Multi-agent system cooperative control method and device, electronic equipment and storage medium

Abstract

The application discloses a kind of multi-agent system cooperative control method, device, electronic equipment and storage medium, it is related to industrial artificial intelligence technical field.The method collects the multi-source sensor data of multiple to be controlled equipment;Based on the multi-source sensor data, drift detection is carried out to build evidence body, and drift detection result is obtained;Based on drift detection result, adaptive reference vector enhancement is carried out to irregular front, and adaptive adjusted reference vector set is obtained;Multi-agent system is modeled as non-cooperative aggregate game, based on adaptive adjusted reference vector set, non-dominated solution set is calculated by differential privacy protection distributed conditional gradient algorithm;According to drift detection result, hierarchical response is executed, and according to the current field demand of multiple to be controlled equipment, control instruction is selected from non-dominated solution set and issued to control equipment.The technical scheme of the application improves dynamic adaptability, breaks through high-dimensional optimization bottleneck, realizes lightweight calculation under privacy protection.

Description

Technical Field

[0001] This invention relates to the field of industrial artificial intelligence technology, and in particular to a method, apparatus, electronic device and storage medium for collaborative control of a multi-agent system. Background Technology

[0002] When complex industrial clusters (such as air compressor stations, injection molding machine groups, etc.) make distributed decisions in a dynamic environment, current technical solutions involve each agent collecting sensor data, using weighted averaging or filtering to remove noise, assuming stable data distribution; constructing a multi-objective optimization model, using a fixed reference vector to decompose the multi-objective problem into single-objective sub-problems; agents calculating gradients and exchanging them over a network, using the projected gradient descent method for iterative updates, projecting variables back into the feasible region; and finally selecting the final solution from the Pareto solution set and issuing it for execution.

[0003] However, existing technologies implicitly assume that training and real-time data are distributed in the same way, making it impossible to dynamically perceive drift; in real industrial problems, there is nonlinear coupling between objectives, the shape of the leading edge is irregular, fixed vectors cannot perceive the shape of the leading edge, a large number of points are pointing to invalid regions, and there is no solution or insufficient accuracy in the key trade-off region (knee point); gradient encoding local data statistical characteristics is vulnerable to gradient inversion attacks; projected gradient descent requires a high-dimensional quadratic programming solution for each step, which cannot meet the real-time requirements of edge devices. Summary of the Invention

[0004] This invention provides a method, device, electronic device, and storage medium for collaborative control of a multi-agent system, in order to improve dynamic adaptability, overcome the bottleneck of high-dimensional optimization, and achieve lightweight computing under privacy protection.

[0005] According to one aspect of the present invention, a cooperative control method for a multi-agent system is provided, the method comprising:

[0006] Collect multi-source sensor data from the multiple devices to be controlled;

[0007] Based on the multi-source sensor data, an evidence body is constructed to perform drift detection, and a drift detection result is obtained.

[0008] Based on the drift detection results, adaptive reference vector enhancement is performed for irregular frontiers to obtain an adaptively adjusted reference vector set.

[0009] The multi-agent system is modeled as a non-cooperative aggregation game, and the non-dominated solution set is calculated based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm.

[0010] A graded response is executed based on the drift detection results, and control commands are selected from the non-dominated solution set and issued to control the devices based on the current field requirements of the multiple devices to be controlled.

[0011] According to another aspect of the present invention, a cooperative control device for a multi-agent system is provided. The multi-agent system includes multiple devices to be controlled, each modeled as an agent, and the multiple devices to be controlled are interconnected. The device includes:

[0012] The data acquisition module is used to acquire multi-source sensor data from the multiple devices to be controlled;

[0013] The drift detection module is used to construct an evidence body based on the multi-source sensor data to perform drift detection and obtain drift detection results;

[0014] The adjusted reference vector acquisition module is used to perform adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtain an adaptive adjusted reference vector set.

[0015] The non-dominated solution set calculation module is used to model the multi-agent system as a non-cooperative aggregation game and calculate the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm.

[0016] The equipment control module is used to execute a graded response based on the drift detection results, and to select control commands from the non-dominated solution set according to the current field requirements of the multiple devices to be controlled in order to control the devices.

[0017] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the multi-agent system cooperative control method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the multi-agent system cooperative control method according to any embodiment of the present invention.

[0022] The technical solution of this invention collects multi-source sensor data from multiple devices to be controlled; constructs an evidence body based on the multi-source sensor data for drift detection, and obtains drift detection results; performs adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtains an adaptively adjusted reference vector set; models the multi-agent system as a non-cooperative aggregation game, and calculates the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm; executes a hierarchical response according to the drift detection results, and selects control commands from the non-dominated solution set to control the devices according to the current field requirements of multiple devices to be controlled. By employing sensing, optimization, game theory, and execution coordination techniques, this invention solves the problems of existing technologies, such as the inability to detect operating condition drift in a timely manner, the difficulty of adapting fixed reference vectors to irregular Pareto fronts, privacy risks in distributed communication, and high computational complexity of projection. This improves dynamic adaptability, overcomes high-dimensional optimization bottlenecks, and achieves lightweight computation under privacy protection.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a multi-agent system cooperative control method provided in Embodiment 1 of the present invention;

[0026] Figure 2 This is a schematic diagram of the structure of a multi-agent system cooperative control device provided in Embodiment 2 of the present invention;

[0027] Figure 3 This is a schematic diagram of the structure of an electronic device that implements the multi-agent system cooperative control method of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0030] Example 1

[0031] Figure 1 This is a flowchart of a multi-agent system cooperative control method provided in Embodiment 1 of the present invention. This embodiment is applicable to the control of equipment in industrial sites under dynamic environments. The method can be executed by a multi-agent system cooperative control device, which can be implemented in hardware and / or software and can be configured in a server. Figure 1 As shown, the method includes:

[0032] S110: Collect multi-source sensor data from multiple devices to be controlled.

[0033] Industrial sites can refer to locations such as air compressor stations or injection molding machine clusters. Controlled equipment can refer to equipment such as air compressors or injection molding machines. Multi-source sensor data includes multimodal sensor data such as vibration, temperature, pressure, current, and images from the industrial site.

[0034] S120. Construct evidence based on multi-source sensor data to perform drift detection and obtain drift detection results.

[0035] In this embodiment, multi-source sensor data can serve as factual evidence for assessing actual working conditions and operational status in industrial settings. Therefore, this embodiment can transform multi-source sensor data into evidence, obtain global reliability through evidence fusion, extract conflict coefficients as statistics, and utilize cumulative sum control charts to detect concept drift in real time.

[0036] In one alternative implementation, constructing an evidence body based on multi-source sensor data to perform drift detection and obtain drift detection results may include:

[0037] Define a state identification framework for multi-source sensor data; for the k-th sensor data and the state identification framework, calculate the normal state confidence and abnormal state confidence of the k-th sensor data, and determine the uncertain state confidence of the k-th sensor data based on the normal state confidence and abnormal state confidence; based on the normal state confidence, abnormal state confidence and uncertain state confidence of the k-th sensor data, construct an evidence body for the k-th sensor data.

[0038] In this embodiment, the complete set of state judgments for the industrial site can be defined first, that is, the state identification framework can be defined. ,in This indicates "operating condition is normal". This indicates an "abnormal operating condition," thus defining a clear judgment boundary for subsequent probability assignments. Data from k sensors is collected over a period t. Real-time readings of the k-th sensor The basic probability assignment of the sensor to "current operating condition is normal" can be calculated using the Gaussian membership function: , , Let represent the mean and standard deviation of the k-th sensor under steady-state normal operating conditions, respectively. The closer the sensor reading is to the mean under normal operating conditions, the better. The closer the value is to 1, the higher the sensor's confidence in the "normal operating condition". Referring to the calculation logic for the normal state, the basic probability assignment for the "abnormal operating condition" of the k-th sensor is calculated similarly: The remaining confidence level can also be allocated to the entire set of identification frames using a normalization formula. : The value represents the degree of "uncertainty" in which the k-th sensor cannot definitively determine whether the operating condition is normal or abnormal due to factors such as noise interference and missing data. The confidence levels for normal, abnormal, and uncertain states of the k-th sensor data collectively constitute the evidence body for the k-th sensor data.

[0039] Taking the air compressor exhaust pressure sensor as an example, It is a cycle Pressure reading; The larger the value, the more the sensor considers the current pressure to be within the normal range; The larger the value, the more the sensor considers the current pressure to be in an abnormal range; The larger the value, the more "uncertain" the sensor is due to factors such as noise or missing data.

[0040] Furthermore, the difference distance between each pair of evidence is calculated, and the credibility of the k-th evidence is calculated based on the difference distance between each pair of evidence. The credibility of all evidence is normalized to obtain the fusion weight of each evidence. Based on the fusion weight, all evidence is fused using the weighted DS rule to obtain the global credibility.

[0041] In this embodiment, weighted optimization can be performed on the independent evidence generated by multiple sensors. This embodiment can first calculate the difference distance between pairwise evidence (such as the evidence from the k-th and j-th sensor data). , This can represent the Jaccard similarity matrix, with elements... , This distance It can precisely quantify the degree of conflict or difference between two sensor pieces of evidence. Furthermore, it can be based on... and The difference distance between them is calculated by averaging the distance between the k-th piece of evidence and all the remaining K-1 pieces of evidence, thereby... Calculate the credibility of the k-th piece of evidence. The higher the value, the more consistent the evidence from the k-th sensor is with the judgments from other sensors, and the higher the reliability of that evidence. This allows for the normalization of the credibility of all evidence, yielding the final fusion weight for each piece of evidence. Sensor evidence with higher reliability is assigned a higher weight in subsequent fusion processes, reducing the interference of conflicting or unreliable evidence on the final result. Then, the obtained fusion weights are used to weight and correct all original evidence before full evidence fusion is completed using the DS fusion rule, ultimately yielding the global overall reliability. .

[0042] In this embodiment, when a sensor generates an abnormal reading due to a fault or strong interference, the difference between it and other evidence will increase significantly. Consequently, By reducing the amount of abnormal evidence, it can be automatically weakened during the final fusion, thus avoiding the pollution of the global judgment by a single point of bad data.

[0043] For example, in an air compressor station group, if the pressure transmitter of a certain unit experiences a jump due to electromagnetic interference, and the evidence differs significantly from that of other units / temperature and vibration evidence, it will be automatically downweighted; after fusion... This indicates whether the entire gas station group is currently in a normal gas supply state.

[0044] Furthermore, the evidence conflict coefficient is extracted from the global reliability, and positive and negative cumulative statistics of the evidence conflict coefficient relative to the steady-state mean are constructed based on the evidence conflict coefficient, the mean of the evidence conflict coefficient at the preset steady state, and the mean offset of the evidence conflict coefficient after the preset drift. If either the positive or negative cumulative statistics exceeds the cumulative threshold, the drift detection result is determined to be a current drift in the industrial field of the controlled equipment.

[0045] In this embodiment, the evidence conflict coefficient can be extracted from the fused global confidence. As a monitoring statistic A larger value indicates a more significant disagreement between different sensors. For example, the global confidence level at the current moment can be quantified into a conflict measure, aggregating the overall degree of disagreement among all sensor evidence over that period into a single scalar, which is the evidence conflict coefficient at the current moment t. It directly characterizes the total level of discrepancies in judgments among multiple sensors at the current moment. By repeating the above steps sequentially according to the control cycle, the time-varying value can be obtained. The time series is used as the input monitoring statistics for drift detection.

[0046] This embodiment can rely on historical data under steady-state conditions to pre-calculate the steady-state time. mean The mean shift after drift occurs is This allows us to construct a two-sided CUSUM cumulative sum statistic: the positive cumulative sum. Negative cumulative sum Set a judgment threshold. ,when or Exceeding the threshold At that time, it was determined that a conceptual drift had occurred at the industrial site of the equipment to be controlled: .

[0047] It should be noted that the judgment threshold in this embodiment... The following method can be used to select the average number of cycles between two false alarms, assuming the production / monitoring process is completely normal and no abnormal deviations occur. Large enough (e.g.) (one control cycle), thereby reducing the false alarm probability under normal operating conditions to an extremely low level; after drift occurs, the average delay period from the occurrence of the drift to the detection of an alarm on the control chart is desired. Small enough to enable rapid anomaly identification. For small shifts in a Gaussian distribution, and Approximately satisfy (Siegmund revised) (This represents a small offset that occurs during the process). In this embodiment, it can be based on the currently acceptable false alarm rate of the controlled equipment in the industrial field. By performing a reverse lookup of this relationship, the judgment threshold can be obtained. .

[0048] S130. Based on the drift detection results, perform adaptive reference vector enhancement for irregular frontiers to obtain an adaptively adjusted reference vector set.

[0049] This embodiment can dynamically adjust the position and direction of the reference vector according to the drift signal and the current non-dominated solution set distribution, so that it fits the irregular Pareto front.

[0050] In one optional implementation, adaptive reference vector enhancement for irregular frontiers is performed based on drift detection results to obtain an adaptively adjusted reference vector set, which may include:

[0051] Generate an initial uniform reference vector set; where each reference vector in the initial uniform reference vector set is a unit direction vector, corresponding to a search direction in the multi-object space; select the current reference vector from the initial uniform reference vector set, associate the current reference vector with the non-dominated solution with the closest angular distance, and obtain the non-dominated solution subset corresponding to the current reference vector; calculate the corresponding coverage density for the current reference vector based on the non-dominated solution subset corresponding to the current reference vector.

[0052] This embodiment can use a simplex lattice point design to generate an initial uniform reference vector set. each This is a unit direction vector, corresponding to an independent search direction in the multi-objective search space, providing a basic uniform search anchor point for subsequent multi-objective optimization. Each reference vector... By associating it with the non-dominated solution that has the closest angular distance to it, we obtain the subset of non-dominated solutions corresponding to the reference vector: Then, the area coverage density corresponding to the reference vector is calculated: .

[0053] Furthermore, if the coverage density corresponding to the current reference vector is less than a preset density threshold, the current reference vector is determined to point to a sparse region. This process continues until each reference vector in the initial uniform reference vector set has been processed, identifying the sparse region corresponding to the initial uniform reference vector set. Valid non-dominated solutions are selected from the sparse region corresponding to the initial uniform reference vector set, and these valid non-dominated solutions are converted into unit direction vectors. Based on the unit direction vectors of the valid non-dominated solutions and a specified adjustment step size, the current reference vector is updated until all reference vectors have been updated, resulting in an adaptively adjusted reference vector set. The specified adjustment step size is determined based on the drift detection results.

[0054] This embodiment allows for pre-setting of the density threshold. ,when At that time, the judgment This points to sparse or invalid regions on the frontier, meaning the solution distribution in the current region is severely uneven and requires further optimization. Then, an effective non-dominated solution is selected at the edge of the identified sparse region. It can Convert to unit direction vector Then, the reference vector at the corresponding position is updated through weighted interpolation: ,in To adjust the step size, if the drift detection result is In other words, the system has detected a conceptual drift in the field conditions of the controlled equipment, and the load can be appropriately increased. To quickly adapt to new frontiers; subsequently to Normalization yields the final adaptively adjusted reference vector set. (This embodiment...) You can decide to adjust the step size. Size.

[0055] For example, in an air compressor station cluster, the three optimization objectives could be "air supply pressure deviation," "total energy consumption," and "equipment wear." These three objectives are inherently conflicting. For instance, forcibly reducing energy consumption can easily lead to unstable air supply pressure and frequent equipment start-ups and shutdowns, exacerbating wear. There is no solution that can simultaneously achieve absolute optimality for all three objectives. The initial uniform vector might place too many search directions in the infeasible region of "low energy consumption but unstable pressure." After density sensing identifies unsolvable regions, it pulls these vectors towards the "stable and energy-saving" trade-off region where the actual Pareto front is densely distributed, increasing the probability of finding the knee solution.

[0056] S140. The multi-agent system is modeled as a non-cooperative aggregation game. Based on the adaptively adjusted reference vector set, the non-dominated solution set is calculated using a differential privacy-preserving distributed conditional gradient algorithm.

[0057] This embodiment can model a multi-agent system as a non-cooperative aggregation game and use a differential privacy-preserving distributed Frank-Wolfe algorithm to solve the Pareto optimal solution set without projection. This can solve the problem that when multiple devices optimize locally, they fight their own battles and damage the global operational constraints.

[0058] In one alternative implementation, modeling the multi-agent system as a non-cooperative aggregation game may include:

[0059] Decision variables are determined for the agents based on the operating state of the equipment to be controlled; global aggregate variables are determined for the industrial site of the equipment to be controlled based on the decision variables of each agent; wherein, each agent has a corresponding cost function, which is a single-objective aggregate cost obtained by projecting along the reference vector direction of the adaptively adjusted reference vector set.

[0060] Taking an air compressor cluster as an example, it can be used to... Taiwan air compressor abstracted as Each is an independent intelligent agent. Decision variables It can represent its load / unload status, pressure setpoint, operating frequency, and other adjustable operating parameters; the global aggregate variable can be represented as the average of the decision variables of each agent. , The operating status of all air compressors is aggregated to characterize the overall operating level of the entire air compressor cluster.

[0061] Cost function for each agent It not only considers the energy consumption and wear costs of a single unit, but also incorporates the overall pressure deviation of the entire station into the calculation, naturally binding the optimization objective of a single unit to the global operating state, ultimately leading to the Nash equilibrium point. It can satisfy the constraint: for any agent i and any decision change ,have This constraint can be understood as follows: under equilibrium conditions, no single air compressor can further reduce its overall cost by unilaterally adjusting its operating decisions. In actual air compressor station operation, each air compressor wants to reduce its own energy consumption, but individually reducing its load may lead to insufficient pressure across the entire station, resulting in a global aggregate variable. By incorporating the overall station pressure level into the cost function of each unit, local optimization naturally takes into account global constraints, avoiding "each fighting its own battle," and ultimately achieving coordinated optimization that balances the energy efficiency of individual units and the stability of the entire station's gas supply.

[0062] In one alternative implementation, the non-dominated solution set is computed using a differential privacy-preserving distributed conditional gradient algorithm based on the adaptively adjusted reference vector set, which may include:

[0063] For each iteration, perform the following operations: Calculate the local gradient for the current agent based on its cost function; inject noise into the local gradient to obtain the noise-injected local gradient; solve for the optimal solution in the reference vector direction of the adaptively adjusted reference vector set based on the noise-injected local gradient; update the decision variables of the current agent according to the optimal solution and the specified optimization step size to obtain the updated decision variables for the next iteration; wherein the updated decision variables fall within the feasible region corresponding to the reference vector; concatenate the optimal solutions in each reference vector direction of the adaptively adjusted reference vector set to obtain the non-dominated solution set.

[0064] Adaptive reference vector set Each unit direction reference vector Each of these corresponds to an independent search direction in the multi-objective space, decomposing the complex multi-objective aggregation game problem into N parallel single-objective subproblems, each subproblem following a specific direction. The specified direction is used to find the optimal trade-off solution for multiple objectives. The cost function in this embodiment... It is itself along the current reference vector The single-object aggregation cost obtained from directional projection, and the calculated gradient. The convergence direction of the optimal solution naturally points to the direction of the reference vector, thus avoiding the confusion and conflict of multi-objective gradients.

[0065] To defend against gradient inversion attacks, the calculated gradient is... Injecting Laplace noise: Among them, noise scale , Gradient Sensitivity For privacy budgets. The noisy gradient after injecting Laplacian noise still retains the gradient along... The main downward trend of the direction is determined without revealing the original operational gradient information of individual agents, and without disrupting the optimization logic of the reference vector-specified direction while protecting data privacy.

[0066] Then solve the linear subproblem. The optimization range of this subproblem is strictly constrained to the reference vector. Within the corresponding feasible subspace, the optimal candidate step size point in that direction is found quickly and directly, significantly improving the search efficiency of the sparse frontier region. Furthermore, the agent's decision variables are updated using a convex combination method. ,because and All are within the feasible region Inside, the updated It must also be within the feasible region, ensuring that each iteration steadily converges towards the optimal solution along the direction of the reference vector, without deviating from the preset search direction. For example, each air compressor calculates locally "how energy consumption / pressure / wear changes under the current decision" (gradient), but adds noise to the gradient before sending it to its neighbors; the neighbors cannot deduce the machine's process parameters from the noise gradient. At the same time, it only requires solving a linear programming problem (such as "finding the frequency that minimizes the weighted objective within the pressure upper and lower limits"), which is an order of magnitude faster than projective quadratic programming, making it suitable for PLC (Programmable Logic Controller) cycles.

[0067] Noisy auxiliary variables are exchanged among agents, and the global average state estimate is updated through a consensus protocol. The iteration converges to the Nash equilibrium of the aggregate game. It is a convex function and Lipschitz continuous, with a step size of satisfy , At that time, the algorithm convergence rate is Superior to projected gradient descent .

[0068] In this embodiment, after all subproblems corresponding to the reference vectors complete the Frank-Wolfe iterations in parallel, each The optimal trade-off solutions obtained in each direction will be pieced together to form a complete set of non-dominated solutions.

[0069] S150. Execute a graded response based on the drift detection results, and select control commands from the non-dominated solution set according to the current field requirements of multiple devices to be controlled, and issue them to control the devices.

[0070] In this embodiment, a graded response can be executed according to the drift intensity. Control commands are selected from the calculated non-dominated solution set and issued, and feedback data is sent back to S110 to form a closed loop.

[0071] In one alternative implementation, performing a graded response based on the drift detection results may include: if a slight drift is determined based on the drift detection results, making slight adjustments to a specified adjustment step size and a specified optimization step size according to a specified adjustment magnitude; if a severe drift is determined based on the drift detection results, re-initializing the partial solutions corresponding to the adaptively adjusted reference vector set and increasing the exploration ratio to adapt to the new operating condition distribution.

[0072] This embodiment can match different optimization strategies according to the severity of the drift in the field conditions, avoiding unnecessary waste of computing power:

[0073] For weak drift scenarios ( but (Slightly exceeding the threshold): Only fine-tune the step size of the Frank-Wolfe iterations. and reference vector adjustment magnitude To maintain the current population's continued evolution and adapt to minor fluctuations in operating conditions with minimal adjustment costs.

[0074] For strong drift scenarios ( (Significantly exceeding limits or multiple consecutive alarms): Triggers partial restart of the optimization population, reinitializes partial solutions, increases the exploration ratio of the algorithm, quickly jumps out of the convergence region of the old working conditions, adapts to the new working condition distribution, and avoids the algorithm getting stuck in local optima in the drastically changing field environment.

[0075] This embodiment exemplifies how the final solution can be selected from the non-dominated solution set based on the operational priorities of the equipment at the industrial site (e.g., whether energy saving or pressure stability is currently more important). It is encapsulated into loading / unloading, pressure setting, and frequency adjustment commands for each unit. The command is executed via PLC.

[0076] Optionally, after selecting control commands from the non-dominated solution set to control the devices based on the current field requirements of multiple devices to be controlled, the process may further include: collecting new actual sensor data from the multiple devices to be controlled; and reconstructing a new evidence body based on the new actual sensor data to re-control the devices.

[0077] The actual operating data of each device to be controlled can be collected and used as new evidence input into S110 to complete the closed-loop iteration of "perception-optimization-game-execution".

[0078] To enable those skilled in the art to better understand the multi-agent system collaborative control method of this embodiment, the following is an implementation scenario. Taking an industrial air compressor station group joint control and energy-saving scheduling scenario as an example, a factory operates 5-10 air compressors in parallel, supplying air to the same pipeline network. The core control objective is synchronous optimization: stable air supply pressure (small pressure deviation), total energy consumption of the units (low energy consumption), and equipment wear life (few start-stop / load switching). It needs to adapt to dynamic changes in air load, equipment performance degradation, and ambient temperature fluctuations, while protecting the privacy of local process parameters for each unit and adapting to the computing power limitations of edge PLCs. Specifically:

[0079] Step 1: Multi-source data access and preprocessing

[0080] The system receives real-time multi-source heterogeneous data from the air compressor station group: exhaust pressure, exhaust temperature, operating current, and vibration spectrum (high-frequency time series) for each unit; ambient temperature / humidity and air flow rate at the user end (operating environment); and safe operating thresholds and maintenance records (structured data) for each unit. The data is then denoised, timestamp aligned, and normalized to obtain the periodicity. The k-th sensor data set .

[0081] Step 2: Multi-source evidence fusion and concept drift detection

[0082] Each preprocessed sensor data point is treated as an independent piece of evidence, and calculations are performed. , , Calculate the difference distance between evidence; calculate the credibility and fusion weight of each evidence body; perform weighted fusion to obtain the global credibility. .

[0083] from Extracting the conflict coefficient Perform cumulative calculations; when the positive or negative cumulative sum statistic exceeds a threshold... When a condition concept drift occurs, output the following: .

[0084] Industrial Example: High summer temperatures cause the overall exhaust temperature of air compressors to rise, and the conflict coefficient between the pressure transmitter and temperature sensor data gradually increases. Accumulation and calculation result in drift being detected within 5 to 10 control cycles, triggering subsequent optimization module adjustments.

[0085] Step 3: Adaptive reference vector dynamic adjustment

[0086] Using "pressure deviation, energy consumption, and wear" as the target space, an initial reference vector set is generated using simplex lattice points. The iteration process calculates the coverage density of each vector to identify sparse regions; it dynamically adjusts the vector directions to align with the Pareto front under the current operating conditions. Then increase the adjustment step size. .

[0087] Industrial example: During peak gas consumption, vectors in the "low energy consumption" direction may fall into unsolvable regions; after the system detects the sparseness of this region, it pulls the vectors toward the actual trade-off front of "medium energy consumption but high stability" to avoid wasting search resources.

[0088] Step 4: Privacy-preserving distributed multi-objective optimization solution

[0089] Treat each air compressor as an intelligent agent. Calculate the global aggregation variables. (e.g., setting the average pressure across the entire station). Optimize the local cost function based on the Nash equilibrium conditions of each agent. (Self-energy consumption + global pressure deviation penalty).

[0090] During iteration, Laplace noise is injected into the local gradient to solve the Frank-Wolfe linear subproblem, and the decision variables are updated using convex combinations. Agents only exchange noisy auxiliary state variables and converge to the Pareto optimal solution set (i.e., the non-dominated solution set) through a consensus protocol.

[0091] Industrial example: The gradient calculated locally by air compressor No. 1 contains the local efficiency curve information, which is then sent to the neighbors after adding noise; the neighbors cannot deduce the specific efficiency parameters of No. 1, but the entire station can still converge to a balanced loading scheme after several iterations.

[0092] Step 5: Drift-graded response and decision execution

[0093] according to Graded response based on drift intensity: fine-tuning for weak drift and The strong drift portion restarts the population. Based on the field priority, the final solution is selected from the non-dominated solution set, encapsulated into loading / unloading, pressure setting, and frequency adjustment instructions for each unit, and sent to the PLC.

[0094] Step 6: Closed-loop feedback and iterative optimization

[0095] After the instruction is executed, new pressure, temperature, current, and energy consumption data are collected and used as evidence for the next cycle in step 1 to complete the closed-loop iteration.

[0096] The technical solution of this invention collects multi-source sensor data from multiple devices to be controlled; constructs an evidence body based on the multi-source sensor data for drift detection, and obtains drift detection results; performs adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtains an adaptively adjusted reference vector set; models the multi-agent system as a non-cooperative aggregation game, and calculates the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm; executes a hierarchical response according to the drift detection results, and selects control commands from the non-dominated solution set to control the devices according to the current field requirements of multiple devices to be controlled. By employing sensing, optimization, game theory, and execution coordination techniques, this invention solves the problems of existing technologies, such as the inability to detect operating condition drift in a timely manner, the difficulty of adapting fixed reference vectors to irregular Pareto fronts, privacy risks in distributed communication, and high computational complexity of projection. This improves dynamic adaptability, overcomes high-dimensional optimization bottlenecks, and achieves lightweight computation under privacy protection.

[0097] Example 2

[0098] Figure 2 This is a schematic diagram of a multi-agent system cooperative control device provided in Embodiment 2 of the present invention. The multi-agent system includes multiple devices to be controlled, each modeled as an agent, and these devices are interconnected, such as... Figure 2 As shown, the device includes: a data acquisition module 210, a drift detection module 220, an adjusted reference vector acquisition module 230, a non-dominated solution set calculation module 240, and an industrial field control module 250. Wherein:

[0099] Data acquisition module 210 is used to acquire multi-source sensor data of the multiple devices to be controlled;

[0100] Drift detection module 220 is used to construct evidence based on the multi-source sensor data to perform drift detection and obtain drift detection results;

[0101] The adjusted reference vector acquisition module 230 is used to perform adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtain an adaptive adjusted reference vector set;

[0102] The non-dominated solution set calculation module 240 is used to model the multi-agent system as a non-cooperative aggregation game, and calculate the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm.

[0103] The industrial field control module 250 is used to execute a graded response based on the drift detection results, and to select control commands from the non-dominated solution set according to the current field requirements of the multiple devices to be controlled, and issue them to control the devices.

[0104] The technical solution of this invention collects multi-source sensor data from multiple devices to be controlled; constructs an evidence body based on the multi-source sensor data for drift detection, and obtains drift detection results; performs adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtains an adaptively adjusted reference vector set; models the multi-agent system as a non-cooperative aggregation game, and calculates the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm; executes a hierarchical response according to the drift detection results, and selects control commands from the non-dominated solution set to control the devices according to the current field requirements of multiple devices to be controlled. By employing sensing, optimization, game theory, and execution coordination techniques, this invention solves the problems of existing technologies, such as the inability to detect operating condition drift in a timely manner, the difficulty of adapting fixed reference vectors to irregular Pareto fronts, privacy risks in distributed communication, and high computational complexity of projection. This improves dynamic adaptability, overcomes high-dimensional optimization bottlenecks, and achieves lightweight computation under privacy protection.

[0105] Optional, the drift detection module 220 can be used for:

[0106] Define a state identification framework for the multi-source sensor data;

[0107] For the k-th sensor data and the state identification framework, calculate the normal state confidence and abnormal state confidence of the k-th sensor data, and determine the uncertain state confidence of the k-th sensor data based on the normal state confidence and the abnormal state confidence.

[0108] Based on the normal state confidence, abnormal state confidence, and uncertain state confidence of the k-th sensor data, an evidence body is constructed for the k-th sensor data.

[0109] Calculate the difference distance between each pair of evidence pieces, and calculate the credibility of the kth evidence piece based on the difference distance between each pair of evidence pieces;

[0110] The credibility of all evidence is normalized to obtain the fusion weight of each evidence.

[0111] Based on the fusion weights, all evidence is fused using the weighted DS rule to obtain the global confidence score.

[0112] The evidence conflict coefficient is extracted from the global confidence, and positive and negative cumulative statistics of the evidence conflict coefficient relative to the steady-state mean are constructed based on the evidence conflict coefficient, the mean of the evidence conflict coefficient at the preset steady state, and the mean offset of the evidence conflict coefficient after the preset drift.

[0113] If either the positive cumulative statistic or the negative cumulative statistic exceeds the cumulative sum threshold, the drift detection result is determined to be a current drift in the industrial field of the controlled equipment.

[0114] Optionally, the adjusted reference vector acquisition module 230 can be used for:

[0115] Generate an initial uniform reference vector set; wherein each reference vector in the initial uniform reference vector set is a unit direction vector, corresponding to a search direction in the multi-object space;

[0116] Select a current reference vector from the initial uniform reference vector set, and associate the current reference vector with the non-dominated solution with the closest included angle distance to obtain a subset of non-dominated solutions corresponding to the current reference vector;

[0117] Calculate the corresponding coverage density for the current reference vector based on the non-dominated solution subset corresponding to the current reference vector;

[0118] If the coverage density corresponding to the current reference vector is less than a preset density threshold, it is determined that the current reference vector points to a sparse region.

[0119] Until each reference vector in the initial uniform reference vector set has been processed, the sparse region corresponding to the initial uniform reference vector set is identified.

[0120] Select effective non-dominated solutions in the sparse region corresponding to the initial uniform reference vector set, and convert the effective non-dominated solutions into unit direction vectors;

[0121] Based on the unit direction vector of the effective non-dominated solution and the specified adjustment step size, the current reference vector is updated until all reference vectors are updated, thus obtaining the adaptively adjusted reference vector set; wherein, the specified adjustment step size is determined based on the drift detection result.

[0122] Optional, the non-dominated solution set calculation module 240 can be used for:

[0123] The device to be controlled is modeled as an intelligent agent, and decision variables are determined for the intelligent agent based on the operating state of the device to be controlled.

[0124] Global aggregate variables are determined for the industrial site of the multiple devices to be controlled based on the decision variables of each agent;

[0125] Each agent has a corresponding cost function, which is a single-objective aggregate cost obtained by projecting along the reference vector direction of the adaptively adjusted reference vector set.

[0126] Optionally, the non-dominated solution set calculation module 240 can also be used for:

[0127] Perform the following operations for each iteration:

[0128] Calculate the local gradient for the current agent based on the cost function of the current agent;

[0129] Noise is injected into the local gradient to obtain the local gradient after noise injection;

[0130] Based on the local gradient after the injection of noise, the optimal solution is solved in the direction of the reference vector in the adaptively adjusted reference vector set.

[0131] Based on the optimal solution and the specified optimization step size, the decision variables of the current agent are updated to obtain the updated decision variables for the next iteration process; wherein, the updated decision variables fall within the feasible region corresponding to the reference vector;

[0132] The optimal solutions in each reference vector direction of the adaptively adjusted reference vector set are concatenated to obtain the non-dominated solution set.

[0133] Optional, the Industrial Field Control Module 250 can be used for:

[0134] If a slight drift is determined based on the drift detection results, a slight adjustment is made to the specified adjustment step size and the specified optimization step size according to the specified adjustment range;

[0135] If a severe drift is determined based on the drift detection results, the partial solutions corresponding to the adaptively adjusted reference vector set are reinitialized, and the exploration ratio is increased to adapt to the new operating condition distribution.

[0136] Optionally, the multi-agent system collaborative control device further includes a data feedback module, used for:

[0137] Collect new actual sensor data from the multiple devices to be controlled;

[0138] Based on the new actual sensor data, a new body of evidence is constructed to re-control the device.

[0139] The industrial control device provided in the embodiments of the present invention can execute the industrial control method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0140] Example 3

[0141] Figure 3 A schematic diagram of an electronic device 300 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers or various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0142] like Figure 3 As shown, the electronic device 300 includes at least one processor 301 and a memory, such as a read-only memory (ROM) 302 or a random access memory (RAM) 303, communicatively connected to the at least one processor 301. The memory stores computer programs executable by the at least one processor. The processor 301 can perform various appropriate actions and processes based on the computer program stored in the ROM 302 or loaded into the RAM 303 from storage unit 308. The RAM 303 can also store various programs and data required for the operation of the electronic device 300. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0143] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0144] Processor 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 301 performs the various methods and processes described above, such as industrial control methods.

[0145] In some embodiments, the industrial control method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by processor 301, one or more steps of the industrial control method described above may be performed. Alternatively, in other embodiments, processor 301 may be configured to perform the industrial control method by any other suitable means (e.g., by means of firmware).

[0146] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0147] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0148] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable 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 thereof.

[0149] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0150] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0151] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0152] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0153] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A cooperative control method for a multi-agent system, characterized in that, The multi-agent system includes multiple devices to be controlled, each modeled as an agent, and these devices are interconnected. The method includes: Collect multi-source sensor data from the multiple devices to be controlled; Based on the multi-source sensor data, an evidence body is constructed to perform drift detection, and a drift detection result is obtained. Based on the drift detection results, adaptive reference vector enhancement is performed for irregular frontiers to obtain an adaptively adjusted reference vector set. The multi-agent system is modeled as a non-cooperative aggregation game, and the non-dominated solution set is calculated based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm. A graded response is executed based on the drift detection results, and control commands are selected from the non-dominated solution set and issued to control the devices based on the current field requirements of the multiple devices to be controlled.

2. The method according to claim 1, characterized in that, An evidence body is constructed based on the multi-source sensor data to perform drift detection, and the drift detection results are obtained, including: Define a state identification framework for the multi-source sensor data; For the k-th sensor data and the state identification framework, calculate the normal state confidence and abnormal state confidence of the k-th sensor data, and determine the uncertain state confidence of the k-th sensor data based on the normal state confidence and the abnormal state confidence. Based on the normal state confidence, abnormal state confidence, and uncertain state confidence of the k-th sensor data, an evidence body is constructed for the k-th sensor data. Calculate the difference distance between each pair of evidence pieces, and calculate the credibility of the kth evidence piece based on the difference distance between each pair of evidence pieces; The credibility of all evidence is normalized to obtain the fusion weight of each evidence. Based on the fusion weights, all evidence is fused using the weighted DS rule to obtain the global confidence score. The evidence conflict coefficient is extracted from the global confidence level, and positive and negative cumulative statistics of the evidence conflict coefficient relative to the steady-state mean are constructed based on the evidence conflict coefficient, the mean of the evidence conflict coefficient at the preset steady state, and the mean offset of the evidence conflict coefficient after the preset drift. If either the positive cumulative statistic or the negative cumulative statistic exceeds the cumulative sum threshold, the drift detection result is determined to be a current drift in the industrial field of the controlled equipment.

3. The method according to claim 1, characterized in that, Based on the drift detection results, adaptive reference vector enhancement is performed for irregular leading edges to obtain an adaptively adjusted reference vector set, including: Generate an initial uniform reference vector set; wherein each reference vector in the initial uniform reference vector set is a unit direction vector, corresponding to a search direction in the multi-object space; Select the current reference vector from the initial uniform reference vector set, and associate the current reference vector with the non-dominated solution with the closest included angle distance to obtain the non-dominated solution subset corresponding to the current reference vector; Calculate the corresponding coverage density for the current reference vector based on the non-dominated solution subset corresponding to the current reference vector; If the coverage density corresponding to the current reference vector is less than a preset density threshold, it is determined that the current reference vector points to a sparse region. Until each reference vector in the initial uniform reference vector set has been processed, the sparse region corresponding to the initial uniform reference vector set is identified. Select effective non-dominated solutions in the sparse region corresponding to the initial uniform reference vector set, and convert the effective non-dominated solutions into unit direction vectors; Based on the unit direction vector of the effective non-dominated solution and the specified adjustment step size, the current reference vector is updated until all reference vectors are updated, thus obtaining the adaptively adjusted reference vector set; wherein, the specified adjustment step size is determined based on the drift detection result.

4. The method according to claim 1, characterized in that, The multi-agent system is modeled as a non-cooperative aggregation game, including: Decision variables are determined for the agent based on the operating state of the device to be controlled; Global aggregate variables are determined for the industrial site of the multiple devices to be controlled based on the decision variables of each agent; Each agent has a corresponding cost function, which is a single-objective aggregate cost obtained by projecting along the reference vector direction of the adaptively adjusted reference vector set.

5. The method according to claim 4, characterized in that, Based on the adaptively adjusted reference vector set, the non-dominated solution set is calculated using a differential privacy-preserving distributed conditional gradient algorithm, including: Perform the following operations for each iteration: Calculate the local gradient for the current agent based on the cost function of the current agent; Noise is injected into the local gradient to obtain the local gradient after noise injection; Based on the local gradient after the injection of noise, the optimal solution is solved in the direction of the reference vector in the adaptively adjusted reference vector set. Based on the optimal solution and the specified optimization step size, the decision variables of the current agent are updated to obtain the updated decision variables for the next iteration process; wherein, the updated decision variables fall within the feasible region corresponding to the reference vector; The optimal solutions in each reference vector direction of the adaptively adjusted reference vector set are concatenated to obtain the non-dominated solution set.

6. The method according to claim 1, characterized in that, A graded response is executed based on the drift detection results, including: If a slight drift is determined based on the drift detection results, a slight adjustment is made to the specified adjustment step size and the specified optimization step size according to the specified adjustment range; If a severe drift is determined based on the drift detection results, the partial solutions corresponding to the adaptively adjusted reference vector set are reinitialized, and the exploration ratio is increased to adapt to the new operating condition distribution.

7. The method according to claim 1, characterized in that, After selecting control commands from the non-dominated solution set based on the current field requirements of the multiple devices to be controlled and issuing them to control the devices, the process further includes: Collect new actual sensor data from the multiple devices to be controlled; Based on the new actual sensor data, a new body of evidence is constructed to re-control the device.

8. A multi-agent system cooperative control device, characterized in that, The multi-agent system includes multiple devices to be controlled, each modeled as an agent, and these multiple devices are interconnected. The device includes: The data acquisition module is used to acquire multi-source sensor data from the multiple devices to be controlled; The drift detection module is used to construct an evidence body based on the multi-source sensor data to perform drift detection and obtain drift detection results; The adjusted reference vector acquisition module is used to perform adaptive reference vector enhancement for irregular frontiers based on the drift detection results, and obtain an adaptive adjusted reference vector set. The non-dominated solution set calculation module is used to model the multi-agent system as a non-cooperative aggregation game and calculate the non-dominated solution set based on the adaptively adjusted reference vector set using a differential privacy-preserving distributed conditional gradient algorithm. The equipment control module is used to execute a graded response based on the drift detection results, and to select control commands from the non-dominated solution set according to the current field requirements of the multiple devices to be controlled in order to control the devices.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a multi-agent system cooperative control method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute and implement the multi-agent system cooperative control method according to any one of claims 1-7.

Images