Switching operation path optimization method and system based on artificial intelligence
By using an AI-based switching operation path optimization method, dynamic modeling and deep reinforcement learning are employed to generate optimized power grid paths. This solves the problems of reliance on human experience and high computational complexity in existing technologies, and achieves efficient and safe power grid operation path planning.
Patent Information
- Application Number
- CN202511707158.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-02-17
AI Technical Summary
Existing switching operation paths rely on manual experience, making it difficult to cope with complex abnormal operating conditions. They also have high computational complexity and lack real-time solutions. Safety verification is done after the fact, resulting in low efficiency.
An artificial intelligence-based approach is adopted, which involves dynamic modeling, task parsing, intelligent path generation, verification and display, and instruction execution. It utilizes deep reinforcement learning and graph neural networks to generate optimized paths, and combines them with anti-misoperation rules for automatic verification and visual confirmation.
It enables the generation of multiple optimized paths that meet safety constraints at the minute or second level, reducing the risk of human error and improving operational efficiency and safety, making it suitable for real-time planning of complex power grids.
Smart Images
Figure CN121546815A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of substation switching operation technology, specifically to a switching operation path optimization method and system based on artificial intelligence. Background Technology
[0002] Electrical equipment is classified into three states: operation, standby (cold standby and hot standby), and maintenance. The process of changing equipment from one state to another is called switching, and the operation performed is called switching operation. Switching operations involve changing electrical equipment from one state to another or changing the operating mode of the system by operating disconnecting switches, circuit breakers, and connecting / disconnecting grounding wires. Switching operations must be carried out in accordance with the operation ticket system and work supervision system. Switching operations are divided into: supervised operation, single-person operation, and maintenance personnel operation. Supervised operation: Two people perform the same operation. During supervised operation, one person who is more familiar with the equipment supervises. Particularly important and complex switching operations are performed by skilled operators, supervised by the shift supervisor. When operating in a substation with a single shift, the operator fills out an operation ticket according to the operation instructions transmitted by the dispatcher by telephone, and performs the operation after verifying that the instructions are correct. Equipment, projects, and operators that are operated by a single person must be approved by the equipment operation management unit, and the personnel should pass a special assessment.
[0003] However, existing switching operation paths have the following drawbacks: they mainly rely on operation experts to manually draft tickets or retrieve and modify them from a typical ticket database under static typical conditions based on procedures and experience. The entire process is time-consuming and dependent on personal experience, making it difficult to cope with complex abnormal operating conditions. Traditional mathematical programming or heuristic algorithms, such as genetic algorithms, often fail to obtain real-time solutions when solving large-scale power grid switching operations due to high computational complexity, or are prone to getting trapped in local optima. Safety verification is usually an independent and passive check after path generation, which is a post-verification mode. It is either entirely manual decision-making, which is inefficient, or a rigid fully automatic system that lacks flexibility. Therefore, there is a lack of an optimization method that can solve the problems of low efficiency, reliance on manual safety, and difficulty in handling complex operating conditions in existing technologies. Summary of the Invention
[0004] This invention aims to provide an artificial intelligence-based method and system for optimizing switching operation paths. It is mainly used to address the shortcomings of existing technologies, which rely heavily on operation experts based on procedures and experience, making it difficult to cope with complex abnormal operation modes. When solving large-scale power grid switching operations, the high computational complexity often makes it difficult to obtain real-time solutions or easily leads to local optima. Safety verification, as an independent and passive inspection step after path generation, is a technical problem of the post-verification mode.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0006] An artificial intelligence-based method for optimizing switching operation paths includes the following steps:
[0007] S1. Dynamic modeling steps: Obtain the topological connection relationship of the power grid, the real-time status of equipment and electrical parameters from the data acquisition and monitoring system in real time, and build and continuously update the dynamic power grid model for path simulation.
[0008] S2. Task parsing steps: Receive and parse the switching operation task, and clarify the initial state and target state of the operation;
[0009] S3. Intelligent path generation step: Using the dynamic power grid model as the simulation environment, a pre-trained artificial intelligence decision-making model is used to optimize the path. The model is trained through deep reinforcement learning, and its reward function integrates power grid safety constraints and operational efficiency objectives to generate candidate operation paths.
[0010] S4. Path verification and visualization steps: The generated candidate paths are automatically verified against the anti-misoperation rule base, and then visualized and confirmed through the human-computer interaction interface.
[0011] S5. Instruction execution steps: Convert the confirmed operation path into a standard instruction sequence and send it to the automated control system for execution.
[0012] An artificial intelligence-based switching operation path system includes a dynamic modeling module, which is used to acquire the topological connections of the power grid, the real-time status of equipment, and electrical parameters from the data acquisition and monitoring system in real time, and to construct and continuously update a dynamic power grid model for path simulation.
[0013] The task parsing module is used to receive and parse switching operation tasks, and to determine the initial and target states of the operation.
[0014] The intelligent decision-making module is used to use the dynamic power grid model as a simulation environment and to perform path optimization using a pre-trained artificial intelligence decision-making model. The model is trained through deep reinforcement learning, and its reward function integrates power grid safety constraints and operational efficiency objectives to generate candidate operation paths.
[0015] The verification and display module is used to automatically verify the generated candidate paths against the anti-misoperation rule base, and to visualize and confirm them through the human-computer interaction interface.
[0016] The instruction execution module is used to convert the confirmed operation path into a standard instruction sequence and send it to the automated control system for execution.
[0017] Working principle and beneficial effects of the present invention:
[0018] 1. Working Principle: The dynamic modeling step defines the input basis of the method, emphasizing that the constructed model is dynamic and real-time, ensuring that the optimization calculation is based on the current real state of the power grid, and achieving the premise of safe and feasible path optimization; the task parsing step clarifies the task input and parsing links of the method, and its role is to transform abstract operation instructions into a state space search problem with clear start and end points that can be processed by the AI model; the intelligent path generation step utilizes embedded domain knowledge, such as safety and efficiency goals, and the reinforcement learning model performs automatic optimization in the simulation environment, replacing the traditional method of relying on manual experience to traverse and verify; the path verification and visualization step defines the safety verification and human-computer interaction mechanism of the output path, and ensures the safety and reliability of the final solution while retaining the final decision-making right of the operator by adding independent automated verification and intuitive visualization; the instruction execution step describes the final output and closed-loop control of the method, transforming the logical path into executable physical control instructions, realizing a complete technical chain from intelligent decision-making to automated execution;
[0019] The dynamic modeling module establishes and maintains a high-fidelity digital twin synchronized with the physical power grid, providing an accurate simulation environment for the intelligent decision-making module. The task parsing module defines the system's task interface module, responsible for converting user instructions into accurate optimization target parameters that are identifiable within the system, connecting user intent with the core algorithm. The intelligent decision-making module defines the system's core computing module, which integrates a trained AI model and simulation environment for performing path optimization calculations and demonstrating the system's intelligence. The verification and display module defines the system's security and interaction module, undertaking the responsibility of secondary security review of AI-generated results and achieving effective collaboration with operators through a human-machine interface. The instruction execution module defines the system's control output module, safely and reliably converting the confirmed optimization scheme into standard instructions executable by the underlying automation system, completing the decision-making closed loop.
[0020] 2. Beneficial effects:
[0021] (1) By using the dynamic power grid model as the simulation environment and using the AI model with embedded power grid safety knowledge for automatic optimization, multiple optimized paths that meet safety constraints can be generated quickly, such as at the minute or even second level. This greatly shortens the preparation time before switching operations and reduces the risk of misoperation due to human negligence from the root. It is especially suitable for scenarios with high timeliness requirements, such as emergency recovery after a fault.
[0022] (2) The deep reinforcement learning model adopted can learn efficient decision-making strategies from massive interactions with the simulation environment through its powerful representation learning ability. In practical applications, the model can perform fast generalization reasoning, directly output high-quality paths, effectively overcome the combination difficulties, and realize real-time path planning in complex real power grids.
[0023] (3) During the training phase of the AI model, anti-misoperation rules, voltage stability, equipment load, and Key constraints such as safety are embedded in the reward function in the form of negative rewards. This makes the AI model inherently guided by safety at every step of the path exploration, and it tends to choose the operation sequence that fundamentally meets the safety requirements, thus making the generated path safer.
[0024] (4) By generating, verifying, visualizing and confirming the process, the rapid computing power of AI is combined with the final judgment of experts. AI is responsible for providing preliminary optimized candidate solutions and clearly showing their advantages and disadvantages. Experts make the final decision based on their extensive experience. This human-machine collaboration mode not only significantly reduces the mental burden on operators, but also ensures human control over key operations, and improves the practicality and acceptability of the system.
[0025] Preferably, the reward function At every decision moment The value of Defined by the following functions: ,in, It is a security reward component. It is a reward for operational efficiency. It is a goal-oriented reward component; , These are the weighting coefficients of each reward component; by clearly defining the reward function as consisting of safety, efficiency, and objective, and quantifying them into a computable formula, the optimization objective is transformed from an abstract concept into a concrete mathematical problem that the AI model can understand and optimize.
[0026] Preferably, the security reward component The calculation method is as follows: ,in, It is a node voltage, It is its rated voltage. It is the maximum permissible deviation voltage; It is the line current. It is its rated current; yes Verify the severity indicators of violations; , , The corresponding penalty coefficient; incorporating voltage deviation and line load. Safety involves encoding expertise from the power sector into the AI training process to ensure that the model's optimization direction aligns with the requirements for safe operation of the power grid.
[0027] Preferably, the artificial intelligence decision-making model is implemented using a graph neural network based on an attention mechanism. This network takes a power grid topology graph as input and dynamically focuses on key equipment related to the current decision by calculating the attention weights of nodes and edges in the graph. Graph neural networks are suitable for processing graph structure data such as power grid topology, and through the attention mechanism, the model can dynamically identify key equipment affected by the current operation, simulating the expert's decision-making focus process.
[0028] Preferably, the intelligent path generation step uses the Monte Carlo tree search algorithm as a planner and uses the action probability distribution output by the artificial intelligence decision model as prior knowledge to guide the search process; through the AI model, the search efficiency of the Monte Carlo tree is improved, and the combination of the generalization ability of AI and the accuracy of the search algorithm results in the generation of high-quality and stable optimized paths.
[0029] Preferably, it also includes an online learning step, in which operational data is collected after actual operation to incrementally learn the artificial intelligence decision-making model in order to optimize its decision-making performance; through the online learning step, the system can use actual operating data to continuously fine-tune the model so that it can adapt to changes in power grid structure, equipment aging, and cope with situations that did not occur during training.
[0030] Preferably, a computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the method as described in any one of claims 1 to 6.
[0031] Preferably, the intelligent decision-making module further includes a strategy network unit and a reward calculation unit. The strategy network unit is used to run an artificial intelligence decision-making model, and the reward calculation unit is used to calculate the reward value of the decision in real time according to predefined rules. Its calculation logic is defined by a reward function. This module contains two key units: a strategy network responsible for decision-making and a reward calculation unit responsible for evaluation. The reward calculation unit is directly associated with the carefully designed mathematical function in the method claims to ensure the consistency between the system implementation and the technical solution.
[0032] Preferably, it also includes an online learning module, which is used to collect operational data after actual operation to perform incremental learning on the artificial intelligence decision-making model; through the online learning module, the system is transformed from a static execution terminal into a terminal with continuous optimization capabilities. Attached Figure Description
[0033] Figure 1 This is a flowchart of an artificial intelligence-based switching operation path optimization method according to the present invention;
[0034] Figure 2 This is a module flowchart of an artificial intelligence-based switching operation path system according to the present invention;
[0035] Figure 3 This is a schematic diagram of the intelligent decision-making module of an artificial intelligence-based switching operation path system according to the present invention. Detailed Implementation
[0036] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0037] Taking the power outage and switching operation of a line in a 220kV substation as an example, Example 1: Figure 1 As shown, an artificial intelligence-based method for optimizing switching operation paths includes the following steps:
[0038] S1. Dynamic modeling steps: Obtain the topological connection relationship of the power grid, the real-time status of equipment and electrical parameters from the data acquisition and monitoring system in real time, and build and continuously update the dynamic power grid model for path simulation.
[0039] S2. Task parsing steps: Receive and parse the switching operation task, and clarify the initial state and target state of the operation;
[0040] S3. Intelligent Path Generation Steps: Using a dynamic power grid model as the simulation environment, a pre-trained artificial intelligence decision-making model is used for path optimization. This model is trained through deep reinforcement learning, and its reward function integrates power grid safety constraints and operational efficiency objectives to generate candidate operational paths. At every decision moment The value of Defined by the following functions: ,in, It is a security reward component. It is a reward for operational efficiency. It is a goal-oriented reward component; , These are the weighting coefficients for each reward component, including the security reward component. The calculation method is as follows: ,in, It is a node voltage, It is its rated voltage. It is the maximum permissible deviation voltage; It is the line current. It is its rated current; yes Verify the severity indicators of violations; , , This represents the corresponding penalty coefficient;
[0041] The artificial intelligence decision-making model is implemented using a graph neural network based on an attention mechanism. This network takes a power grid topology graph as input and dynamically focuses on key equipment relevant to the current decision by calculating the attention weights of nodes and edges in the graph.
[0042] The intelligent path generation process employs the Monte Carlo tree search algorithm as a planner and uses the action probability distribution output by the artificial intelligence decision model as prior knowledge to guide the search process.
[0043] S4. Path verification and visualization steps: The generated candidate paths are automatically verified against the anti-misoperation rule base, and then visualized and confirmed through the human-computer interaction interface.
[0044] S5. Instruction execution steps: Convert the confirmed operation path into a standard instruction sequence and send it to the automation control system for execution;
[0045] It also includes online learning steps, and after actual operation, the operation data is collected to incrementally learn the artificial intelligence decision-making model in order to optimize its decision-making performance;
[0046] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as claimed in any one of claims 1 to 6.
[0047] Example 2, as follows Figures 2-3 As shown, an AI-based switching operation path system includes a dynamic modeling module. This module acquires real-time topology connections, equipment status, and electrical parameters from a data acquisition and monitoring system to construct and continuously update a dynamic power grid model for path simulation. The system connects to the SCADA / EMS system via standard communication protocols (such as IEC 61850 and IEC 104) to periodically (e.g., once per second) acquire real-time data from the entire network. Based on this data, the system maintains a graphical power grid model in memory. This model uses substations and lines as nodes and circuit breakers and disconnectors as edges. Each node and edge is associated with current electrical parameters (such as voltage, current, and power) and status (such as closed or open). This model forms the basis for all subsequent simulations and decisions.
[0048] The task parsing module is used to receive and parse switching operation tasks, clarify the initial state and target state of the operation. The operator issues tasks through graphical drag and drop or selection of preset task templates on the human-machine interface. For example, switching 110kV line L1 from operation to maintenance. After the system receives the instruction, the task parsing module will automatically parse the initial state (the state of all switches associated with L1 at present) and the target state (the state of all switches required by the maintenance state) of the task, and convert it into a state transition problem consisting of a sequence of equipment operations.
[0049] The intelligent decision-making module uses a dynamic power grid model as a simulation environment and a pre-trained artificial intelligence decision-making model for path optimization. The model is trained through deep reinforcement learning, and its reward function integrates power grid safety constraints and operational efficiency objectives to generate candidate operation paths. The intelligent decision-making module also includes a policy network unit and a reward calculation unit. The policy network unit is used to run the artificial intelligence decision-making model, and the reward calculation unit is used to calculate the reward value of the decision in real time according to predefined rules. Its calculation logic is defined by the reward function.
[0050] Intelligent path generation: Simulation environment. The system is based on a dynamic power grid model and develops a power grid operation simulator. This simulator can simulate the changes in power grid topology and power flow after the execution of any compliant operation command (such as "disconnecting circuit breaker CB1").
[0051] AI Decision Model Training: Model Selection: In this embodiment, a graph neural network (GAT) based on attention mechanism is preferred as the policy network because it can effectively process the graph structure data of the power grid and focus on the key equipment affected by the current operation through the attention mechanism;
[0052] Training method: The model is trained using the Proximal Policy Optimization (PPO) deep reinforcement learning algorithm. The training process is carried out in a simulation environment, where the AI agent explores the path from the initial state to the target state by trying various operations.
[0053] Reward Function Design: The reward function is key to guiding AI learning, and its specific design is as follows: Reward Function At every decision moment The value of Defined by the following functions: Security reward component The calculation method is as follows: Anything that causes equipment overload, voltage exceeding limits, or damage Operating according to safety guidelines will result in significant negative rewards (penalties), while efficiency will result in rewards. : =-1 (A fixed small penalty is given for each additional step, encouraging the shortest path), target reward : =+100 (Only awarded upon successfully reaching the target state, to encourage task completion);
[0054] Path generation: For a new task, the pre-trained model is loaded. Based on the current power grid state, the model calculates the probability of each legal operation through the GAT network, and then selects the operation with the highest probability. The system sends this operation to the simulator, updates the state, and repeats this process until the target state is reached, thereby generating a complete operation path. To obtain a better solution, Monte Carlo Tree Search (MCTS) can be combined, using the model output as a prior probability to guide the search and evaluate more possibilities.
[0055] The verification and display module is used to automatically verify the generated candidate paths against the anti-misoperation rule base, and to visualize and confirm them through the human-machine interface. The generated path is immediately sent to the anti-misoperation rule base for verification. This rule base includes mandatory rules such as the five preventions (e.g., preventing disconnection switches from being pulled under load). At the same time, the system displays the path on the human-machine interface in two forms: sequential operation ticket and topology coloring animation. The operator can see the changes in the power grid status after each step of the operation and has the final confirmation right.
[0056] The instruction execution module is used to convert the confirmed operation path into a standard instruction sequence and send it to the automation control system for execution. After the operator confirms the instruction, the system uses the instruction execution module to convert the operation path into a standard instruction sequence that conforms to the field automation equipment protocol (such as GOOSE or SV messages of IEC 61850) and send it out for execution in sequence. The system will strictly monitor the verification and execution results of each instruction to ensure that the operation is accurate.
[0057] System Hardware and Software Architecture: The physical implementation of this system is based on a high-performance computing platform, which mainly includes: a data server, an application server (with a built-in high-performance GPU computing card), network communication equipment, and operator workstations;
[0058] The data server is responsible for acquiring real-time grid topology information, circuit breaker / disconnector status, power flow data, etc. from the data acquisition and monitoring system (SCADA) and energy management system (EMS) of the dispatch center, and building a dynamic grid model;
[0059] The application server is the intelligent core of the system, carrying the intelligent decision-making module, the simulation training environment module, and the online learning module. Its built-in GPU is used to accelerate the training and inference process of deep neural networks (such as the graph neural network).
[0060] Network communication equipment ensures secure and reliable data exchange between the system and SCADA / EMS and station control layer automation equipment;
[0061] The operator workstation provides a human-computer interaction interface for tasks such as task assignment, path visualization, and execution confirmation.
[0062] It also includes an online learning module, which is used to collect operational data after actual operation to perform incremental learning on the artificial intelligence decision-making model. With the online learning function, the system has the ability to learn online. After each actual operation is completed, the relevant state, action and reward sequence will be recorded and stored in the experience playback buffer. The system will use this real data during idle time (such as at night) to fine-tune the AI model in the form of offline reinforcement learning, so that it can adapt to the actual operating characteristics of the power grid and continuously optimize its performance.
[0063] As can be seen from the above, the specific embodiments of the present invention are as follows:
[0064] System Hardware and Software Architecture: The physical implementation of this system is based on a high-performance computing platform, which mainly includes: a data server, an application server (with a built-in high-performance GPU computing card), network communication equipment, and operator workstations;
[0065] The data server is responsible for acquiring real-time grid topology information, circuit breaker / disconnector status, power flow data, etc. from the data acquisition and monitoring system (SCADA) and energy management system (EMS) of the dispatch center, and building a dynamic grid model;
[0066] The application server is the intelligent core of the system, carrying the intelligent decision-making module, the simulation training environment module, and the online learning module. Its built-in GPU is used to accelerate the training and inference process of deep neural networks (such as the graph neural network).
[0067] Network communication equipment ensures secure and reliable data exchange between the system and SCADA / EMS and station control layer automation equipment;
[0068] The operator workstation provides a human-computer interaction interface for tasks such as task assignment, path visualization, and execution confirmation.
[0069] The specific implementation of the method flow is as follows: Step 1: Dynamic power grid model construction. The system connects to the SCADA / EMS system through standard communication protocols (such as IEC61850, IEC 104) and periodically (e.g., once per second) acquires real-time data of the entire network. Based on this data, the system maintains a graphical power grid model in memory. This model uses substations and lines as nodes and circuit breakers and disconnect switches as edges. Each node and edge is accompanied by the current electrical parameters (such as voltage, current, power) and status (such as closed and open). This model is the basis for all subsequent simulations and decisions.
[0070] Step 2: Switching operation task analysis. The operator issues tasks through graphical drag-and-drop or by selecting preset task templates on the human-machine interface. For example, switching 110kV line L1 from operation to maintenance. After the system receives the instruction, the task analysis module will automatically analyze the initial state (the state of all switches associated with L1) and the target state (the state of all switches required for maintenance) of the task, and convert it into a state transition problem consisting of a sequence of equipment operations.
[0071] Step 3: Intelligent Path Generation: Simulation Environment. Based on a dynamic power grid model, the system develops a power grid operation simulator that can simulate the changes in power grid topology and power flow after the execution of any compliant operation command (such as "disconnecting circuit breaker CB1").
[0072] AI Decision Model Training: Model Selection: In this embodiment, a graph neural network (GAT) based on attention mechanism is preferred as the policy network because it can effectively process the graph structure data of the power grid and focus on the key equipment affected by the current operation through the attention mechanism;
[0073] Training method: The model is trained using the Proximal Policy Optimization (PPO) deep reinforcement learning algorithm. The training process is carried out in a simulation environment, where the AI agent explores the path from the initial state to the target state by trying various operations.
[0074] Reward Function Design: The reward function is key to guiding AI learning, and its specific design is as follows: Reward Function At every decision moment The value of Defined by the following functions: Security reward component The calculation method is as follows: Anything that causes equipment overload, voltage exceeding limits, or damage Operating according to safety guidelines will result in significant negative rewards (penalties), while efficiency will result in rewards. : =-1 (A fixed small penalty is given for each additional step, encouraging the shortest path), target reward : =+100 (Only awarded upon successfully reaching the target state, to encourage task completion);
[0075] Path generation: For a new task, the pre-trained model is loaded. Based on the current power grid state, the model calculates the probability of each legal operation through the GAT network, and then selects the operation with the highest probability. The system sends this operation to the simulator, updates the state, and repeats this process until the target state is reached, thereby generating a complete operation path. To obtain a better solution, Monte Carlo Tree Search (MCTS) can be combined, using the model output as a prior probability to guide the search and evaluate more possibilities.
[0076] Step 4: Path Verification and Visualization. The generated path will be immediately sent to the anti-misoperation rule base for verification. This rule base includes mandatory rules such as the five preventions (e.g., preventing disconnection switches from being pulled under load). At the same time, the system displays the path on the human-machine interface in two forms: sequential operation ticket and topology coloring animation. The operator can see the changes in the power grid status after each operation and has the final confirmation right.
[0077] Step 5: Command execution. After operator confirmation, the system converts the operation path into a standard command sequence that conforms to the field automation equipment protocol (such as GOOSE or SV messages of IEC 61850) through the command execution module, and sends them out for execution in sequence. The system will strictly monitor the verification and execution results of each command to ensure that the operation is accurate.
[0078] The system has an online learning function. After each actual operation is completed, the relevant status, actions and reward sequences will be recorded and stored in the experience playback buffer. The system will use this real data during idle time (such as at night) to fine-tune the AI model in an offline reinforcement learning manner so that it can adapt to the actual operating characteristics of the power grid and continuously optimize its performance.
[0079] Taking the power outage and switching operation of a line in a 220kV substation as an example, in the traditional way, it takes 10-15 minutes for an experienced operator to draft and verify the ticket. After adopting this system, it only takes about 30 seconds from the time the task is issued to the generation of 3 safe and optimized candidate paths. The operator selects the one with the fewest steps (5 steps in total), and executes it with one click after visual confirmation. The whole process is completed within 2 minutes, improving efficiency by more than 80% and completely avoiding the risk of human error.
[0080] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. An artificial intelligence-based switching operation path optimization method, characterized in that, Comprise the following steps: S1, dynamic modeling step: real-time acquisition of the topological connection relationship, real-time state and electrical parameters of the equipment of the power grid from the data acquisition and monitoring system, building and continuously updating the dynamic power grid model for path simulation; S2, task analysis step: receiving and analyzing the switching operation task, and clarifying the initial state and target state of the operation; S3, intelligent path generation step: taking the dynamic power grid model as the simulation environment, using the pre-trained artificial intelligence decision model for path optimization, wherein the model is trained by deep reinforcement learning, and its reward function integrates power grid safety constraints and operation efficiency targets to generate candidate operation paths; S4, path verification and visualization step: automatically checking the generated candidate path with the anti-misoperation rule library, and visualizing and confirming through the man-machine interface; S5, instruction execution step: converting the confirmed operation path into a standard instruction sequence and issuing it to the automation control system for execution.
2. The artificial intelligence-based switching operation path optimization method according to claim 1, characterized in that: The reward function At each decision time The value of Defined by the following function: Where, Is the safety reward component, Is the operation efficiency reward component, Is the goal-oriented reward component; , Is the weight coefficient of each reward component.
3. The artificial intelligence-based switching operation path optimization method according to claim 2, characterized in that: The security reward component The calculation method is as follows: Wherein, is the voltage of the node , is the rated voltage thereof, is the maximum allowable deviation voltage; is the line current, is the rated current thereof; is a severity index of the check violation; , , corresponding penalty coefficients.
4. The artificial intelligence-based switching operation path optimization method of claim 1, wherein: The artificial intelligence decision model is implemented by a graph neural network based on an attention mechanism, which takes the power grid topology graph as input, dynamically focuses on key equipment related to current decision by calculating the attention weight of nodes and edges in the graph.
5. The artificial intelligence-based switching operation path optimization method according to claim 1, characterized in that: The intelligent path generation step uses the Monte Carlo tree search algorithm as a planner, and uses the action probability distribution output by the artificial intelligence decision model as prior knowledge to guide the search process.
6. The artificial intelligence-based switching operation path optimization method according to claim 1, characterized in that: It also includes an online learning step, which collects operation data to incrementally learn the artificial intelligence decision model after actual operation to optimize its decision performance.
7. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the method of any one of claims 1-6.
8. An artificial intelligence-based switching operation path system, characterized by: Comprise: A dynamic modeling module for real-time acquisition of the topological connection relationship, real-time state and electrical parameters of the equipment of the power grid from the data acquisition and monitoring system, building and continuously updating the dynamic power grid model for path simulation; A task analysis module for receiving and analyzing the switching operation task, and clarifying the initial state and target state of the operation; An intelligent decision module for taking the dynamic power grid model as the simulation environment, using the pre-trained artificial intelligence decision model for path optimization, wherein the model is trained by deep reinforcement learning, and its reward function integrates power grid safety constraints and operation efficiency targets to generate candidate operation paths; A verification and display module for automatically checking the generated candidate path with the anti-misoperation rule library, and visualizing and confirming through the man-machine interface; An instruction execution module for converting the confirmed operation path into a standard instruction sequence and issuing it to the automation control system for execution.
9. The artificial intelligence-based switching operation path system according to claim 8, characterized in that: The intelligent decision module further comprises a policy network unit and a reward calculation unit, the policy network unit is used to run the artificial intelligence decision model, and the reward calculation unit is used to calculate the reward value of the decision in real time according to the pre-defined rules, and the calculation logic is defined by the reward function.
10. The artificial intelligence-based switching operation path system of claim 8, wherein: It also includes an online learning module for incrementally learning the artificial intelligence decision model based on operation data after actual operation.