrial control link and determines whether the control start conditions are met. Furthermore, if the conditions are met, the memory and CPU resource data of each sub-operating condition are mapped to Socket service requests, and the preset data storage area is called to configure Socket service parameters; the PLC control program sends control signals through the Ethernet unit according to the configured parameters, and synchronously receives the status feedback of the sub-processes to realize energy consumption monitoring and management.

[0014] As another important technical solution, this invention also provides a system for grid-based management of energy consumption in industrial control processes using a multi-agent system, comprising: The data aggregation and training matrix generation module is used to extract data from the multi-agent system, classify the data features based on the structural and behavioral characteristics of the agents, and aggregate the classified data using the Pareto optimal solution algorithm. The intelligent industrial control grid segmentation module is used to calculate the grid segmentation index of each industrial control link based on the classification data of the multi-agent system, and then perform adaptive grid segmentation on the data of different industrial control links to obtain the training matrix of grid marking. The reinforcement learning optimization module is used to solve the training matrix of the grid marking using the Minimax-Q algorithm to obtain the optimal matrix training result set; The energy consumption monitoring and execution module is used to collect the memory and CPU resource data of the training matrix of the grid marker corresponding to each sub-industrial control link in real time using the PLC control program, and compare it with the optimal matrix training result set: if the preset control start conditions are met, the CPU resource data is mapped to a Socket service request, and a control signal is sent through the Ethernet unit to execute the optimization action specified by the optimal matrix training result set; otherwise, monitoring is maintained and the data is recorded.

[0015] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for grid-based management of energy consumption in industrial control links of a multi-agent system as described above.

[0016] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute a method for managing energy consumption in industrial control components using a multi-agent system grid, as described above.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention solves the Nash equilibrium of multi-agent games using the Minimax-Q algorithm, obtains the optimal matrix training result set, effectively coordinates the game problem caused by the strong autonomy and conflicting goals of each agent in the system, achieves optimal energy consumption at the overall system level, and avoids the overall performance degradation caused by local optimization.

[0018] This invention employs adaptive grid segmentation by calculating multi-dimensional indicators such as the quantity density and command density of industrial control components, ensuring that the size of the management unit matches the actual data density and importance of the industrial control components. It enables refined management of high-density areas and macro-level control of low-density areas, thereby optimizing resource allocation and computational load while maintaining management accuracy.

[0019] This invention uses the Pareto optimal solution algorithm to aggregate agent data, ensuring that each state-action pair in the generated training matrix is ​​a non-dominated solution under multiple objectives such as coverage and redundancy. This provides a high-quality data foundation for subsequent reinforcement learning training and improves the reliability and effectiveness of the final decision strategy.

[0020] This invention seamlessly integrates the optimal strategy set with the underlying hardware execution through a PLC control program, and uses Socket communication to send control signals and receive status feedback in real time. The system can not only proactively intervene based on the optimal strategy set, but also maintain silent monitoring when the status is normal, thus balancing energy efficiency optimization and system stability. Attached Figure Description

[0021] Figure 1The flowchart of the present invention, which uses a PLC control program to manage the industrial control grid and each link, is as follows. Detailed Implementation The present invention will be further described in detail below through specific embodiments, but it should not be construed as limiting the scope of the subject matter of the present invention to the following embodiments. All technologies implemented based on the above content of the present invention fall within the scope of the present invention.

[0022] The following is a brief explanation of some of the terms related to this invention: Pareto optimality algorithm: refers to an ideal state of resource allocation, in which it is impossible to improve at least one objective metric without worsening any other objective metric by changing the resource allocation. In this invention, the Pareto optimality algorithm is used to screen non-dominated solutions from multi-objective decisions to optimize data aggregation strategies.

[0023] This invention provides a method for grid-based management of energy consumption in industrial control processes within a multi-agent system, comprising the following steps: S1. Extract data from the multi-agent system, classify features based on the agents' structural and behavioral characteristics, and aggregate the classified data using the Pareto optimal solution algorithm to generate a training matrix. This algorithm filters non-dominated solutions by judging the dominance relationship of decision vectors, ensuring the effectiveness of data aggregation. Preferably, step 1 involves extracting data from a multi-agent system and classifying it based on the agents' autonomous and adaptive characteristics, specifically including: Based on the physical distribution and logical architecture of the agents in the system, their structural characteristics are classified as centralized, distributed, or hybrid; and based on the decision-making patterns and interaction capabilities of the agents, their behavioral characteristics are classified as autonomous or proactive.

[0024] In some implementations, the structure is classified as distributed, with each agent having independent input, decision-making and output. They are loosely combined and can be flexibly added or removed, which fits the characteristics of distributed parallelism, flexibility and robustness. The classification and aggregation points are based on the main characteristics of multi-agent systems: Communication and collaboration: The agents in the system communicate information and collaborate through various direct or indirect means, enabling the entire system to have problem-solving capabilities that surpass those of a single agent.

[0025] Distributed parallelism: Each agent can independently input, make a decision, and output, giving the entire system a distributed characteristic. The agents can operate independently and in parallel, allowing the entire system to solve complex problems in parallel or in a distributed manner.

[0026] Flexible and robust: In a multi-agent system, the agents are relatively loosely combined, making it easy to add or remove an agent without causing the entire system to crash due to the failure of a single component.

[0027] The Pareto optimal solution algorithm specifically includes: Define decision vectors x and y and objective function f:

[0028] If f(x) is not greater than f(y) in any objective and at least one objective is less than f(y), then x dominates y. Select x that are not dominated by any vector to form a Pareto optimal solution set, and use a linear weighting method (the weights are set according to the importance of the industrial control link) to determine the final aggregation strategy from the solution set.

[0029] In some implementations, an m×n training matrix is ​​constructed using the agent state (real-time data volume, device load) and actions (aggregation frequency, data filtering rules) corresponding to the optimal solution as row vectors, and the industrial control link identifier and target parameters (coverage ≥95%, redundancy ≤10%) as column vectors (m is the number of solutions, n is the state + action + target dimension).

[0030] In some implementations, an industrial control system has 10 intelligent agents. The algorithm selects 3 non-dominated solutions, chooses the solution with the highest weighted score, and generates a 10-row, 8-column (2 states + 3 actions + 3 targets) training matrix.

[0031] S2. Based on the classification data of the multi-agent system, calculate the grid segmentation index of each industrial control link, and perform adaptive grid segmentation on different industrial control links accordingly to obtain the training matrix of grid marking. In some implementations, multi-agent system classification data is used to analyze the data usage density of each industrial control link; specifically, the overall industrial control network usage density is calculated using the density of the number of industrial control links, the density of industrial control commands, and the density of connections, using the following formula:

[0032] in, It is the density of integrated industrial control network usage. It refers to the quantity density of industrial control components. It is the density of industrial control commands. It is the connection number density. , and These are the weights used for the number of industrial control links, industrial control commands, and the number of connections, respectively.

[0033] Step S2 adjusts the grid size of each industrial control link based on the usage density of the integrated industrial control network: high-density areas are divided into larger grids, and low-density areas are divided into smaller grids. This increases the proportion of grids in industrial control links with weaker signals and poorer perception, ensuring that the performance data of the industrial control network is more accurate and meaningful for analysis. The specific calculation formula is as follows:

[0034] in, This is the adjusted grid size. This is the initial grid size. It refers to the network usage density of each sub-industrial control link. It represents the average network usage density across the entire industrial control system.

[0035] In some implementations, the raw data of the multi-agent system is extracted and classified according to the structural features and behavioral characteristics of the agents. Then, the Pareto optimal solution algorithm is used to aggregate the classified data. By defining decision vectors and objective functions, filtering non-dominated solution sets, and determining the final aggregation strategy based on the importance of industrial control links, an initial training matrix is ​​generated, with the agent states and actions corresponding to Pareto optimal solutions as rows and the industrial control link identifiers and target parameters as columns. Next, based on the classification data of the multi-agent system, the grid segmentation index of each industrial control link is calculated, including the density of industrial control links, the density of industrial control commands, and the density of connections. The industrial control network usage density is then comprehensively derived, and adaptive grid segmentation is performed on different industrial control links accordingly, dynamically adjusting the grid size of each link. Finally, the segmented grid coordinates or identifiers are assigned to each data unit in the initial training matrix, thereby forming a training matrix with grid labels containing spatial structure information.

[0036] S3. Receive the training matrix marked with grids, solve it using the Minimax-Q algorithm, and obtain the optimal matrix training result set; The algorithm mines data and uses training metrics to construct an accurate mapping model. This mapping model is described by the following equation:

[0037] Where Y is the set of control objectives (dependent variables), such as the expected yield of a certain end product; X is the set of independent variables (independent variables) that are controllable in real time, such as the current feeding rate; Z is the set of independent variables (independent variables) that are not controllable in real time, such as the properties of the current input deliverable; t±{△} represents the time lag relationship between the variables.

[0038] Preferably, the Minimax-Q algorithm is used for solving the problem, specifically including: In a zero-sum random game between two players, given a state, the state-value function V(s) of the i-th agent is defined as:

[0039] In the formula, -i represents the opponent of agent i. Let be the joint action state-value function. This formula means that each agent i maximizes the expected reward value in the worst-case scenario against the opponent. In multi-agent reinforcement learning, 2 is unknown, so Q-learning is used to approximate the actual reward. The value is then used to solve for the Nash equilibrium strategy at state s using linear programming.

[0040] NashQ-Learning extends Minimax-Q from zero-sum games to multi-player general sum games. The algorithm requires observation of the actions of all other agents. and We use quadratic programming to solve for the Nash equilibrium point.

[0041] In some implementations, the training matrix based on grid marking is iteratively solved using the Minimax-Q reinforcement learning algorithm. This process simulates the game interaction of multiple agents in an industrial control grid environment. The algorithm continuously approximates the action value function of each agent in a specific state through the Q-learning mechanism, and solves the Nash equilibrium policy with the help of linear programming or quadratic programming. After a large number of iterations until the policy converges, a stable policy set is finally output. This set is the optimal matrix training result set, which defines the optimal actions and target parameters that each agent should take under different industrial control grid states, providing a decision-making basis for subsequent energy consumption management.

[0042] S4. The PLC control program collects the memory and CPU resource data of the training matrix corresponding to the grid markers of each sub-industrial control link in real time, and compares it with the optimal matrix training result set: if the preset control start conditions are met, the CPU resource data is mapped to a Socket service request, and a control signal is sent through the Ethernet unit to execute the optimization action specified by the optimal matrix training result set; otherwise, monitoring is maintained and the data is recorded.

[0043] In some implementations, the PLC control program receives memory and CPU resource data from each sub-control unit and compares it with the preset reasonable threshold for equipment load and energy consumption optimization target parameters of the optimal matrix training result set. If the data deviates from the reasonable range or the target parameters are not met, the control start-up conditions are determined to be met; otherwise, the start-up conditions are determined not to be met.

[0044] If the control start conditions are met, the Socket service parameters are configured based on the optimal action and target parameters in the optimal matrix training result set. Then, control signals are accurately sent through the Ethernet unit (such as adjusting the resource allocation of high-load sub-links). The status feedback is verified based on the target parameters set in the result set (such as confirming whether the coverage and redundancy of sub-link energy consumption data meet the optimal standards). Finally, the global optimal monitoring and management of the energy consumption of the industrial control system is achieved.

[0045] Preferably, when the control start-up conditions are not met, the PLC control program continues to collect data and monitor the status of each sub-control link in real time: It continuously receives dynamic data on memory and CPU resources from sub-control links, and periodically repeats the comparison process with the preset conditions of the optimal matrix training result set to ensure that once the operating status of the sub-link changes (such as resource load exceeding a reasonable range or energy consumption target parameters not meeting the standard), the control start conditions can be triggered immediately. The system synchronously records the operating data of sub-processes that do not meet the current conditions (such as memory occupancy rate that meets the threshold, redundancy that meets the standard, etc.) to the system storage module. This provides actual operating data support for the subsequent optimization matrix training result set (such as adjusting the load threshold and optimizing the target parameters). At the same time, it maintains the current stable operating state of the industrial control system, avoids unnecessary control operations from interfering with the normal production process, and ultimately realizes the dynamic energy consumption monitoring logic of not intervening unless necessary and responding immediately when intervention is required.

[0046] In some implementations, Socket service parameters include, for example, setting the signal transmission interval according to the optimal aggregation frequency and determining the range of resource data to be transmitted according to data filtering rules. The preset data storage area provides standardized basic resources for parameter configuration, while the optimal matrix training result set provides an optimized basis for energy consumption management for parameter adaptation and adjustment. The two are combined through the calling logic of the PLC control program to ensure that the configured Socket service parameters can not only meet the stability requirements of industrial communication, but also adapt to the energy consumption monitoring and management needs of the current industrial control process.

[0047] In some implementations, such as Figure 1 As shown, attached Figure 1 This paper clearly describes the complete workflow of data exchange between a PLC and a host computer server via Socket network communication in an industrial control energy management system. The entire process begins with the establishment of the connection and ends with the closure of the connection, forming a complete control loop.

[0048] First, there is the connection establishment phase. The left side of the flowchart represents the server (PC), and the right side represents the PLC. At the start of the process, the server, as the passive party in the communication, enters a listening state, waiting for connection requests from clients on a specific network port. Simultaneously, the PLC, as the active party in the communication, performs an active open operation, initiating a Socket connection request to the server's specified address and port. Once the server accepts the request, the connection is successfully established, preparing a channel for subsequent data transmission.

[0049] After the connection is established, the PLC first sends a command to the server. Then, the process enters a crucial data request and receive loop. The PLC first sets the value of its internal data memory address D01000 to 3030. This indicates to the server that the PLC currently wishes to request or is ready to receive the first set of specific data. After setting this status flag, the PLC executes the data receive operation, retrieving the first set of data from the server.

[0050] Preferably, after successfully receiving the first set of data, the PLC updates the value of D01000 to 3131. This indicates that the PLC has issued the next command, instructing the server to prepare to send the second set of data, or to enter another processing stage. Subsequently, the PLC executes the data receiving operation again to obtain the second set of data. This sub-process, consisting of setting a specific D01000 value and receiving data, demonstrates how the system can orderly schedule different data exchange tasks by changing the value of a status register.

[0051] After all necessary data and command exchanges are completed, the process enters the final stage. The PLC initiates a socket close operation, actively disconnecting the network connection. Upon detecting the connection disconnection, the server also performs a socket close operation. Thus, a complete communication session ends. This closed-loop process ensures efficient use of communication resources and prepares for the next connection request.

[0052] In another embodiment of the present invention, a system for grid-based management of energy consumption in industrial control processes using a multi-agent system is proposed, comprising: The data aggregation and training matrix generation module is used to extract data from the multi-agent system, classify the data features based on the structural and behavioral characteristics of the agents, and aggregate the classified data using the Pareto optimal solution algorithm. The intelligent industrial control grid segmentation module is used to calculate the grid segmentation index of each industrial control link based on the classification data of the multi-agent system, and then perform adaptive grid segmentation on the data of different industrial control links to obtain the training matrix of grid marking. The reinforcement learning optimization module is used to solve the training matrix of the grid marking using the Minimax-Q algorithm to obtain the optimal matrix training result set; The energy consumption monitoring and execution module is used to collect the memory and CPU resource data of the training matrix of the grid marker corresponding to each sub-industrial control link in real time using the PLC control program, and compare it with the optimal matrix training result set: if the preset control start conditions are met, the CPU resource data is mapped to a Socket service request, and a control signal is sent through the Ethernet unit to execute the optimization action specified by the optimal matrix training result set; otherwise, monitoring is maintained and the data is recorded.

[0053] In another embodiment of the present invention, an electronic device is proposed, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for grid-based management of energy consumption in industrial control links of a multi-agent system as described above.

[0054] In another embodiment of the present invention, a computer-readable storage medium is provided, storing a computer program that enables a computer to execute a method for grid-based management of energy consumption in industrial control components of a multi-agent system, as described above.

[0055] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent substitutions, and improvements made by those skilled in the art to the above embodiments without departing from the scope of the technical solution of the present invention, based on the technical essence of the present invention, shall still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for grid management of energy consumption of an industrial control link by a multi-agent system, characterized in that, The method comprises the following steps: S1, extracting data of the multi-agent system, classifying data features according to structural features and behavior characteristics of the agent, and aggregating the classified data by using a Pareto optimal solution algorithm; S2, calculating grid segmentation indexes of each industrial control link according to the classified data of the multi-agent system, and performing adaptive grid segmentation on data of different industrial control links to obtain a grid-labeled training matrix; S3, solving the grid-labeled training matrix by using a Minimax-Q algorithm to obtain an optimal matrix training result set; S4, collecting memory and CPU resource data of the grid-labeled training matrix corresponding to each sub-industrial control link in real time by using a PLC control program, comparing the data with the optimal matrix training result set, and if a preset control starting condition is met, mapping the CPU resource data into a Socket service request, sending a control signal through an Ethernet unit, and performing an optimization action specified by the optimal matrix training result set; otherwise, maintaining monitoring and recording data. 2.The method for managing energy consumption of industrial control links in a grid form according to claim 1, characterized in that: In step S1, the classified data is aggregated by using the Pareto optimal solution algorithm, and the specific steps are as follows: Define decision vectors x and y and a target function f, and if f(x) is not greater than f(y) on any target and is less than f(y) on at least one target, then x dominates y; Screen all vectors in the decision space that are not dominated by any other decision vector to form a Pareto optimal solution set; Determine the final data aggregation strategy from the Pareto optimal solution set by using a linear weighting method according to the importance of each industrial control link; The row vector of the training matrix is the agent state and agent action corresponding to the Pareto optimal solution, and the column vector is the industrial control link identifier and target parameter; The agent state includes real-time data volume and device load, the agent action includes aggregation frequency and data filtering rule, and the target parameter includes coverage and redundancy.

3. The method for grid management of energy consumption of industrial control links by multi-agent system according to claim 1, characterized in that: In step S1, the data of the multi-agent system is extracted, and the feature classification is performed according to the structural features and behavior characteristics of the agent, specifically including: According to the physical distribution and logical architecture of the agent in the system, the structural features of the agent are divided into centralized, distributed or hybrid; and according to the decision mode and interaction ability of the agent, the behavior characteristics of the agent are divided into autonomous or active adaptive.

4. The method for grid management of energy consumption of industrial control links by multi-agent system according to claim 1, characterized in that: In step S2, the grid segmentation indexes of each industrial control link include: The comprehensive industrial control network usage density is calculated by using the industrial control link number density, industrial control command density and connection number density, and the calculation formula is: wherein, is the integrated industrial network usage density, is the industrial loop quantity density, is the industrial command density, is the connection number density, , and are weights for industrial loop quantity, industrial command, and connection number, respectively.

5. The method for energy consumption of industrial control links in grid management of multi-agent system according to claim 4, characterized in that: In step S2, the grid size of each industrial control link is adjusted according to the comprehensive industrial control network usage density, and the calculation formula is: wherein, is the adjusted grid size, is the initial grid size, is the network usage density of each sub-industrial control segment, is the average network usage density of the entire industrial control segment.

6. The method for grid management of energy consumption of industrial control links by multi-agent system according to claim 1, characterized in that: In step S3, the grid-labeled training matrix is solved by using the Minimax-Q algorithm, specifically including: Define the state value function of agent i at state s as V (s) = maxa∈A (R (s, a) + γV (s where -i denotes the opponent of agent i, and a^i denotes the action of the opponent, is the joint action state value function; The real Q value is approximated by using the Q-learning algorithm, and the Nash equilibrium strategy is solved by using linear programming or quadratic programming.

7. The method for energy consumption of industrial control links in grid management of multi-agent system according to claim 1, characterized in that: The step S4 specifically includes the following processes: The PLC control program receives the memory and CPU resource data of each sub-industrial control link, and if the optimal state range in the optimal matrix training result set is deviated, it is judged that the control starting condition is met; If the condition is met, the memory and CPU resource data of each sub-process link is mapped to a Socket service request, and the preset data storage area is configured with Socket service parameters; the PLC control program sends control signals through the Ethernet unit according to the configured parameters, and synchronously receives the state feedback of the sub-link, thereby realizing energy consumption monitoring and management.

8. A multi-agent system grid management system for energy consumption of an industrial control link, characterized in that, Comprise: The data aggregation and training matrix generation module is used for extracting the data of the multi-agent system, classifying the data features according to the structural features and behavior characteristics of the agents, and aggregating the classified data by using a Pareto optimal solution algorithm; The intelligent industrial control grid segmentation module is used for calculating the grid segmentation indexes of each industrial control link according to the classified data of the multi-agent system, and adaptively segmenting the data of different industrial control links to obtain a grid-labeled training matrix; The reinforcement learning optimization module is used for solving the grid-labeled training matrix by using a Minimax-Q algorithm to obtain an optimal matrix training result set; The energy consumption monitoring and execution module is used for collecting the memory and CPU resource data of the grid-labeled training matrix corresponding to each sub-industrial control link in real time by using a PLC control program, comparing the data with the optimal matrix training result set, mapping the CPU resource data to a Socket service request if a preset control starting condition is met, sending a control signal through an Ethernet unit, and executing the optimization action specified by the optimal matrix training result set; otherwise, the monitoring is maintained and the data is recorded.

9. An electronic device, comprising: Comprise: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program enables the computer to execute the method of any one of claims 1-7.

Images