A new energy power scientific research multi-agent collaborative whole-process intelligent optimization method

Abstract

This invention discloses a multi-agent collaborative intelligent optimization method for the entire process of new energy power research, relating to the field of new energy power technology. The method includes: receiving user-submitted control requirements, experimental objectives, and optimization objectives for new energy power research; constructing an initial strategy scheme through multiple agents in an intelligent optimization server; performing multi-agent collaborative verification in a digital twin environment; outputting a preliminary optimization strategy; outputting semi-physical verification results and optimization suggestions; further optimizing the preliminary optimization strategy; converting the optimized strategy into code adapted to a digital signal processor real-time control unit; downloading the compiled code to a testing device; collecting real-time data during the operation of the testing device; analyzing the real-time data; and visually monitoring the entire process's operational status. This invention improves the efficiency of new energy power research, achieving seamless integration and intelligent optimization of the entire research process.

Description

Technical Field

[0001] This invention relates to the field of new energy power technology, specifically to a multi-agent collaborative intelligent optimization method for the entire process of new energy power research. Background Technology

[0002] With the high proportion of new energy sources being connected to the power grid and the increasing level of electronicization in the power system, the complexity and coupling of new energy power research objects are constantly increasing, and the demand for accuracy and efficiency in scientific research experiments is becoming increasingly urgent. At present, new energy power research work usually relies on decentralized experimental equipment and independent operating procedures. A complete research task needs to go through multiple stages in sequence, such as demand analysis, strategy design, model building, digital twin verification, semi-physical testing, control deployment, data acquisition, and analysis and optimization. Each stage is undertaken independently by different servers, controllers or software, lacking a unified collaborative mechanism and scheduling system.

[0003] In existing technologies, the research process suffers from significant fragmentation: the strategy transfer efficiency between the digital twin verification stage and the semi-physical testing stage is low; optimized strategies in the digital twin environment are difficult to adapt quickly to real hardware scenarios, requiring extensive manual parameter adjustments, which prolongs the research cycle; experimental data, strategy solutions, and optimization conclusions generated at each stage are scattered across different devices, making centralized storage, sharing, and reuse impossible, resulting in numerous data silos. This not only increases repetitive research workload but also hinders the accumulation of research experience, leading to extremely low utilization of research assets; furthermore, the coordination of various devices and stages during the research process relies on manual scheduling. Problems such as process bottlenecks and rhythm imbalances are prone to occur, and it is difficult to achieve real-time monitoring and dynamic adjustment of the entire process, resulting in low research efficiency and insufficient reliability of experimental results. In addition, existing research methods lack intelligent guidance mechanisms in the iterative optimization process. When a strategy that meets the preset evaluation indicators is not obtained, it is impossible to quickly identify invalid iterative paths and parameter redundancy problems, resulting in an excessively large iterative search range and low optimization efficiency. At the same time, there is a lack of intelligent means to monitor the status of core components such as printed circuit boards and real-time control units of digital signal processors during the research process. Component defects can easily lead to experimental deviations and affect the accuracy of research results. Summary of the Invention

[0004] To address the aforementioned technical problems, this paper presents a multi-agent collaborative intelligent optimization method for the entire process of new energy power research. This technical solution resolves the problems mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A multi-agent collaborative intelligent optimization method for the entire process of new energy power research includes: Receive user-submitted research and control requirements, experimental objectives, and optimization goals for new energy power; An initial strategy scheme is constructed by multiple intelligent agents in the intelligent optimization server; The initial strategy plan is sent to multiple agents in the digital twin intelligent server, and multi-agent collaborative verification is performed in the digital twin environment; Based on user interaction feedback and verification results, the initial strategy scheme is revised in multiple rounds through a multi-agent iterative optimization mechanism to output a preliminary optimized strategy. The initial optimization strategy and test data are sent to multiple agents in the semi-physical intelligent server. Real-time closed-loop verification of the multi-agent system is performed in the semi-physical environment, and the semi-physical verification results and optimization suggestions are output. Based on the semi-physical verification results and optimization suggestions, the preliminary optimization strategy is further optimized through the collaboration of multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server. The optimized strategy is then converted into code that adapts to the real-time control unit of the digital signal processor. The code is compiled, and the compiled code is downloaded to the test device; The data acquisition agent in the semi-physical intelligent server collects real-time data during the operation of the test device. Analyze real-time data and feed the analysis results and test data back to the intelligent optimization server to achieve data sharing and asset accumulation among multiple agents; Through the industrial touch control and integrated management system, each intelligent agent is uniformly scheduled, and the entire process operation status is visualized and monitored.

[0006] Preferably, the initial strategy scheme is constructed by multiple intelligent agents in the intelligent optimization server, specifically including: Read historical scientific research experimental data, existing strategy results, and user requirement parameters; The model-trained agent is used to train an optimized model adapted to new energy power research scenarios based on historical scientific research experimental data; The model-inference agent, in conjunction with user requirement parameters, invokes the optimized model to perform policy inference. The intelligent optimization agent makes initial adjustments to the policy reasoning results; Output an initial strategy solution that meets user needs and is compatible with subsequent digital twin and verification processes.

[0007] Preferably, the initial strategy is modified multiple times through a multi-agent iterative optimization mechanism to output a preliminary optimized strategy, specifically including: A digital twin verification environment is built collaboratively by digital twin agents in a digital twin intelligent server and agents constructed from the training environment. Import the initial strategy scheme into the digital twin verification environment; The strategy is executed by a digital twin intelligent agent, which simulates the dynamic changes of a real scientific research scenario. The intelligent agent constructed by the training environment dynamically adjusts its training parameters based on the digital twin verification results; Multi-agent collaborative analysis of digital twin verification data identifies invalid parameter combinations, invalid control paths, and erroneous policy directions; Based on user interaction feedback, the control parameters, model structure, constraints, and objective function were adjusted in multiple rounds. Select preliminary optimization strategies that align with user needs.

[0008] Preferably, multi-agent real-time closed-loop verification is performed in a semi-physical environment, and semi-physical verification results and optimization suggestions are output, specifically including: The verification agent, data acquisition agent, and analysis agent in the semi-physical intelligent server work together. Couple the initial optimization strategy with the actual control hardware and the object under test; Real-time closed-loop verification is performed by the verification agent; The data acquisition agent synchronously collects strategy response performance, stability indicators, dynamic process data, and error data. The analytical agent performs in-depth analysis of the collected data to identify the adaptation defects of the strategy in the semi-physical environment; Generate optimization suggestions that adapt to the defects of the semi-physical environment, and output the semi-physical verification results.

[0009] Preferably, multi-agent collaboration further optimizes the initial optimization strategy, converting the optimized strategy into code, specifically including: The intelligent optimization server agent receives the semi-physical verification results and optimization suggestions. Based on historical data and real-time feedback, develop further optimization plans; Multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server work together to modify the parameters and structure of the initial optimization strategy and form the final optimization strategy. The code-generated agent translates the final optimization strategy into code that is adapted to the real-time control unit of the digital signal processor. Compile and run the code; Download the code that passes the test to the test device.

[0010] Preferably, the implementation of multi-agent data sharing and asset accumulation includes: The data acquisition agent in the semi-physical intelligent server collects control input data, control output data, operating status data, fault data, timing data, and experimental result data during the operation of the test device; The collected data is transmitted to the intelligent optimization server; The intelligent storage agent in the intelligent optimization server centrally classifies and stores the transmitted data; The analytical agent in the intelligent optimization server extracts features from the stored data; The extracted feature data is synchronously shared with each intelligent agent, forming a research asset that can be used for subsequent strategy optimization.

[0011] Preferably, unified scheduling and visual monitoring are achieved through intelligent agents in industrial touch control and integrated management systems, specifically including: Establish communication connections between the intelligent agents of the industrial touch control and integrated management system and the intelligent optimization server, digital twin intelligent server, semi-physical intelligent server, digital signal processor real-time control unit and intelligent printed circuit board visual analysis intelligent agent; Real-time scheduling of the operational status and work progress of each intelligent agent; Coordinate the rhythm of multi-agent collaboration; The experimental plan, running process, test data and analysis results are presented through a human-computer interaction interface.

[0012] Preferably, the semi-physical real-time closed-loop verification process also includes: The intelligent printed circuit board visual analysis agent acquires images of the digital signal processor and real-time control unit of the printed circuit board and the testing device in real time. Perform defect detection, device identification, and experimental status observation on the acquired images; Output board-level defect information, device identification results, and experimental observation images; The output results are fed back to the analytical agent in the semi-physical intelligent server.

[0013] Preferably, the multi-round iterative optimization process also includes: When no strategy solution that meets the preset evaluation indicators is obtained, multiple agents work together to summarize the generated iterative data. Analyze iterative data to identify invalid iteration paths and parameter redundancy issues; Optimize the iterative algorithm based on the recognition results; Narrow the search scope for subsequent iterations.

[0014] Preferably, the digital signal processor real-time control unit executes control code, specifically including: Receive the compiled control code; Low-latency real-time control of the test device is implemented based on control code instructions; The data acquisition agent in the semi-physical intelligent server provides real-time feedback of control execution status data.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By leveraging the collaborative efforts of multiple intelligent agents within intelligent optimization servers, digital twin intelligent servers, and semi-physical intelligent servers, and combined with the unified scheduling of intelligent agents in industrial touch control and integrated management systems, the existing problems of fragmented and uncoordinated processes in new energy power research, such as demand analysis, strategy design, digital twin verification, and semi-physical testing, have been improved. This has enabled seamless integration of the entire research process, reduced manual parameter adjustments and process switching between different stages, effectively shortened the research cycle, and improved the overall efficiency of research work. 2. By using a semi-physical intelligent server to collect various real-time data during the operation of the testing device, the data is transmitted to the storage intelligent agent of the intelligent optimization server for centralized classification and storage. Then, the analysis intelligent agent extracts features and shares them with each intelligent agent. This improves the problem of data being scattered across different devices and unable to be shared and reused in existing scientific research, avoids the formation of data silos, reduces repetitive scientific research work, improves the utilization rate of scientific research data and related assets, and facilitates the accumulation and subsequent reuse of experience in the scientific research process. 3. By collaboratively summarizing and analyzing iterative data through multiple agents, invalid iterative paths and parameter redundancy issues can be quickly identified and iterative algorithms optimized, thus improving the problems of excessively large search range and low efficiency in existing scientific research iterative optimization processes. At the same time, through intelligent printed circuit board visual analysis agents, images of core components such as printed circuit boards and digital signal processor real-time control units are acquired and defects are detected, which improves the problem of lack of intelligent means for monitoring the status of core components in existing scientific research and the susceptibility to experimental deviations due to component defects, thereby improving the reliability of experimental results. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] This application provides a multi-agent collaborative intelligent optimization method for the entire process of new energy power research. Please refer to [link / reference]. Figure 1 The present application provides a multi-agent collaborative intelligent optimization method for the entire process of new energy power research, which includes the following steps: Step S101: Receive the user's research and control requirements for new energy power, experimental objectives, and optimization objectives; Step S102: Construct an initial strategy scheme through multiple intelligent agents in the intelligent optimization server; Step S103: Send the initial strategy plan to multiple agents in the digital twin intelligent server and perform multi-agent collaborative verification in the digital twin environment; Step S104: Combining user interaction feedback and verification results, the initial strategy scheme is modified in multiple rounds through a multi-agent iterative optimization mechanism to output a preliminary optimized strategy; Step S105: Send the preliminary optimization strategy and test data to multiple agents in the semi-physical intelligent server, perform real-time closed-loop verification of multiple agents in the semi-physical environment, and output the semi-physical verification results and optimization suggestions. Step S106: Based on the semi-physical verification results and optimization suggestions, the preliminary optimization strategy is further optimized through the collaboration of multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server. Step S107: Convert the optimized strategy into code adapted to the real-time control unit of the digital signal processor; Step S108: Compile the code and download the compiled code to the test device; Step S109: Collect real-time data during the operation of the test device through the data acquisition agent in the semi-physical intelligent server; Step S110: Analyze the real-time data and feed the analysis results and test data back to the intelligent optimization server to realize multi-agent data sharing and asset accumulation; Step S111: Through the industrial touch control and integrated management system, the intelligent agents are uniformly scheduled and the entire process operation status is visually monitored.

[0019] The above-mentioned scheme achieves seamless integration of the entire new energy power research process through multi-agent collaboration, solving the problems of fragmentation and poor coordination in existing research stages; through the division of labor and cooperation among intelligent optimization servers, digital twin intelligent servers, and semi-physical intelligent servers, it realizes intelligentization of the entire process from strategy construction, verification to optimization and deployment, reducing manual intervention; through centralized data storage and sharing, it avoids data silos and improves the utilization rate of research assets; and through unified scheduling and visual monitoring, it ensures the smooth operation of the research process, improving research efficiency and the reliability of experimental results.

[0020] In some embodiments, step S102, constructing an initial policy scheme through multiple agents in the intelligent optimization server, specifically includes: Step S1021: Read historical scientific research experimental data, existing strategy results, and user requirement parameters; Step S1022: The model training agent trains an optimized model adapted to the new energy power research scenario based on historical scientific research experimental data; Step S1023: The model reasoning agent calls the optimized model to perform policy reasoning based on the user requirement parameters; Step S1024: The intelligent optimization agent makes preliminary adjustments to the policy reasoning results; Step S1025: Output an initial strategy scheme that meets user needs and is compatible with subsequent digital twin and verification processes.

[0021] Specifically, in step S1021, the storage module of the intelligent optimization server pre-stores historical scientific research experimental data in the field of new energy power research, previously verified effective strategy results, and basic data such as parameter specifications of various new energy power equipment and scientific research scenario parameters. The user inputs specific scientific research control requirements through the human-machine interface of the industrial touch control and integrated management system. The scientific research control requirements include new energy power generation regulation and grid connection stability control. The user also inputs experimental objectives, such as improving control accuracy and shortening response time. The user also needs to input optimization objectives, such as reducing energy consumption and improving strategy adaptability. After receiving the above information, the intelligent optimization server's requirement receiving agent extracts the user requirement parameters and reads them together with historical data and existing strategy results into the model training agent.

[0022] In step S1022, the model training agent uses common machine learning algorithms such as gradient descent and random forest; it uses historical scientific research experimental data as training samples and user demand parameters as constraints to train an optimized model adapted to new energy power research scenarios; the optimized model can output a corresponding preliminary strategy framework according to the input control requirements and experimental objectives; the preliminary strategy framework covers core contents such as control parameter range, model structure, and constraints; the optimized model is adapted to the needs of different scenarios in new energy power research, including photovoltaic research, wind power research, and other scenarios.

[0023] In step S1023, the model inference agent calls the optimized model trained in step S1022, substitutes the user requirement parameters into the model for inference calculation, and obtains the initial policy inference result based on the model output. During the inference process, the model inference agent combines the experimental goal and optimization goal set by the user to perform preliminary screening of the core parameters of the policy to ensure that the inference result meets the user's needs.

[0024] In step S1024, the intelligent optimization agent makes preliminary adjustments to the policy reasoning results. The adjustment focuses on parameters in the policy that are incompatible with subsequent digital twin and semi-physical verification steps. The adjustment methods include adjusting the parameter range to adapt to the computing power of the digital twin server and optimizing the policy logic to adapt to the hardware characteristics of the semi-physical environment. At the same time, the intelligent optimization agent will correct problems such as parameter redundancy and logical contradictions that occur during the reasoning process.

[0025] In step S1025, the intelligent optimization agent performs compliance testing on the adjusted strategy. The testing includes confirming that the strategy meets user needs, experimental objectives, and optimization goals, and is compatible with the requirements of subsequent digital twin and verification stages. Upon successful testing, the initial strategy scheme is output, synchronously transmitted to the digital twin intelligent server, and fed back to the industrial touch control and integrated management system for user review.

[0026] This step involves multiple intelligent agents working together to build an initial strategy plan based on historical data and user needs. This avoids the subjectivity and limitations of traditional manual strategy design, improves the rationality and adaptability of the initial strategy, and lays the foundation for subsequent digital twin verification, optimization, and iteration.

[0027] In some embodiments, step S104 involves revising the initial policy scheme multiple times through a multi-agent iterative optimization mechanism to output a preliminary optimized policy, specifically including: Step S1041: The digital twin intelligent agent in the digital twin intelligent server and the intelligent agent constructed by the training environment collaboratively build a digital twin verification environment; Step S1042: Import the initial strategy scheme into the digital twin verification environment; Step S1043: The digital twin agent performs strategy verification, and the digital twin agent simulates the dynamic changes of a real scientific research scenario; Step S1044: The intelligent agent constructed from the training environment dynamically adjusts the training parameters based on the digital twin verification results; Step S1045: Multi-agent collaborative analysis of digital twin operation data to identify invalid parameter combinations, invalid control paths, and erroneous policy directions; Step S1046: Based on user interaction feedback, adjust the control parameters, model structure, constraints and objective function in multiple rounds; Step S1047: Select preliminary optimization strategies that are consistent with user needs.

[0028] Specifically, in step S1041, the digital twin agent is responsible for building the digital twin computing framework; the digital twin agent constructs a digital twin model based on the physical parameters and operating characteristics of real new energy power research scenarios. Real new energy power research scenarios include photovoltaic power plants, wind farms, and grid-connected systems; the training environment construction agent adjusts the parameters of the digital twin operating environment according to the requirements of the initial strategy scheme; the parameters of the digital twin operating environment include grid voltage, load variation range, and environmental interference factors; the three work together to build a digital twin verification environment that fits the real scenario and adapts to the initial strategy, ensuring the reliability of the digital twin verification results.

[0029] In step S1042, the digital twin intelligent server receives the initial strategy scheme sent by the intelligent optimization server, and the digital twin intelligent agent imports it into the completed digital twin verification environment to clarify the execution logic, control nodes and parameter thresholds of the strategy, so as to prepare for the subsequent execution of the digital twin.

[0030] In step S1043, the digital twin agent executes the complete strategy digital twin operation process according to the instructions of the initial strategy scheme, simulating the operation process of the strategy in a real scientific research scenario; at the same time, the digital twin agent simulates the dynamic changes of the real scientific research scenario; dynamic changes include interference factors such as changes in light intensity, wind speed, and power grid load fluctuations; to ensure that the digital twin operation process fits the actual scientific research scenario and avoids the result deviation caused by static digital twins.

[0031] In step S1044, the training environment construction agent collects the digital twin operation data output by the digital twin agent in real time, and analyzes the deviation between the digital twin verification results and the experimental and optimization objectives. After the analysis is completed, the training parameters of the digital twin operation environment are dynamically adjusted; the training parameters include the intensity of interference factors, parameter adjustment step size, etc., so that the digital twin environment better meets the policy optimization requirements and improves the efficiency of subsequent iterative optimization.

[0032] In step S1045, the digital twin agent and the training environment-constructed agent collaboratively analyze the digital twin's operational data. The analysis focuses on identifying invalid parameter combinations, invalid control paths, and erroneous policy directions that occur during the digital twin's operation. Invalid parameter combinations refer to parameter combinations with conflicting values ​​that prevent the policy from executing normally. Invalid control paths refer to control paths with redundant control logic that cause response delays. Erroneous policy directions refer to policy directions that deviate from the policy objective and user requirements. After identification, the identification results are compiled into a digital twin verification analysis report and transmitted to the intelligent optimization server.

[0033] In step S1046, after receiving the digital twin verification and analysis report, the intelligent optimization server combines the adjustment suggestions from the user through the industrial touch and integrated management system. These suggestions include adjusting control accuracy requirements and optimizing response speed. The intelligent optimization agent and the model inference agent collaborate to adjust the control parameters, model structure, constraints, and objective function of the initial strategy scheme. After each round of adjustments, the adjusted strategy is resent to the digital twin intelligent server for digital twin verification, forming an iterative cycle of adjustment, digital twin creation, and analysis.

[0034] In step S1047, after multiple rounds of iterative adjustments, when the strategy digital twin verification result meets the preset evaluation indicators, the strategy is selected as the preliminary optimization strategy, output to the semi-physical intelligent server, and synchronously fed back to the user for user confirmation; the preset evaluation indicators include control accuracy reaching the experimental target, response speed reaching the experimental target, and deviation within the allowable range, etc.

[0035] In some embodiments, step S104, during the multi-round iterative optimization process, further includes: Step S1048: When no strategy solution that meets the preset evaluation index is obtained, multiple agents collaboratively summarize the generated iterative data. Step S1049: Analyze the iterative data to identify invalid iterative paths and parameter redundancy issues; Step S10410: Optimize the iterative algorithm based on the recognition results; Step S10411: Narrow the search range for subsequent iterations.

[0036] Specifically, in step S1048, if a strategy solution that meets the preset evaluation indicators is still not obtained after a preset number of iterations, the analytical agent of the intelligent optimization server and the digital twin agent of the digital twin intelligent server collaborate to summarize all iteration data generated during the iteration process; the iteration data includes strategy parameters, digital twin data, adjustment records, etc. After summarizing, the iteration data is centrally organized and stored.

[0037] In step S1049, the iterative data after being summarized and analyzed collaboratively by multiple agents is analyzed. The analysis method involves comparing the policy parameters and digital twin results of different iteration rounds. Invalid iteration paths and parameter redundancy issues are identified through comparison. Invalid iteration paths refer to parameter adjustment directions where the digital twin verification results do not show significant improvement after multiple adjustments; parameter redundancy issues refer to problems such as redundant control parameters and repetitive policy logic; after identification, the bottlenecks in the iterative optimization process are clarified.

[0038] In step S10410, the iterative algorithm is optimized based on the identified invalid iteration paths and parameter redundancy issues. The optimization methods include adjusting the parameter search step size, optimizing the objective function weight, and eliminating the search directions corresponding to redundant parameters. Through these optimization methods, invalid iterations are avoided from consuming computing power.

[0039] In step S10411, the parameter search range for subsequent iterations is narrowed based on the optimized iterative algorithm. The search range is adjusted to focus on core parameters that significantly affect the digital twin verification results; by narrowing the search range, the efficiency of iterative optimization is improved, the iteration cycle is shortened, and a preliminary optimization strategy that meets the requirements is obtained as soon as possible.

[0040] In some embodiments, step S105 involves performing multi-agent real-time closed-loop verification in a semi-physical environment and outputting semi-physical verification results and optimization suggestions, specifically including: Step S1051: The verification agent, data acquisition agent, and analysis agent in the semi-physical intelligent server work together. Step S1052: Couple the preliminary optimization strategy with the actual control hardware and the object under test; Step S1053: Real-time closed-loop verification is performed by the verification agent; Step S1054: The data acquisition agent synchronously collects strategy response performance, stability indicators, dynamic process data, and error data; Step S1055: The analytical agent performs in-depth analysis on the collected data to identify the adaptation defects of the strategy in the semi-physical environment. Step S1056: Generate optimization suggestions to adapt to the defects of the semi-physical environment and output the semi-physical verification results.

[0041] Specifically, in step S1051, after the semi-physical intelligent server is started, the verification agent, data acquisition agent, and analysis agent are started synchronously and establish communication connections, clarifying the division of labor and cooperation mechanism; the verification agent is responsible for executing the closed-loop verification process; the data acquisition agent is responsible for collecting various types of data during the verification process; and the analysis agent is responsible for data parsing and defect identification; through clear division of labor, the verification process is ensured to be highly efficient and collaborative.

[0042] In step S1052, the verification agent couples the initial optimization strategy with the real control hardware and the object under test in the semi-physical environment; the real control hardware includes controllers, sensors, etc.; the object under test includes new energy power generation simulation devices, grid simulation loads, etc.; after coupling, a linkage relationship between strategy control and hardware execution is established to ensure that strategy instructions can directly act on real hardware and simulate the strategy execution process in real scientific research scenarios.

[0043] In step S1053, the verification agent executes real-time closed-loop verification according to the instructions of the preliminary optimization strategy, controlling the real control hardware to drive the test object to run; simultaneously, it receives real-time operating status data from the hardware, forming a closed-loop process of strategy instructions, hardware execution, and status feedback. Through this closed-loop process, the feasibility and stability of the strategy in a real hardware environment are verified.

[0044] In step S1054, the data acquisition agent synchronously collects various types of data during the verification process through sensors and data acquisition modules. The collected data includes strategy response performance data, stability indicators, dynamic process data, and error data. Strategy response performance data includes response time and adjustment speed; stability indicators include operational volatility and error range; dynamic process data includes real-time parameter change curves; and error data includes deviations between actual operating results and theoretical values. The acquisition frequency is set according to research needs to ensure the real-time nature and completeness of the data. After acquisition, the data is transmitted to the analysis agent.

[0045] In step S1055, the analytical agent performs in-depth analysis on the collected data. The analysis method is to compare the preset standard values ​​with the collected data. The preset standard values ​​include the response time and error threshold required by the experimental target. Through comparison, the adaptation defects of the strategy in the semi-physical environment are identified. The adaptation defects include response delay and insufficient stability due to mismatch between the strategy parameters and the real hardware, and error exceeding the standard due to loopholes in the strategy logic. After identification, the cause of the defects is clarified.

[0046] In step S1056, the intelligent agent generates targeted optimization suggestions based on the identified adaptation defects. The optimization suggestions include adjusting the value range of a certain type of control parameter, optimizing the execution logic of the strategy, and correcting the parameter adjustment step size. At the same time, the data and defect identification results in the verification process are organized to form a semi-physical verification result. The semi-physical verification result is synchronously transmitted to the intelligent optimization server and the industrial touch and integrated management system.

[0047] In some embodiments, step S105, during the semi-physical real-time closed-loop verification process, further includes: Step S1057: The intelligent printed circuit board visual analysis agent acquires images of the printed circuit board and the digital signal processor real-time control unit of the testing device. Step S1058: Perform defect detection, device identification, and experimental status observation on the acquired images; Step S1059: Output board-level defect information, device identification results, and experimental observation images; Step S10510: Feed the output results back to the analysis agent in the semi-physical intelligent server.

[0048] Specifically, in step S1057, the intelligent printed circuit board visual analysis agent uses a high-definition camera to perform multi-angle, real-time image acquisition of the printed circuit board and the digital signal processor real-time control unit of the testing device; the acquisition range covers the device layout, solder joints and operating status of the digital signal processor control unit of the printed circuit board; ensuring that the acquired images are clear and complete.

[0049] In step S1058, the intelligent printed circuit board visual analysis agent uses machine vision detection algorithms to perform defect detection, device identification, and experimental status observation on the acquired images. Defect detection focuses on identifying soldering defects on the printed circuit board, including cold solder joints and missing solder joints. Defect detection also includes identifying device damage, including capacitor bulging and resistor burnout. Device identification focuses on confirming whether the models and specifications of various devices on the printed circuit board are consistent with the design requirements. Experimental status observation focuses on monitoring the operation indicator lights and heat dissipation status of the digital signal processor control unit to determine whether it is working properly.

[0050] In step S1059, the intelligent printed circuit board visual analysis agent organizes the defect detection results, device identification results, and experimental observation images, and outputs board-level defect information, device identification results, and experimental observation images; the board-level defect information includes defect location, defect type, etc.; the device identification results include whether the device model matches, whether there are any missing parts, etc.; ensuring that the output information is accurate and intuitive.

[0051] In step S10510, the above output results are fed back to the analysis agent in the hardware-in-the-loop intelligent server. The analysis agent, based on these results, determines whether the data deviations during the hardware-in-the-loop verification process are caused by defects in the printed circuit board or the digital signal processor control unit. If defects are found, hardware repair suggestions are added to the optimization recommendations to improve the targeting of the strategy optimization.

[0052] In some embodiments, in step S106, the multi-agent collaborative optimization of the preliminary optimization strategy is further optimized, and the optimized strategy is converted into code, specifically including: Step S1061: The intelligent optimization server agent receives the semi-physical verification results and optimization suggestions; Step S1062: Combine historical data and real-time feedback to formulate further optimization plans; Step S1063: Multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server work together to modify the parameters and structure of the preliminary optimization strategy to form the final optimization strategy. Step S1064: The code-generating agent converts the final optimization strategy into code adapted to the real-time control unit of the digital signal processor; Step S1065: Compile and run the code for testing; Step S1066: Download the code that has passed the test to the test device.

[0053] Specifically, in step S1061, the intelligent agent of the intelligent optimization server receives the semi-physical verification results and optimization suggestions sent by the semi-physical intelligent server; at the same time, it retrieves historical scientific research data and past strategy optimization experience from the storage module to provide data support for further optimization.

[0054] In step S1062, the intelligent agent of the intelligent optimization server combines the semi-physical verification results, optimization suggestions, historical data, and real-time user feedback to formulate a further optimization plan; real-time user feedback includes supplementary requirements from users for strategy optimization; the optimization plan must clearly define the optimization focus, optimization steps, and expected goals; the optimization focus includes parameter adjustment, logic optimization, etc.; ensuring that the optimization plan aligns with the defects found in the semi-physical verification.

[0055] In step S1063, the intelligent optimization agent and model reasoning agent of the intelligent optimization server, the digital twin agent of the digital twin intelligent server, and the analysis agent of the semi-physical intelligent server work together. The intelligent optimization agent is responsible for parameter adjustment and logic optimization; the model reasoning agent is responsible for verifying the rationality of the optimized strategy; the digital twin agent is responsible for rapid digital twin verification of the optimized strategy; and the analysis agent is responsible for providing optimization references based on semi-physical verification experience. Through the collaboration of multiple agents, the parameters and structure of the initial optimization strategy are corrected, and finally, a final optimization strategy that meets the experimental and optimization objectives and is adapted to the semi-physical environment is formed.

[0056] In step S1064, the code generation agent in the intelligent optimization server converts the code into code adapted to the digital signal processor real-time control unit according to the logic and parameters of the final optimization strategy; the code format and syntax meet the operating requirements of the digital signal processor control unit, ensuring that the code can be correctly parsed and executed by the digital signal processor control unit.

[0057] In step S1065, the code generation agent compiles the converted code to generate an executable file. Simultaneously, a runtime test is performed to simulate the code's execution within the digital signal processor control unit. The test includes checking for syntax errors, logical flaws, and whether the code's response speed and stability during execution meet requirements. If the test fails, the process returns to the code generation stage for correction until the test passes.

[0058] In step S1066, the code that has passed the test is downloaded to the real-time control unit of the digital signal processor of the test device through the communication module, completing the code deployment and preparing for the actual operation of the test device in the future.

[0059] In some embodiments, in step S106, the digital signal processor real-time control unit executes control code, specifically including: Step S1067: Receive the compiled control code; Step S1068: Implement low-latency real-time control of the test device based on control code instructions; Step S1069: Real-time feedback of control execution status data to the data acquisition agent in the semi-physical intelligent server.

[0060] Specifically, in step S1067, the digital signal processor real-time control unit receives the compiled control code downloaded from the intelligent optimization server, parses the code, clarifies the control instructions, parameter settings and execution logic in the code, and prepares for control execution.

[0061] In step S1068, the digital signal processor real-time control unit implements low-latency real-time control of various components of the test device based on the control instructions in the code. The various components of the test device include new energy power generation simulation modules, grid load modules, etc., to ensure that the control instructions can respond quickly, the control accuracy meets the experimental requirements, and conforms to the control needs of real new energy power research scenarios.

[0062] In step S1069, the digital signal processor real-time control unit collects its own operating status data and the response data of the test device in real time during the execution of control commands; it feeds back the collected data to the data acquisition agent of the semi-physical intelligent server to provide real-time data support for subsequent data acquisition, analysis and asset accumulation; at the same time, it forms a closed loop of code execution and status feedback, which makes it easier for multiple agents to discover problems in the control process in a timely manner.

[0063] In some embodiments, step S110, which implements multi-agent data sharing and asset accumulation, specifically includes: Step S1101: The data acquisition agent in the semi-physical intelligent server collects control input data, control output data, operating status data, fault data, timing data and experimental result data during the operation of the test device; Step S1102: Transmit the collected data to the intelligent optimization server; Step S1103: The storage agent in the intelligent optimization server performs centralized classification and storage of the transmitted data; Step S1104: The analytical agent in the intelligent optimization server extracts features from the stored data; Step S1105: Synchronously share the extracted feature data to each intelligent agent to form a scientific research asset that can be used for subsequent strategy optimization.

[0064] Specifically, in step S1101, the data acquisition agent comprehensively collects various types of data during the operation of the testing device; control input data includes control commands and parameter settings output by the digital signal processor real-time control unit; control output data includes the response results and execution status of the testing device; operating status data includes the operating parameters of each component of the testing device and the working status of the digital signal processor control unit; fault data includes the fault type, fault time, and fault cause that occur during the test; time-series data includes curves of various data changing over time; experimental result data includes the final result after the test is completed and the deviation from the experimental target, etc.; ensuring that the data covers the entire testing process.

[0065] In step S1102, the data acquisition agent transmits the collected data to the intelligent optimization server in real time through a high-speed communication module; during the transmission process, data encryption is used to ensure the security and integrity of the data and to prevent data loss or tampering.

[0066] In step S1103, after receiving the transmitted data, the storage agent in the intelligent optimization server centrally classifies and stores the data according to dimensions such as data type, research scenario, and experimental time. Data types include control data, operational data, and fault data. Through classified storage, a standardized data storage system is established, facilitating subsequent data query, retrieval, and analysis.

[0067] In step S1104, the analytical agent in the intelligent optimization server extracts features from the stored data. The extraction method is to use a data mining algorithm to extract the core features from the data. The core features include key parameters of strategy optimization, core indicators of experimental results, and patterns of fault occurrence. After extraction, feature data is formed, data redundancy is removed, and the reuse value of the data is improved.

[0068] In step S1105, the analytical agent synchronously shares the extracted feature data with each agent in the intelligent optimization server, digital twin intelligent server, and semi-physical intelligent server. After sharing, a research asset is formed that can be used for subsequent strategy optimization, digital twin verification, and semi-physical testing. Subsequent research tasks can directly call upon this asset to reduce repetitive research work and improve research efficiency.

[0069] In some embodiments, step S111 involves unified scheduling and visual monitoring achieved through an intelligent agent in the industrial touch and integrated management system, specifically including: Step S1111: Establish communication connections between the intelligent agent of the industrial touch control and integrated management system and the intelligent optimization server, digital twin intelligent server, semi-physical intelligent server, digital signal processor real-time control unit and intelligent printed circuit board visual analysis intelligent agent. Step S1112: Real-time scheduling of the operating status and work progress of each intelligent agent; Step S1113: Coordinate the rhythm of multi-agent collaboration; Step S1114: Present the experimental plan, running process, test data and analysis results through the human-computer interaction interface.

[0070] Specifically, in step S1111, after the intelligent agent of the industrial touch control and integrated management system is started, it establishes a stable communication connection with the intelligent optimization server, digital twin intelligent server, semi-physical intelligent server, digital signal processor real-time control unit and intelligent printed circuit board visual analysis intelligent agent through Ethernet, wireless communication and other means; by establishing the connection, a unified communication network is built to ensure smooth information transmission between various devices and intelligent agents.

[0071] In step S1112, the industrial touch control and integrated management system's intelligent agents collect the real-time operating status and work progress of each intelligent agent. The operating status includes whether each intelligent agent is operating normally and whether any faults have occurred. The work progress includes strategy construction progress, digital twin verification progress, code download progress, etc.; warning signals are issued to intelligent agents with abnormal operation, and the lagging links are scheduled to ensure the orderly progress of the entire process.

[0072] In step S1113, the intelligent agent of the industrial touch control and integrated management system coordinates the collaborative rhythm of multiple agents according to the sequence of the research process and the time consumption of each step; for example, after the digital twin verification is completed, the intelligent optimization server is promptly scheduled to start iterative optimization. After the semi-physical verification is completed, the intelligent optimization server is scheduled to further optimize the strategy; by coordinating the rhythm, problems such as process bottlenecks and excessive waiting time are avoided, thereby improving the overall collaborative efficiency.

[0073] In step S1114, the industrial touch and integrated management system visualizes the experimental plan, operation process, test data, and analysis results through a human-machine interface. The experimental plan includes the initial strategy and the final optimization strategy; the operation process includes the progress of each stage and the remaining time; the test data includes digital twin operation data and semi-physical verification data; and the analysis results include defect identification results and optimization suggestions. The presentation uses intuitive forms such as charts and curves, making it easy for users to view and grasp the research progress in real time. At the same time, it supports users to input adjustment opinions, realize human-machine interaction, and improve the controllability of the research process.

[0074] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A multi-agent collaborative intelligent optimization method for the entire process of new energy power research, characterized in that, include: Receive user-submitted research and control requirements, experimental objectives, and optimization goals for new energy power; An initial strategy scheme is constructed by multiple intelligent agents in the intelligent optimization server; The initial strategy plan is sent to multiple agents in the digital twin intelligent server, and multi-agent collaborative verification is performed in the digital twin environment; Based on user interaction feedback and verification results, the initial strategy scheme is revised in multiple rounds through a multi-agent iterative optimization mechanism to output a preliminary optimized strategy. The initial optimization strategy and test data are sent to multiple agents in the semi-physical intelligent server. Real-time closed-loop verification of the multi-agent system is performed in the semi-physical environment, and the semi-physical verification results and optimization suggestions are output. Based on the semi-physical verification results and optimization suggestions, the preliminary optimization strategy is further optimized through the collaboration of multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server. The optimized strategy is then converted into code that adapts to the real-time control unit of the digital signal processor. The code is compiled, and the compiled code is downloaded to the test device; The data acquisition agent in the semi-physical intelligent server collects real-time data during the operation of the test device. Analyze real-time data and feed the analysis results and test data back to the intelligent optimization server to achieve data sharing and asset accumulation among multiple agents; Through the industrial touch control and integrated management system, each intelligent agent is uniformly scheduled, and the entire process operation status is visualized and monitored.

2. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, An initial policy scheme is constructed by multiple intelligent agents in an intelligent optimization server, specifically including: Read historical scientific research experimental data, existing strategy results, and user requirement parameters; The model-trained agent is used to train an optimized model adapted to new energy power research scenarios based on historical scientific research experimental data; The model-inference agent, in conjunction with user requirement parameters, invokes the optimized model to perform policy inference. The intelligent optimization agent makes initial adjustments to the policy reasoning results; Output an initial strategy solution that meets user needs and is compatible with subsequent digital twin and verification processes.

3. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, The initial policy scheme is revised multiple times through a multi-agent iterative optimization mechanism to output a preliminary optimized policy, which includes: The training environment and verification environment for digital twin intelligent agents are located in the digital twin intelligent server. Import the initial strategy scheme into the digital twin verification environment; The digital twin agent performs strategy verification and simulates the dynamic changes of real scientific research scenarios. The intelligent agent, constructed from the training environment, dynamically adjusts its training parameters based on the digital twin verification results. Multi-agent collaborative analysis of digital twin verification data identifies invalid parameter combinations, invalid control paths, and erroneous policy directions; Based on user interaction feedback, the control parameters, model structure, constraints, and objective function were adjusted in multiple rounds. Select preliminary optimization strategies that align with user needs.

4. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, Perform real-time closed-loop verification of multi-agent systems in a semi-physical environment, and output semi-physical verification results and optimization suggestions, specifically including: The verification agent, data acquisition agent, and analysis agent in the semi-physical intelligent server work together. Couple the initial optimization strategy with the actual control hardware and the object under test; Real-time closed-loop verification is performed by the verification agent; The data acquisition agent synchronously collects strategy response performance, stability indicators, dynamic process data, and error data. The analytical agent performs in-depth analysis of the collected data to identify the adaptation defects of the strategy in the semi-physical environment; Generate optimization suggestions that adapt to the defects of the semi-physical environment, and output the semi-physical verification results.

5. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, Multi-agent collaboration further optimizes the initial optimization strategy, converting the optimized strategy into code, specifically including: The intelligent optimization server agent receives the semi-physical verification results and optimization suggestions. Based on historical data and real-time feedback, develop further optimization plans; Multiple intelligent agents in the intelligent optimization server, digital twin intelligent server and semi-physical intelligent server work together to modify the parameters and structure of the initial optimization strategy and form the final optimization strategy. The code-generated agent translates the final optimization strategy into code that is adapted to the real-time control unit of the digital signal processor. Compile and run the code; Download the code that passes the test to the test device.

6. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, Achieving multi-agent data sharing and asset accumulation specifically includes: The data acquisition agent in the semi-physical intelligent server collects control input data, control output data, operating status data, fault data, timing data, and experimental result data during the operation of the test device; The collected data is transmitted to the intelligent optimization server; The intelligent storage agent in the intelligent optimization server centrally classifies and stores the transmitted data; The analytical agent in the intelligent optimization server extracts features from the stored data; The extracted feature data is synchronously shared with each intelligent agent, forming a research asset that can be used for subsequent strategy optimization.

7. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 1, characterized in that, Unified scheduling and visual monitoring are achieved through intelligent agents in industrial touch control and integrated management systems, specifically including: Establish communication connections between the intelligent agents of the industrial touch control and integrated management system and the intelligent optimization server, digital twin intelligent server, semi-physical intelligent server, digital signal processor real-time control unit and intelligent printed circuit board visual analysis intelligent agent; Real-time scheduling of the operational status and work progress of each intelligent agent; Coordinate the rhythm of multi-agent collaboration; The experimental plan, running process, test data and analysis results are presented through a human-computer interaction interface.

8. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 4, characterized in that, The semi-physical real-time closed-loop verification process also includes: The intelligent printed circuit board visual analysis agent acquires images of the digital signal processor and real-time control unit of the printed circuit board and the testing device in real time. Perform defect detection, device identification, and experimental status observation on the acquired images; Output board-level defect information, device identification results, and experimental observation images; The output results are fed back to the analytical agent in the semi-physical intelligent server.

9. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 3, characterized in that, The multi-round iterative optimization process also includes: When no strategy solution that meets the preset evaluation indicators is obtained, multiple agents work together to summarize the generated iterative data. Analyze iterative data to identify invalid iteration paths and parameter redundancy issues; Optimize the iterative algorithm based on the recognition results; Narrow the search scope for subsequent iterations.

10. The intelligent optimization method for the entire process of multi-agent collaborative research in new energy power as described in claim 5, characterized in that, The real-time control unit of the digital signal processor executes control code, specifically including: Receive the compiled control code; Low-latency real-time control of the test device is implemented based on control code instructions; The data acquisition agent in the semi-physical intelligent server provides real-time feedback of control execution status data.

Images