Human-machine hybrid intelligent power grid regulation method and system based on plase system
By introducing dispatcher experience data and historical data into the PLASE system, and combining it with AI intelligent quantitative evaluation and autonomous evolution modules, the problems of high training cost, opaque decision-making, and insufficient security of power grid control agents have been solved, achieving efficient and reliable human-machine collaborative power grid control.
Patent Information
- Application Number
- CN202611183699.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-05
- Publication Date
- 2026-11-10
AI Technical Summary
Existing power grid control agents suffer from high training costs and uncertain cycles, opaque decision-making processes, unfriendly human-machine interaction, and difficulty in meeting security and reliability requirements. PLASE systems lack human-machine collaboration mechanisms.
By introducing dispatcher experience data and historical data to construct a tutoring dataset, the PLASE system is used to conduct intelligent evaluation and evolution of intelligent agents. Combined with the AI intelligent quantitative evaluation module and the autonomous evolution module, the dispatcher can review and guide the intelligent agents.
It significantly reduces the training cost and cycle of intelligent agents, improves the transparency and reliability of decision-making, ensures the safety and reliability of power grid regulation, and improves the channels for human-machine communication.
Smart Images

Figure CN123074101A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent dispatch and control technology for power systems, specifically a human-machine hybrid intelligent power grid control method and system based on the PLASE system. Background Technology
[0002] Power grid dispatching is a core component in ensuring the safe, high-quality, and economical operation of the power grid. With the rapid development of artificial intelligence (AI) technology, using intelligent agents to solve power grid dispatching and control problems is gradually becoming an effective method. Power grid control intelligent agents built based on AI technologies such as deep reinforcement learning (DRL) can achieve intelligent correction and control of the power grid to a certain extent.
[0003] However, the existing technology still has the following significant shortcomings:
[0004] High training costs and uncertain training time: DRL-based power grid control agents require a large amount of interactive trial and error training, which is costly and the training time is difficult to predict, thus restricting the engineering application of related technologies.
[0005] The decision-making process is opaque and the human-computer interaction is unfriendly: the decision-making logic of AI agents is usually a "black box" model, and its output actions lack interpretability, making it difficult for power dispatchers to understand and trust its decisions, and the human-computer communication channels are not smooth.
[0006] Safety and reliability requirements are difficult to meet: Power systems have extremely high requirements for safe operation. Even if existing AI agents are deployed in actual systems, they can only serve as auxiliary references and are usually not allowed to directly perform control operations, thus failing to fully realize the regulatory potential of AI.
[0007] Existing evaluation and evolution platforms lack human-machine collaboration mechanisms: Although PLASE (AI quantitative evaluation and autonomous evolution based on parallel systems) system can perform quantitative evaluation and optimization evolution of intelligent agents, in practical applications there are still problems such as opaque decision-making actions of intelligent agents, unfriendly human-machine communication channels, and difficulty in understanding and mastering actual decisions by power industry practitioners. There is a lack of a systematic platform that can effectively integrate dispatchers' experience and knowledge with AI autonomous decision-making.
[0008] To address the aforementioned issues, this invention proposes a human-machine hybrid smart grid control system based on the PLASE system. The aim is to enhance the transparency, reliability, and security of the intelligent agent's decision-making by introducing the dispatcher's experience and knowledge to manually review and intervene in the intelligent assessment and evolution process of the intelligent agent. Summary of the Invention
[0009] To overcome the shortcomings of existing power grid control agents, such as high training costs, opaque decision-making, insufficient security, and poor human-machine collaboration in existing PLASE systems, this invention provides a human-machine hybrid intelligent power grid control method and system based on the PLASE system. This enables intelligent power grid control under the guidance of dispatchers, improving the decision-making level and control effect of the agent in actual power grid scenarios.
[0010] To achieve the above-mentioned objectives, the present invention provides the following technical solution:
[0011] A human-machine hybrid smart grid control method based on PLASE system includes the following steps:
[0012] Step S1: The tutoring dataset, which is based on dispatcher experience data and historical data of the dispatching system, is used as the initial action decision of the power grid control agent and input into the PLASE system;
[0013] Step S2: Input the actual power grid operation scenario into the PLASE system;
[0014] Step S3: The PLASE system uses the input actual power grid operation scenario as the test scenario, and uses its built-in AI intelligent quantitative evaluation module to perform intelligent level quantitative evaluation on the power grid control agent, generate intelligent evaluation results, and output the intelligent evaluation results, the decision action output list of the agent in the current test scenario, and the power grid simulation operation data after the decision action. The intelligent evaluation results, the decision action output list, and the power grid simulation operation data are submitted to the dispatcher for review.
[0015] Step S4: If the dispatcher confirms after review that the intelligent evaluation results, decision action output list, and power grid simulation operation data meet its expected requirements, then the power grid control intelligent agent will be handed over to the PLASE system for execution and saved.
[0016] Step S5: If the dispatcher confirms after review that the intelligent evaluation results, decision action output list and power grid simulation operation data do not meet the expected requirements, then the autonomous evolution module built into the PLASE system is used to perform optimization evolution on the power grid control agent to generate the evolved agent.
[0017] Step S6: Using the evolved agent as the current agent to be evaluated, and the actual power grid operation scenario input in step S2 as the test scenario, the AI intelligent quantitative evaluation module is used again to perform intelligent level quantitative evaluation, generate the evolved intelligent evaluation result, and output the decision action output list of the evolved agent and the corresponding power grid simulation operation data. The evolved intelligent evaluation result, decision action output list and power grid simulation operation data are then submitted to the dispatcher for review again.
[0018] Step S7: If the scheduler confirms that the evolved agent meets its expected requirements after review, proceed to step S4 for processing; if the scheduler confirms that the evolved agent still does not meet its expected requirements after review, repeat steps S5 and S6 until the current agent meets the scheduler's expected requirements.
[0019] Furthermore, the intelligent evaluation results in step S3 include at least the intelligent quantitative scoring results of the machine intelligence level of the power grid control intelligent agent.
[0020] Furthermore, the decision action output list in step S3 includes a set of controllable device control instructions output by the power grid control agent for the current actual operation scenario of the power grid.
[0021] Furthermore, the optimization-oriented evolution in step S5 refers to the autonomous evolution module iteratively updating the neural network structure and parameters of the power grid control agent with the goal of optimizing the defects pointed out by the dispatcher during the review.
[0022] A human-machine hybrid smart grid control system based on PLASE includes:
[0023] The PLASE system has a built-in AI intelligent quantitative evaluation module and an autonomous evolution module, which are used to quantitatively evaluate the intelligence level of the power grid control agent and promote its optimization.
[0024] A human-computer interaction interface is used to enable information exchange between the dispatcher and the PLASE system;
[0025] The dispatcher's workbench allows dispatchers to view approval information and input approval instructions.
[0026] The PLASE system receives the input tutoring dataset and the actual operation scenario of the power grid. Using the actual operation scenario as a test scenario, it calls the AI intelligent quantitative evaluation module to quantitatively evaluate the intelligence level of the power grid control agent. The evaluation results and agent decision information are then output to the dispatcher's workbench through the human-computer interaction interface. Based on the review instructions input by the dispatcher's workbench, the PLASE system selectively calls the autonomous evolution module to perform optimization evolution on the power grid control agent, or saves the agents that meet the requirements.
[0027] Furthermore, the tutoring dataset is established based on dispatcher experience data and historical data of the dispatching system, and is used as the initial action decision input for the power grid control agent.
[0028] Furthermore, the evaluation results include the intelligent evaluation results generated by the AI intelligent quantitative evaluation module, the decision action output list of the power grid control intelligent agent, and the power grid simulation operation data after the decision action.
[0029] Furthermore, the review instructions include: an instruction to instruct the PLASE system to save the current agent's confirmation, and a "negation / evolution" instruction to instruct the PLASE system to call the autonomous evolution module to perform optimization evolution on the current agent.
[0030] Furthermore, the human-computer interaction interface includes at least a result display submodule, an instruction input submodule, and a dialogue communication submodule. The result display submodule is used to present the evaluation results and agent decision information to the dispatcher. The instruction input submodule is used to receive the review instructions input by the dispatcher. The dialogue communication submodule is used to realize text or voice interaction between the dispatcher and the PLASE system.
[0031] Furthermore, after receiving the "negation / evolution" command from the dispatcher's workbench, the PLASE system automatically uses the current actual operation scenario of the power grid as the test scenario, drives the autonomous evolution module to perform optimization evolution on the power grid control agent, and automatically triggers the AI intelligent quantitative evaluation module to re-evaluate after the evolution is completed, until the dispatcher's expected requirements are met.
[0032] The present invention has the following beneficial effects:
[0033] 1. By introducing scheduler experience data and historical data to construct a tutoring dataset, a high-quality initial decision-making benchmark is provided for the agent, reducing the cost of blind trial and error and improving training efficiency.
[0034] 2. By embedding the dispatcher review process into the closed-loop process of agent evaluation and evolution, the agent's decision-making actions continuously improve under the supervision and guidance of the dispatcher, significantly enhancing the transparency of decision-making.
[0035] 3. Through the iterative mechanism of human-machine hybrid intelligence, the rapid computing power of AI and the experience judgment of human experts are taken into account, which effectively ensures the safety and reliability of power grid control decisions and reduces the risks of practical application.
[0036] 4. This invention relies on the quantitative evaluation and autonomous evolution capabilities of the PLASE system to provide dispatchers with objective and visual evaluation basis, improve the human-machine communication channel, and make the evolution direction of the intelligent agent more in line with the actual power grid operation needs. Attached Figure Description
[0037] Figure 1 This is a flowchart of the human-machine hybrid smart grid control method based on the PLASE system of the present invention.
[0038] Figure 2 This is a structural block diagram of the human-machine hybrid smart grid control system based on the PLASE system of the present invention. Detailed Implementation
[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0040] Example 1: Human-Machine Hybrid Smart Grid Control Method Based on PLASE System
[0041] This embodiment uses the correction and control task of a regional power grid as an example to illustrate the specific implementation process of the method described in this invention. This regional power grid includes 10 adjustable generators, 39 electrical loads, and 39 220kV transmission lines. The power grid dispatch center is equipped with a SCADA / EMS system and a senior dispatcher. Because the adjustable generators are installed on busbars #30 to #39, they are referred to as generators #30 to #39. Figure 1 As shown, the method includes the following steps:
[0042] (I) Initial preparation stage – corresponding to step S1
[0043] First, historical operating data of the regional power grid for the past three years and control operation logs from three senior dispatchers were collected. From this data, optimal or suboptimal control actions and modes for different operating conditions (such as heavy load, light load, and post-N-1 fault conditions) were extracted to form a guidance dataset. This dataset is stored in the form of "state-action" pairs. For example, when the power flow of a certain line reaches 95% of its rated value and the reserve capacity is insufficient, the corresponding guidance action is "increase the active power output of generator #30 by 5MW and simultaneously activate capacitor bank #1".
[0044] In this embodiment, the power grid control agent to be trained adopts a reinforcement learning model based on a competitive deep Q-neural network (DDQN) architecture. Its policy network is a three-layer fully connected neural network (the input layer dimension is the number of power grid state features, the hidden layer has 256 neurons, and the output layer dimension is the number of adjustable devices). The aforementioned tutoring dataset is input into the PLASE system, and the "state-action" pairs are used as initial training samples. The "agent construction and training" module of the PLASE system is used to perform supervised learning pre-training on the agent's policy network, so that the agent's initial decision policy is biased towards the dispatcher's experience. After the pre-training is completed, the "agent construction and training" module of the PLASE system will generate and initially save the network model parameters of the agent.
[0045] (II) Test Scenario Input Phase – Corresponding Step S2
[0046] The system acquires real-time data on the actual operation of the power grid through a Supervisory Control and Data Acquisition (SCADA) system. This data includes 201 dimensions of state characteristics, such as the voltage amplitude (per unit value) of each bus, the output of each generator, and the active / reactive power of each load. The PLASE system receives this real-time data through a dedicated data interface and converts it into a state vector format that the intelligent agent can recognize.
[0047] Meanwhile, if the current power grid is operating normally, the dispatcher can set a predictive "hypothetical scenario" (e.g., the temperature is expected to rise by 5°C at 14:00 in the afternoon, and the air conditioning load will increase by 10%) based on the weather forecast and load prediction results, and input it into the PLASE system as an additional test scenario.
[0048] (III) Intelligent Quantitative Evaluation Stage – Corresponding Step S3
[0049] After receiving a scenario depicting the actual operation of the power grid, the PLASE system invokes its built-in AI intelligent quantitative evaluation module. This module, based on parallel system theory, constructs a simulated virtual power system model in digital space that is equivalent to the real power grid. The specific operation is as follows:
[0050] Simulation and deduction: The current agent is placed in the simulated virtual power system model and asked to make decisions based on the current power grid operating state. For example, the agent outputs decisions such as: increasing the output of generator #30 by 3.2MW and decreasing the output of generator #32 by 1.8MW.
[0051] Simulation operation: Apply the above decision actions to power grid simulation software such as PSASP to perform transient or steady-state power flow calculations, and obtain the power grid simulation operation data after the decision actions, including: adjusted node voltages, power flow of each line, system frequency deviation, network loss rate, etc.
[0052] Intelligent Quantitative Scoring: The PLASE system's AI intelligent quantitative evaluation module is used to quantitatively score the intelligence level of the current intelligent agent. In the intelligent quantitative scoring process, firstly, appropriate intelligent indicators to be tested are selected based on factors such as the power grid dispatch control tasks performed by the current intelligent agent and the dispatcher's requirements and expectations for the agent. Then, using appropriate international power grid operation criteria, the simulated power grid operation data obtained from the simulation operation is mapped to the test scores of each intelligent indicator to be tested. Finally, the test scores of each intelligent indicator to be tested are combined to obtain the intelligent quantitative evaluation result of the current intelligent agent. For example, in this instance, for the intelligent agent performing the power grid correction control task, "safe and stable operation" and "emergency response capability" are selected as the intelligent indicators to be tested. The corresponding evaluation results are: a score of 57.21 for the safe and stable operation intelligent indicator, a score of 57.24 for the emergency response capability intelligent indicator, and an intelligent quantitative score of 53.61 for the current intelligent agent.
[0053] Subsequently, the PLASE system displays the aforementioned intelligent evaluation results, a complete list of decision-making actions (including the adjustment amounts of all controllable devices), and power grid simulation operation data (in the form of visual power flow diagrams and tables) on the dispatcher's workbench display screen through the "Result Display Submodule" of the human-computer interaction interface.
[0054] (iv) Dispatcher review and decision-making branch stage – corresponding to steps S4 and S5
[0055] The dispatcher reviews the above information at the workstation and makes a comprehensive judgment based on their own operational experience:
[0056] Scenario 1 (Requirements Met) — Step S4: If the dispatcher believes the agent's decision aligns with their expectations (e.g., the dispatcher originally planned to balance the power flow by adjusting generators #30 and #32), and all safety constraints are met, then click the "Confirm Pass" button on the workbench. This instruction is transmitted to the PLASE system through the "Instruction Input Submodule" of the human-machine interface. The PLASE system then stores the agent's network model parameters in the model library and marks it as "deployable." Subsequently, the system can periodically deploy the agent for online decision support.
[0057] Scenario 2 (Requirements Not Met) – Step S5: If the dispatcher finds that the agent's decision does not meet their expectations or requirements—for example, although the decision maintains the voltage within limits, the adjusted tap changer operation of transformer #30 causes reactive power backflow from adjacent substations, increasing network losses—and the dispatcher believes there should be a more economical adjustment scheme (e.g., "the reactive power output of generator #35 should be adjusted first"), the dispatcher notes the reason for rejection in the workbench comment box (e.g., "improper tap changer adjustment; generator reactive power regulation should be prioritized") and clicks the "Reject / Evolve" button. Upon receiving this rejection / evolution instruction, the PLASE system calls its built-in autonomous evolution module.
[0058] This autonomous evolution module employs a Bayesian optimization-based evolutionary algorithm. It uses the current agent's policy network parameters as the initial unit, and the dispatcher's specified decision actions (such as "reducing network losses" and "reducing the number of tap changer actions"), along with improving the agent's current intelligence level, as the optimization and evolution goals. All safety constraints are kept hard constraints. In the PLASE system's autonomous evolutionary model, an intelligent evaluation is performed after each iteration to autonomously adjust the agent's evolutionary direction. The evolutionary module performs 64 iterations in a simulated virtual power grid environment, searching for better policy parameter combinations using a Gaussian process based on the Bayesian optimization algorithm. By comparing the intelligence quantification scores of the agents generated in each iteration, it autonomously adjusts the evolutionary direction to ensure that the evolved agents have a higher intelligence level and more suitable power grid control strategies that better meet the dispatcher's expectations and requirements. Finally, the evolved agent is generated, and its policy network parameters are updated.
[0059] (v) Post-evolution evaluation and iterative phase – corresponding to steps S6 and S7
[0060] After the evolution is completed, the PLASE system automatically uses the evolved agent as the current agent to be evaluated, and uses the same actual power grid operation scenario input in step S2 as the test scenario to repeatedly call the AI intelligent quantitative evaluation module for a new round of evaluation.
[0061] In this embodiment, the evolved agent, in the new evaluation, makes the following decision: adjust voltage and power flow solely by increasing the active power output of generator #32 by 3.0MW and the reactive power output of generator #35 by 5MVar, without changing any transformer taps. The new agent's intelligence quantification score is improved to 61.55 points, with the intelligent indicator score for safe and stable operation increasing from 57.21 points to 73.08 points, and the intelligent indicator score for emergency response capability increasing from 57.24 points to 73.13 points.
[0062] The new evaluation results, decision action list, and simulation data are once again presented to the dispatcher through the human-computer interaction interface. After reviewing the plan, the dispatcher confirms that it meets the safety requirements, has a higher level of intelligence, and is easy to operate. The dispatcher then clicks "Confirm Pass," and the system proceeds to step S4 to save the evolved intelligent agent.
[0063] If the dispatcher is still not satisfied (for example, believing that the adjusted voltage distribution still has room for optimization), steps S5 and S6 are repeated, that is, the autonomous evolution module is triggered again to carry out a new round of evolution, and a re-evaluation is performed after the evolution until the dispatcher is satisfied. In this embodiment, the dispatcher's requirements are met after one round of evolution.
[0064] Example 2: Human-Machine Hybrid Smart Grid Control System Based on PLASE System
[0065] like Figure 2 As shown, this embodiment describes in detail the specific hardware deployment and software module composition of the system of the present invention.
[0066] (I) Overall System Architecture
[0067] The system is deployed on a server cluster in the power grid dispatch center and includes the following physical devices:
[0068] The PLASE system utilizes dual Xeon Gold 6248 processors, 256GB of DDR4 memory, and dual NVIDIA A100 GPUs. It runs on Linux but can also run on Windows. The server houses the PLASE system's core software, including an AI-powered intelligent quantitative evaluation module and an autonomous evolution module.
[0069] AI Intelligent Quantitative Evaluation Module: Developed based on Python 3.8, it integrates power grid simulation calculation engines such as PSASP (based on an improved Newton-Raphson power flow algorithm) and is responsible for multi-dimensional scoring of the intelligence level of intelligent agents.
[0070] Autonomous Evolution Module: Developed based on PyTorch 2.4, it has a built-in evolutionary algorithm library (including Bayesian optimization algorithm, BOHB, etc.) and can drive the agent's policy parameters to update according to the optimization goal specified by the scheduler and the agent's intelligence level improvement goal.
[0071] Human-computer interface: Utilizing a web server architecture, developed based on the Vue.js front-end framework and the Flask back-end framework, and connected to the PLASE system via gigabit Ethernet. This human-computer interface includes:
[0072] The results display submodule uses the ECharts chart library to dynamically display intelligent scoring radar charts, decision action comparison tables and corresponding bar charts, power grid flow simulation diagrams (displaying the voltage of each node and the line load rate in the form of a single-line diagram), and power grid operation status data tables.
[0073] The instruction input submodule provides buttons such as "Confirm Pass", "Negative / Evolution", "Manually Specify Action for Fine-tuning", and "Print Results", as well as option drop-down boxes and text comment input boxes, allowing dispatchers to provide detailed suggestions for improving control for different power equipment.
[0074] Dialogue Communication Submodule: Integrates a lightweight dialogue engine based on Natural Language Processing (NLP), which allows dispatchers to input commands via voice or text (such as "Please adjust the output of generator #30 to 5MW"), and the system automatically parses and converts them into corresponding parameter constraints.
[0075] Dispatcher workstation: Equipped with four 27-inch 4K high-definition touchscreen displays, a professional-grade industrial control computer, and human-machine interface peripherals (keyboard, mouse, microphone array), connected to the human-machine interface via an encrypted network. The workstation features a two-factor authentication mechanism to ensure secure access permissions.
[0076] (II) System Workflow
[0077] The actual operation flow of the system is completely consistent with the method steps described in Example 1, and its data flow and control flow relationship is as follows:
[0078] Data Input Phase: The dispatcher uploads a pre-processed tutor dataset file (in JSON or NPY format) via the command input submodule on the workbench. The PLASE system receives and loads this dataset. Simultaneously, the PLASE system acquires real-time data on the actual operation of the power grid from the SCADA system using the standard IEC104 protocol.
[0079] Evaluation Execution Phase: The PLASE system uses the actual operation scenario of the power grid as the test scenario and calls the AI intelligent quantitative evaluation module to perform quantitative evaluation on the current intelligent agent. The evaluation calculation process is executed in parallel on the GPU, and a single evaluation takes about 30 minutes.
[0080] Results feedback phase: Evaluation results (including intelligent quantitative scoring, decision action list, power grid simulation operation data, etc.) are pushed to the dispatcher's workbench through the human-computer interaction interface and presented in a large-screen visualization.
[0081] Dispatcher review stage: After reviewing the information on the workbench, the dispatcher clicks the corresponding button to generate a review instruction. If the instruction is "Confirm Pass", the PLASE system saves the agent; if the instruction is "Reject / Evolve", the PLASE system automatically starts the autonomous evolution module and automatically triggers a new round of evaluation after the evolution is completed, without the need for manual restart of the process (corresponding to the automatic loop mechanism of claim 10).
[0082] Loop Iteration Control: The PLASE system has an internal loop counter with a default maximum of 64 iterations. Exceeding this limit will forcibly terminate the loop and trigger an alarm, preventing infinite loops. The agent parameters, evaluation results, and scheduler comments for each iteration are recorded in the system log database for subsequent analysis and auditing.
[0083] In the actual test of this embodiment, the system of the present invention was used to train and optimize a power grid control agent. The test results show:
[0084] Compared with pure AI autonomous training, this invention reduces the number of iterations required for the agent to reach the scheduler's satisfaction level from an average of 1,000 to less than 100, shortening the training time by about 90%.
[0085] The dispatcher's satisfaction score (out of 10) for the agent's decision-making improved from 3.2 to 8.7.
[0086] All intelligent agents deployed by this platform did not violate any power grid safety constraints during actual online operation, verifying the effectiveness of the human-machine hybrid audit mechanism.
[0087] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0088] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A human-machine hybrid smart grid control method based on PLASE system, characterized in that, Includes the following steps: Step S1: The tutoring dataset, which is based on dispatcher experience data and historical data of the dispatching system, is used as the initial action decision of the power grid control agent and input into the PLASE system. Step S2: Input the actual power grid operation scenario into the PLASE system; Step S3: The PLASE system uses the input actual power grid operation scenario as the test scenario, and uses its built-in AI intelligent quantitative evaluation module to perform intelligent level quantitative evaluation on the power grid control agent, generate intelligent evaluation results, and output the intelligent evaluation results, the decision action output list of the agent in the current test scenario, and the power grid simulation operation data after the decision action. The intelligent evaluation results, the decision action output list, and the power grid simulation operation data are submitted to the dispatcher for review. Step S4: If the dispatcher confirms after review that the intelligent evaluation results, decision action output list, and power grid simulation operation data meet its expected requirements, then the power grid control intelligent agent will be handed over to the PLASE system for execution and saved. Step S5: If the dispatcher confirms after review that the intelligent evaluation results, decision action output list and power grid simulation operation data do not meet the expected requirements, then the autonomous evolution module built into the PLASE system is used to perform optimization evolution on the power grid control agent to generate the evolved agent. Step S6: Using the evolved agent as the current agent to be evaluated, and the actual power grid operation scenario input in step S2 as the test scenario, the AI intelligent quantitative evaluation module is used again to perform intelligent level quantitative evaluation, generate the evolved intelligent evaluation result, and output the decision action output list of the evolved agent and the corresponding power grid simulation operation data. The evolved intelligent evaluation result, decision action output list and power grid simulation operation data are then submitted to the dispatcher for review again. Step S7: If the scheduler confirms that the evolved agent meets its expected requirements after review, proceed to step S4 for processing; if the scheduler confirms that the evolved agent still does not meet its expected requirements after review, repeat steps S5 and S6 until the current agent meets the scheduler's expected requirements.
2. The method according to claim 1, characterized in that, The intelligent evaluation results in step S3 include at least the intelligent quantitative scoring results of the machine intelligence level of the power grid control intelligent agent.
3. The method according to claim 1, characterized in that, The decision action output list in step S3 includes a set of control instructions for each controllable device output by the power grid control agent in response to the current actual operation scenario of the power grid.
4. The method according to claim 1, characterized in that, The optimization-oriented evolution in step S5 refers to the autonomous evolution module iteratively updating the neural network structure and parameters of the power grid control agent with the goal of optimizing the defects pointed out by the dispatcher during the review.
5. A human-machine hybrid smart grid control system based on PLASE system, characterized in that, include: The PLASE system has a built-in AI intelligent quantitative evaluation module and an autonomous evolution module, which are used to quantitatively evaluate the intelligence level of the power grid control agent and promote its optimization. A human-computer interaction interface is used to enable information exchange between the dispatcher and the PLASE system; The dispatcher's workbench allows dispatchers to view approval information and input approval instructions. The PLASE system receives the input tutoring dataset and the actual operation scenario of the power grid, uses the actual operation scenario of the power grid as the test scenario, calls the AI intelligent quantitative evaluation module to quantitatively evaluate the intelligence level of the power grid control agent, and outputs the evaluation results and agent decision information to the dispatcher's workbench through the human-computer interaction interface. The PLASE system selectively invokes the autonomous evolution module to optimize the power grid control agent based on the review instructions input by the dispatcher's workbench, or saves the agents that meet the requirements.
6. The system according to claim 5, characterized in that, The tutoring dataset is built based on dispatcher experience data and historical data of the dispatching system, and is used as the initial action decision input for the power grid control agent.
7. The system according to claim 5, characterized in that, The evaluation results include the intelligent evaluation results generated by the AI intelligent quantitative evaluation module, the decision action output list of the power grid control intelligent agent, and the power grid simulation operation data after the decision action.
8. The system according to claim 5, characterized in that, The review instructions include: an instruction to instruct the PLASE system to save the current agent's confirmation, and a "negation / evolution" instruction to instruct the PLASE system to call the autonomous evolution module to perform optimization evolution on the current agent.
9. The system according to claim 5, characterized in that, The human-computer interaction interface includes at least a result display submodule, an instruction input submodule, and a dialogue communication submodule. The result display submodule is used to present the evaluation results and agent decision information to the dispatcher. The instruction input submodule is used to receive the review instructions input by the dispatcher. The dialogue communication submodule is used to realize text or voice interaction between the dispatcher and the PLASE system.
10. The system according to claim 5, characterized in that, After receiving the "negation / evolution" command from the dispatcher's workbench, the PLASE system automatically uses the current actual operation scenario of the power grid as the test scenario, drives the autonomous evolution module to perform optimization evolution on the power grid control agent, and automatically triggers the AI intelligent quantitative evaluation module to re-evaluate after the evolution is completed, until the dispatcher's expected requirements are met.
