Generation method and device of power grid task operation scheme, equipment, medium and product

Abstract

The embodiment of the invention provides a power grid task operation scheme generation method and device, equipment, a medium and a product, and is applied to the technical field of power system automation. The method comprises the following steps: in response to an operation instruction of a dispatcher, inputting the operation instruction into a pre-trained task generation model to obtain a first operation scheme corresponding to the operation instruction; performing logic verification on the first operation scheme based on a preset verification rule to obtain a second operation scheme; inputting the second operation scheme and dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation scheme; wherein the pre-trained task generation model and the dispatcher preference model are obtained by performing model training based on historical operation records of the power grid task. The technical effect of improving the generation efficiency of the operation task is achieved.

Description

Technical Field

[0001] The present application relates to the field of power system automation technology, and in particular to a method, device, equipment, medium and product for generating a power grid task operation plan. Background Art

[0002] In the fields of power grid dispatching and industrial operations, operational tasks involve multiple aspects of equipment management, process management, and safety management. The overall operational process is highly complex. Original task generation is primarily manual, based on the dispatcher's experience, and is then manually reviewed to determine the final task. Due to the high cost of manual review and writing, existing technologies have proposed automated task generation based on predefined rules.

[0003] In the prior art, predefined rules are used to assist dispatchers in generating operation tasks, and the operation tasks are reviewed through a preset review process to determine the final operation tasks.

[0004] Since the existing technology relies on manual participation, there is a technical problem in that the generation efficiency of operation tasks is low. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, equipment, medium and product for generating a power grid task operation plan, so as to achieve the technical effect of improving the efficiency of generating operation tasks.

[0006] In a first aspect, an embodiment of the present application provides a method for generating a power grid task operation plan, comprising:

[0007] In response to the dispatcher's operation instruction, the operation instruction is input into the pre-trained task generation model to obtain a first operation plan corresponding to the operation instruction;

[0008] Performing a logic check on the first operation plan based on a preset check rule to obtain a second operation plan;

[0009] Inputting the second operation plan and the dispatcher information corresponding to the operation instruction into the pre-trained dispatcher preference model to obtain an optimized third operation plan;

[0010] Among them, the pre-trained task generation model and dispatcher preference model are both trained based on the historical operation records of power grid tasks.

[0011] In one possible implementation, performing a logic check on the first operation plan based on a preset check rule to obtain a second operation plan includes:

[0012] Analyze the device operation sequence, operation task execution order, and operation task execution time in the first operation plan;

[0013] Based on the preset verification rules, the device operation sequence, the operation task execution order, and the logical conflicts and time conflicts in the operation task execution time are detected to obtain the verification results;

[0014] When the verification result indicates that the verification fails, the first operation plan is repaired based on the verification result to obtain a second operation plan.

[0015] In one possible implementation, the second operation plan and the dispatcher information corresponding to the operation instruction are input into a pre-trained dispatcher preference model to obtain an optimized third operation plan, including:

[0016] extracting solution features based on the second operation solution;

[0017] Extract dispatcher features based on dispatcher information;

[0018] Inputting the scheme characteristics and the dispatcher characteristics into the dispatcher preference model to obtain an optimization strategy for the second operation scheme;

[0019] The second operation plan is optimized based on the optimization strategy to obtain a third operation plan.

[0020] In a possible implementation, after obtaining the optimized third operation solution, the method further includes:

[0021] Push the third operation plan to the dispatcher's corresponding device terminal and instruct the dispatcher to confirm and execute the plan;

[0022] Collecting the dispatcher's operational feedback information on the third operation plan;

[0023] Based on the operation feedback information, the task generation model and the dispatcher preference model are incrementally learned online to obtain an updated task generation model and an updated dispatcher preference model.

[0024] In one possible implementation, the method further includes:

[0025] In response to the grid state mutation request, interrupting the currently executed third operation plan and acquiring real-time operation state data corresponding to the grid task;

[0026] The unexecuted operation tasks in the third operation plan are adjusted based on the real-time operation status data to obtain a fourth operation plan.

[0027] In one possible implementation, online incremental learning is performed on the task generation model and the dispatcher preference model based on the operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, including:

[0028] The operation feedback information is stored in the incremental database, and the size of the accumulated operation feedback information in the incremental database is calculated;

[0029] When the size of the accumulated operation feedback information in the incremental database reaches a preset number, online incremental learning is performed on the task generation model and the dispatcher preference model based on the accumulated operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, and the incremental database is cleared.

[0030] In one possible implementation, before responding to the dispatcher's operation instruction and inputting the operation instruction into a pre-trained task generation model to obtain a first operation plan corresponding to the operation instruction, the method further includes:

[0031] Obtain historical operation records of power grid tasks and construct the first sample training data set;

[0032] Fine-tune the initial task generation model based on the first sample training data set to obtain a task generation model to be determined;

[0033] Fixing the model parameters of the task generation model to be determined, and generating a second sample training data set based on the task generation model to be determined and the first sample training data set;

[0034] Performing comparative training on the initial dispatcher preference model based on the second sample training data set to obtain a dispatcher preference model to be determined;

[0035] Calculate the joint loss function value based on the to-be-determined task generation model and the to-be-determined dispatcher preference model;

[0036] When the joint loss function value is lower than the preset loss function value, the task generation model to be determined is determined as the target task generation model, and the dispatcher preference model to be determined is determined as the target preference model.

[0037] In a second aspect, an embodiment of the present application provides a device for generating a power grid task operation plan, comprising:

[0038] An acquisition module, configured to respond to an operation instruction of the dispatcher, input the operation instruction into a pre-trained task generation model, and obtain a first operation plan corresponding to the operation instruction;

[0039] A first processing module is configured to perform a logic check on the first operation plan based on a preset check rule to obtain a second operation plan;

[0040] A second processing module is used to input the second operation plan and the dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation plan;

[0041] Among them, the pre-trained task generation model and dispatcher preference model are both trained based on the historical operation records of power grid tasks.

[0042] In a possible implementation, the first processing module is further configured to:

[0043] Analyze the device operation sequence, operation task execution order, and operation task execution time in the first operation plan;

[0044] Based on the preset verification rules, the device operation sequence, the operation task execution order, and the logical conflicts and time conflicts in the operation task execution time are detected to obtain the verification results;

[0045] When the verification result indicates that the verification fails, the first operation plan is repaired based on the verification result to obtain a second operation plan.

[0046] In a possible implementation, the second processing module is further configured to:

[0047] extracting a solution feature based on the second operation solution;

[0048] Extract dispatcher features based on dispatcher information;

[0049] Inputting the scheme characteristics and the dispatcher characteristics into the dispatcher preference model to obtain an optimization strategy for the second operation scheme;

[0050] The second operation plan is optimized based on the optimization strategy to obtain a third operation plan.

[0051] In a possible implementation, the device further includes a fourth processing module, configured to:

[0052] Push the third operation plan to the dispatcher's corresponding device terminal and instruct the dispatcher to confirm and execute the plan;

[0053] Collecting the dispatcher's operational feedback information on the third operation plan;

[0054] Based on the operation feedback information, the task generation model and the dispatcher preference model are incrementally learned online to obtain an updated task generation model and an updated dispatcher preference model.

[0055] In a possible implementation, the fourth processing module is further configured to:

[0056] In response to the grid state mutation request, interrupting the currently executed third operation plan and acquiring real-time operation state data corresponding to the grid task;

[0057] The unexecuted operation tasks in the third operation plan are adjusted based on the real-time operation status data to obtain a fourth operation plan.

[0058] In a possible implementation, the fourth processing module is further configured to:

[0059] The operation feedback information is stored in the incremental database, and the size of the accumulated operation feedback information in the incremental database is calculated;

[0060] When the size of the accumulated operation feedback information in the incremental database reaches a preset number, online incremental learning is performed on the task generation model and the dispatcher preference model based on the accumulated operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, and the incremental database is cleared.

[0061] In a possible implementation, the acquisition module is further configured to:

[0062] Obtain historical operation records of power grid tasks and construct the first sample training data set;

[0063] Fine-tune the initial task generation model based on the first sample training data set to obtain a task generation model to be determined;

[0064] Fixing the model parameters of the task generation model to be determined, and generating a second sample training data set based on the task generation model to be determined and the first sample training data set;

[0065] Performing comparative training on the initial dispatcher preference model based on the second sample training data set to obtain a dispatcher preference model to be determined;

[0066] Calculate the joint loss function value based on the to-be-determined task generation model and the to-be-determined dispatcher preference model;

[0067] When the joint loss function value is lower than the preset loss function value, the task generation model to be determined is determined as the target task generation model, and the dispatcher preference model to be determined is determined as the target preference model.

[0068] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor;

[0069] Memory stores computer-executable instructions;

[0070] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and various possible implementations of the first aspect.

[0071] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the above-mentioned first aspect and various possible implementation methods of the first aspect.

[0072] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and various possible implementation methods of the first aspect.

[0073] The embodiments of the present application provide a method, device, equipment, medium and product for generating a power grid task operation plan. The method uses a pre-trained task generation model to analyze the operation instructions input by the dispatcher to obtain a first operation plan corresponding to the operation instruction; uses a preset verification rule to perform a logical verification on the first operation plan to obtain a second operation plan after the logical verification; uses a pre-trained dispatcher preference model combined with dispatcher information to optimize the second operation plan to obtain a third operation plan after optimization. Compared with the existing technology, the present application uses a task generation model to realize the rapid generation of operation plans, and uses preset verification rules and a dispatcher preference model to optimize the operation plans, thereby improving the effectiveness of the operation plans and achieving the technical effect of improving the generation efficiency of operation tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0075] Figure 1 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 1 ;

[0076] Figure 2 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 2 ;

[0077] Figure 3 A flowchart of the model joint training method provided in this application;

[0078] Figure 4 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 3 ;

[0079] Figure 5 A schematic diagram of the structure of a device for generating a power grid task operation plan provided in this application;

[0080] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application.

[0081] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0082] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0083] First, the proper nouns involved in this application are explained:

[0084] Proximal Policy Optimization (PPO): A reinforcement learning algorithm based on policy gradients. Its core idea is to ensure the stability of the training process by limiting the magnitude of policy updates, thus avoiding the performance crash caused by large updates in traditional policy gradient methods.

[0085] In the existing technology, based on the accumulation of historical power grid task operation records, relevant rules are summarized to obtain a set of predefined rules for generating operation tasks. The predefined rules are used to assist dispatchers in writing and generating operation tasks, and the operation tasks are reviewed and optimized in combination with the review process, so as to obtain an operation plan for the power grid task.

[0086] However, in the prior art, predefined rules can only realize simple task writing, and manual participation is still required when facing complex tasks. Therefore, there is a technical problem of low efficiency in generating operation tasks in the prior art.

[0087] In response to the above technical problems, the present application proposes the following technical ideas: Compared with the method of using rules to assist in manually generating operation plans in the prior art, the present application uses a trained model to quickly generate and optimize tasks. Specifically: when receiving the dispatcher's operation instructions, the dispatcher's operation instructions are input into the pre-trained task generation model to obtain a first operation plan; the logic of the first operation plan is verified using preset verification rules to obtain a second operation plan after logic verification; wherein, the second operation plan has been preliminarily optimized in the logic verification stage; the second operation plan is optimized using a pre-trained dispatcher preference model, so that the operation plan is more inclined to the current dispatcher's work habits, and a third operation plan is obtained. Compared with the prior art, the present application uses a task generation model to achieve efficient solution generation, and uses preset verification rules and a dispatcher preference model to optimize the operation plan, thereby ensuring the efficiency of solution generation and achieving the technical effect of improving the efficiency of operation task generation.

[0088] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0089] Figure 1 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0090] S101. In response to an operation instruction from a dispatcher, the operation instruction is input into a pre-trained task generation model to obtain a first operation plan corresponding to the operation instruction.

[0091] In this step, the information in the operation instruction includes but is not limited to: target device, operation type, time requirement, appliance status, and associated devices. The information in the first operation plan includes: at least one operation task, the execution order of each operation task, and the target device corresponding to each operation task.

[0092] For example, the input operation instruction is: [Please perform maintenance operations on transformer No. 1 of station A. The operation steps must include power outage, hanging ground wires, etc.]; the corresponding operation plan is: [{Step 1, operation task: power outage; equipment: transformer No. 1}, {Step 2, operation task: hanging ground wires, equipment: transformer No. 1}].

[0093] In this step, the pre-trained task generation model is a model obtained through machine learning or deep learning training. The model type can be: sequence-to-sequence model, reinforcement learning model, and hybrid model.

[0094] Optionally, when the operation instruction is input into the task generation model, a possible implementation method of obtaining the first operation solution is:

[0095] S1011: Parse the operation instructions, and convert the natural language type operation instructions into formatted instructions.

[0096] S1012: Extract features from the formatting instructions, convert them into a numerical format acceptable to the model, and obtain preprocessing instructions.

[0097] S1013: Input the preprocessing instruction into the task generation model to obtain the numerical result of the output of the model reasoning.

[0098] S1014: Decode the numerical result to obtain a first operation solution in a readable format.

[0099] In this step, the numerical result may be an index, and the decoding method may be to convert a word or phrase according to the index, thereby obtaining a readable first operation solution.

[0100] S102: Perform a logic check on the first operation plan based on a preset check rule to obtain a second operation plan.

[0101] Optionally, a possible implementation of obtaining the second operation solution through logic verification is:

[0102] S1021. Analyze the device operation sequence, operation task execution order, and operation task execution time in the first operation plan.

[0103] In this step, the device operation sequence refers to obtaining the operation sequence for different target devices according to the execution order of the operation tasks.

[0104] S1022: Detect the device operation sequence, the operation task execution order, and the logical conflicts and time conflicts in the operation task execution time based on the preset verification rules to obtain a verification result.

[0105] In this step, the preset verification rules include time conflict verification and logic conflict verification, which are used to determine whether there is a conflict in the execution time of the operation task, and whether there is a logic conflict between the execution order of the operation task and the device operation sequence.

[0106] S1023: When the verification result indicates that the verification fails, repair the first operation plan based on the verification result to obtain a second operation plan.

[0107] In this step, when the verification result indicates that the verification fails, it is necessary to repair the first operation plan according to the correction suggestion or correction plan in the verification result, thereby obtaining the second operation plan.

[0108] For example, the logic check part is described with reference to steps S1021 to S1023:

[0109] a1. Enter the first operation plan. The specific information in the plan is:

[0110] [{Step Operation Task: Power Outage, Equipment: A Substation No. 1 Main Transformer, Planned Time: 05:00},

[0111] {Step 1, Operation Task: Hanging Ground Wire, Equipment: High Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 05:15},

[0112] {Step 2, Operation Task: Hanging Ground Wire, Equipment: Low Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 05:20},

[0113] {Step 3, Operation Task: Maintenance Work, Equipment: Main Transformer No. 1 of Substation A, Planned Time: 05:30-06:30},

[0114] {Step 4, Operation Task: Remove the Grounding Wire, Equipment: Low Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 06:35},

[0115] {Step 5, Operation Task: Remove the Grounding Wire, Equipment: High Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 06:40},

[0116] {Step 6, Operation Task: Restore power supply, Equipment: Main transformer No. 1 of substation A, Planned time: 06:45}].

[0117] a2. Analyze the first operation plan to obtain the device operation sequence, the operation task execution order, and the operation task execution time.

[0118] a3. Logical conflict check: Check the logical consistency of all operation steps; Time conflict check: Ensure that multiple operations are not performed on the same device in the same time period.

[0119] a4. Get the verification results, specifically:

[0120] [{Problem type: Equipment status conflict, Description: The No. 1 main transformer of substation A is currently carrying a load of 30MW. A direct power outage will result in load loss. Severity level: High. Related steps and tasks: Power outage operation of the No. 1 main transformer of substation A},

[0121] {Problem type: Insufficient operation interval, Description: The operation interval of the grounding wire of substation A (5 minutes) is less than the safety requirement (15 minutes), Severity level: Medium, Related step operation task: Grounding wire operation of substation A}].

[0122] a5. Determine the revision plan, specifically:

[0123] [{Suggestion type: load transfer, suggestion content: before the main transformer of station A is powered off, transfer the load to the No. 2 main transformer of station A},

[0124] {Suggestion type: Time adjustment, Suggestion content: Extend the grounding wire operation interval of Station A to more than 15 minutes}].

[0125] a6. Modify the first operation plan based on the revised plan to obtain the second operation plan:

[0126] [{Step 1, Operation Task: Load Transfer, Equipment: Main Transformer No. 1 → Main Transformer No. 2 of Substation A, Planned Time: 04:30-05:00},

[0127] {Step 2, Operation Task: Power Outage, Equipment: Main Transformer No. 1 of Substation A, Planned Time: 05:00},

[0128] {Step 3, Operation Task: Hanging Ground Wire, Equipment: High Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 05:15},

[0129] {Step 4, Operation Task: Hanging Ground Wire, Equipment: Low Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 05:30},

[0130] {Step 5, Operation Task: Maintenance Work, Equipment: Main Transformer No. 1 of Substation A, Planned Time: 05:45-06:45},

[0131] {Step 6, Operation Task: Remove the Grounding Wire, Equipment: Low Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 06:50},

[0132] {Step 7, Operation Task: Remove the Grounding Wire, Equipment: High Voltage Side of No. 1 Main Transformer in Substation A, Planned Time: 07:05},

[0133] {Step 8, Operation Task: Restore Power Supply, Equipment: Main Transformer No. 1 of Substation A, Planned Time: 07:10},

[0134] {Step 9, Operation Task: Load Shedding, Equipment: Main Transformer No. 2 → Main Transformer No. 1 of Substation A, Planned Time: 07:10-07:30}].

[0135] It should be noted that the examples in this embodiment are only for illustrative purposes.

[0136] S103: Input the second operation plan and the dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation plan.

[0137] In this step, the pre-trained task generation model and dispatcher preference model are trained based on the historical operation records of power grid tasks. The dispatcher preference model is used to adjust the second operation plan based on the dispatcher's behavioral habits to make it more consistent with the dispatcher's work habits.

[0138] Optionally, a possible implementation of the third operation plan optimized based on the dispatcher preference model is:

[0139] S1031. Extract solution features based on the second operation solution.

[0140] In this step, the types of solution features include but are not limited to: task structure features, time distribution features, equipment-related features, safety features, and efficiency features.

[0141] Exemplarily, a possible implementation method of extracting the solution feature of the second operation solution is:

[0142] b1. Perform structured data analysis on the second operation plan to obtain plan features in multiple dimensions.

[0143] b2. The solution features of multiple dimensions are combined to obtain the solution features of the second operation solution.

[0144] S1032. Extract dispatcher features based on dispatcher information.

[0145] In this step, the dispatcher information includes: the dispatcher's historical work records and the dispatcher's identity information. The dispatcher's characteristics include but are not limited to: identity characteristics, historical preference characteristics, performance characteristics, recent behavior characteristics, and personalized characteristics.

[0146] Exemplarily, one possible implementation method for extracting dispatcher features is to extract dispatcher features using a deep learning model.

[0147] S1033: Input the scheme characteristics and the dispatcher characteristics into the dispatcher preference model to obtain an optimization strategy for the second operation scheme.

[0148] In this step, the dispatcher characteristics and the scheme characteristics are input into the dispatcher preference model, and the obtained optimization strategy is used to adjust the operation task execution order, operation task execution time, and addition and deletion of operation tasks of the second operation scheme.

[0149] For example, the purpose of the second operation scheme is to perform busbar maintenance. The second operation scheme is specifically as follows:

[0150] [{Step 1, Operation Task: Load Transfer, Equipment: 220kV I mother load transfer to II mother, Planned Time: 05:00-05:30},

[0151] {Step 2, Operation Task: Power Outage Operation, Equipment: 220kV I / O Bus, Planned Time: 05:30-05:45},

[0152] {Step 3, Operation Task: Electrical Test, Equipment: 220kV I / O Bus, Planned Time: 05:45-05:50},

[0153] {Step 4, Operation Task: Hanging Ground Wire, Equipment: Both Sides of 220kVI Bus, Planned Time: 05:50-06:00},

[0154] {Step 5, Operation Task: Maintenance Work, Equipment: 220kV I female insulator replacement, Planned Time: 06:00-08:00},

[0155] {Step 6, Operation Task: Remove the ground wire, Equipment: Both sides of the 220kV VI bus, Planned Time: 08:00-08:10},

[0156] {Step 7, Operation Task: Restore power supply, Equipment: 220kV I bus, Planned time: 08:10-08:30}].

[0157] The optimization strategy generated based on the dispatcher preference model is:

[0158] Optimization strategy: {temporary optimization: {recommended strategy: postpone the start time by 1 hour, reason: match the dispatcher's morning operation preference, range: {update start time: 06:00, time increase: +0.15}},

[0159] Safety Enhancement: {Recommended strategy: Add electrical testing steps and extend the grounding wire interval. Reason: Historical modifications show that 80% of tasks have added safety measures.}

[0160] Parallel optimization: {Recommended strategy: Execute safety measures and maintenance preparation in parallel, Reason: Reduce total time by approximately 30 minutes, Scope: {Parallel step operation tasks: [Step 3, Step 4]}}.

[0161] S1034: Optimize the second operation plan based on the optimization strategy to obtain a third operation plan.

[0162] Exemplarily, combined with the example in step S1033, the third operation scheme obtained is:

[0163] [{Step 1, Operation Task: Load Transfer, Equipment: 220kV I mother load transfer to II mother, Planned Time: 06:00-06:30, Modification Content: Time Adjustment},

[0164] {Step 2, Operation Task: Power Outage Operation, Equipment: 220kV I / O Bus, Planned Time: 06:30-06:45},

[0165] {Step 3, Operation Task: Electrical Test, Equipment: 220kV VI Bus, Planned Time: 06:45-06:50},

[0166] {Step 4, Operation Task: Hang Ground Wire, Equipment: Both Sides of 220kV I Bus, Planned Time: 06:50-07:05, Modification Content: Interval Extended to 15 Minutes},

[0167] {Step 5, Operation Task: Secondary Electrical Test, Equipment: 220kV I / O Bus, Planned Time: 07:05-07:08, Modification Content: Add a new step},

[0168] {Step 6, Operation Task: Maintenance Work, Equipment: 220kV I female insulator replacement, Planned Time: 07:08-09:08},

[0169] {Step 7, Operation Task: Remove the Grounding Wire, Equipment: Both Sides of the 220kVI Bus, Planned Time: 09:08-09:23, Modification Content: Extend the Interval},

[0170] {Step 8, Operation Task: Restore power supply, Equipment: 220kV I bus, Planned time: 09:23-09:40}].

[0171] The embodiment of the present application provides a method for generating a power grid task operation plan. The method uses a pre-trained task generation model to analyze the operation instructions input by the dispatcher to obtain a first operation plan corresponding to the operation instruction; uses preset verification rules to perform a logical verification on the first operation plan to obtain a second operation plan after the logical verification; and uses a pre-trained dispatcher preference model combined with dispatcher information to optimize the second operation plan to obtain a third operation plan after optimization. Compared with the existing technology, the present application uses a task generation model to achieve rapid generation of operation plans, and uses preset verification rules and a dispatcher preference model to optimize operation plans, thereby improving the effectiveness of operation plans and achieving the technical effect of improving the efficiency of generating operation tasks.

[0172] Figure 2 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 2 ,like Figure 2 As shown, the method includes:

[0173] S201. Push the third operation plan to the device terminal corresponding to the dispatcher, and instruct the dispatcher to confirm and execute the plan.

[0174] In this step, the method of implementing the plan push and confirming the execution can be: pushing the third operation plan to multiple device terminals related to the dispatcher at the same time, and displaying the detailed text information of the third operation plan and the timeline of the operation task on the device terminal; prompting the dispatcher to perform data signature confirmation or password verification to verify whether the current device operator has the operation authority.

[0175] Optionally, when instructing the dispatcher to confirm and execute the plan, the execution status of the plan of the power grid can be monitored in real time, and the third operation plan can be dynamically updated, specifically:

[0176] S2011. In response to a grid state mutation request, interrupt the currently executing third operation plan and obtain real-time operation state data corresponding to the grid task.

[0177] In this step, the real-time operating status data corresponding to the power grid task includes: device status data, device electrical data, and mutation data corresponding to the power grid status mutation request.

[0178] For example, in response to a grid state sudden change request, the execution of the interruption may be:

[0179] c1. Monitor the operating status of the power grid in real time and obtain real-time data flow in the power grid.

[0180] c2. Generate an operation interrupt instruction when the real-time data flow exceeds the safety threshold.

[0181] c3. Based on the operation interrupt instruction, the execution of the third operation plan is interrupted, the context corresponding to the operation task being executed at the time of interruption is recorded, and real-time status operation data is obtained.

[0182] S2012: Adjust the unexecuted operation tasks in the third operation plan based on the real-time operation status data to obtain a fourth operation plan.

[0183] In this step, the operation tasks that are not executed in the third operation plan are adjusted to obtain the fourth operation plan in the following manner:

[0184] d1. Determine the unexecuted operation tasks in the third operation plan based on the context corresponding to the operation tasks at the time of interruption.

[0185] d2. Evaluate the safety impact level of the unexecuted operation tasks to obtain the safety impact level of each unexecuted operation task.

[0186] In this step, the safety impact level can be generated by inputting the operation task into a pre-trained level output model to obtain the corresponding safety impact level; wherein the level output model is a prediction model obtained by deep learning using historical level assessment records.

[0187] d3. When the security impact level corresponding to the unexecuted operation task indicates low risk, the operation task will not be modified.

[0188] d4. When the security impact level corresponding to the unexecuted operation task indicates medium risk, adjust the parameters of the operation task, and modify the execution time and / or execution order of the operation task.

[0189] d5. When the security impact level corresponding to the unexecuted operation task indicates high risk, the steps corresponding to the operation task in the third operation plan are canceled, and a replacement task for the operation task is generated.

[0190] d6. Generate a fourth operation plan based on the adjusted operation task.

[0191] Exemplarily, the third operation plan is used for load transfer and power outage operations, specifically: [{Step 1, Operation Task: Power Test, Equipment: 220KV I Bus, Planned Time: 14:15-14:20}, {Step 2, Operation Task: Restore Power, Equipment: 220KV I Bus, Planned Time: 14:30-14:45}]. The grid mutation request includes: mutation time: 14:18, mutation event: line failure. At this time, step 1 of the third operation plan is immediately interrupted, and the context of the interruption is automatically recorded. The unexecuted steps include steps 1 and 2. An analysis of the unexecuted steps yields the following results: Step 1: The original target device fails, so the adjacent device is changed; Step 2: The line failure makes the plan unfeasible, so the backup plan is changed.

[0192] The fourth operation plan after update is: [{Step 1, operation task: fault isolation, equipment: line cut-off switch, planned time: 14:25-14:30}, {Step 2, operation task: backup line switching, equipment: backup line, planned time: 14:35-14:50}, {Step 3, operation task: restricted restoration, equipment: 220KV I bus, planned time: 15:00-15:15}].

[0193] In this step, in addition to obtaining the fourth operation plan by adjusting the third operation plan, an operation instruction may be generated based on the real-time operation status data, and the operation instruction may be input into the task generation model to regenerate the operation plan.

[0194] S202: Collect the dispatcher's operational feedback information on the third operation plan.

[0195] In this step, the operation feedback information includes execution feedback on each operation task in the third operation scheme, specifically: time arrangement feedback, execution sequence feedback, task arrangement feedback, and parameter value feedback.

[0196] Optionally, the collection of operation feedback information in this step includes: explicit feedback collection and implicit feedback collection.

[0197] Among them, display feedback collection includes: obtaining operation feedback information by pushing mandatory scores and optional opinions; obtaining the dispatcher's annotation information for each operation task.

[0198] Implicit feedback collection includes: collecting execution records, operation change records, and execution time change records of the third operation plan during execution; or obtaining feedback information by analyzing the interactive data of the dispatcher when performing the operation task.

[0199] S203 : Performing online incremental learning on the task generation model and the dispatcher preference model based on the operation feedback information to obtain an updated task generation model and an updated dispatcher preference model.

[0200] Optionally, a possible implementation of online incremental learning is:

[0201] S2031. Store the operation feedback information in an incremental database, and calculate the size of the accumulated operation feedback information in the incremental database.

[0202] In this step, the incremental database may be designed in a ring buffer data structure, an event sequence storage format, or distributed shard storage.

[0203] The size of the accumulated operation feedback information needs to be calculated. The calculation method can be: calculation based on the accumulated number of records, calculation based on the amount of information, or a mixed calculation based on the number of records and the amount of information.

[0204] S2032. When the size of the accumulated operation feedback information in the incremental database reaches a preset number, online incremental learning is performed on the task generation model and the dispatcher preference model based on the accumulated operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, and the incremental database is cleared.

[0205] In this step, the method of performing online incremental learning based on the accumulated operation feedback information can be:

[0206] e1. Perform data preprocessing based on the accumulated operation feedback information to obtain an incremental training dataset.

[0207] e2. Use the PPO reinforcement learning method combined with the incremental training data set to train the task generation model and obtain an updated task generation model.

[0208] e3. Use the gradient descent method combined with the incremental training data set to train the dispatcher preference model and obtain an updated dispatcher preference model.

[0209] It should be noted that, in addition to updating based on the size of the accumulated operation feedback information, updating may also be performed based on a fixed time period; or updating may be performed based on a combination of the size of the accumulated operation feedback information and the time period.

[0210] In this embodiment, operation feedback information is collected during the execution of the third operation plan, and the task generation model and the dispatcher preference model are updated based on the operation feedback information, thereby further improving the accuracy of the operation plan generation.

[0211] Figure 3This is a flow chart of the model joint training method provided in this application. Based on the above embodiment, this embodiment further explains the training method of the task generation model and the dispatcher preference model used in the above embodiment, such as Figure 3 As shown, the method includes:

[0212] S301: Obtain historical operation records of power grid tasks and construct a first sample training data set.

[0213] In this step, the historical operation record consists of three parts: the operation instruction, the operation plan corresponding to the operation instruction, and the dispatcher feedback information. The sources of historical operation records include but are not limited to: power outages, power supply, equipment maintenance, and load adjustment tasks.

[0214] Optionally, the historical operation records of the power grid task are obtained and the first sample training data set is constructed in the following manner:

[0215] S3011. Collect historical operation records of power grid tasks, extract operation instructions corresponding to different operation tasks, operation plans corresponding to operation instructions, and dispatcher feedback information.

[0216] S3012: Perform data cleaning and normalization processing on the extracted information to obtain a first sample training data set with a unified data format.

[0217] In this step, data cleaning can be done by using regular expressions to remove noise, filling in missing data, and deleting duplicate data. The goal is to align the operation instructions with the operation plans. The data in the first sample training set includes the operation instructions and the operation plans corresponding to the operation instructions.

[0218] S302 : Fine-tune the initial task generation model based on the first sample training data set to obtain a to-be-determined task generation model.

[0219] In this step, the initial task generation model can use a large language model of deep learning to parse the text and generate operation plans.

[0220] For example, a pre-trained language model is used as the initial task generation model to generate the sequence output corresponding to the operation scheme. Specifically:

[0221] S3021, using the loss function shown in Formula 1 as the loss function of the initial task generation model, when the loss value of the loss function is lower than the preset loss function value, the initial task generation model is determined as the task generation model to be determined. Formula 1 is specifically:

[0222]

[0223] Among them, Y true Refers to the actual operation plan corresponding to each operation instruction in the sample training data; Y pred Refers to the predicted operation plan corresponding to each operation instruction; N is the size of a training batch of the initial task generation model; L(θ) is the loss value calculated by the loss function, and θ is the model parameter of the initial task generation model.

[0224] It should be noted that the loss function in this step is to calculate the difference between the predicted output and the actual data. Reinforcement learning can also be used to further emphasize the difference between the predicted output and the actual data for each step corresponding to the operation task in the operation plan.

[0225] For example, the Q-value reward mechanism can be used for reinforcement learning. Specifically, the predicted operation plan generated by the initial task generation model is compared with each step of the actual operation plan. When the comparison results are consistent, the reward value is increased, and when the comparison results are inconsistent, the reward value is reduced. The purpose of maximizing the reward value is to achieve reinforcement learning of the initial task generation model. The Q-value reward mechanism can refer to Formula 2:

[0226]

[0227] Among them, α is the learning factor, γ is the discount factor; Q(S,A) represents the expected cumulative reward of executing step A under the current state S, also called Q value; R refers to the immediate reward, which represents the immediate reward obtained from the environment after specifying step A; max a' Q(S', a') refers to the maximum Q value corresponding to all possible steps a' in the next state S'; R+max a' Q(S', a') represents the total expected reward value of the current step A.

[0228] S3022. Based on the loss value calculated by the loss function, when the loss value is higher than the preset loss function value, the model parameter θ is adjusted through the back propagation algorithm, thereby gradually learning the specifications and formats corresponding to the operation plan of the power grid. The model parameter θ follows the standard gradient descent formula, as shown in Formula 3:

[0229]

[0230] Wherein, η refers to the learning rate of the initial task generation model, L is the loss value calculated in step S3021, and θ is the model parameter.

[0231] S303: Fix the model parameters of the task generation model to be determined, and generate a second sample training data set based on the task generation model to be determined and the first sample training data set.

[0232] In this step, a possible implementation method of generating the second sample data set is:

[0233] S3031. For each operation instruction in the first sample data set, generate a plurality of candidate operation plans using a to-be-determined task generation model with fixed parameters.

[0234] S3032. Perform logic verification on multiple candidate operation plans using preset verification rules, and evaluate the execution efficiency of each candidate operation plan.

[0235] S3033. Determine the dispatcher feedback information corresponding to each operation instruction in the first sample data set based on the historical operation records, label each candidate operation plan based on the dispatcher feedback information, and determine a preference label for each candidate plan.

[0236] In this step, the preference label of the candidate operation plan refers to the dispatcher's selection intention for the plan. Labels such as optimal plan and suboptimal plan can be used as the preference label of the candidate operation plan.

[0237] S3034: Construct a second training sample data set based on multiple candidate operation plans and the preference label corresponding to each candidate operation plan.

[0238] S304: Perform comparative training on the initial dispatcher preference model based on the second sample training data set to obtain a dispatcher preference model to be determined.

[0239] In this step, the initial dispatcher preference model can adopt a neural network model with discriminative function.

[0240] Optionally, a possible implementation method of obtaining the dispatcher preference model to be determined through comparative training is:

[0241] S3041. Select different candidate operation plans from the second sample training data set to construct positive and negative sample pairs.

[0242] S3042. Perform comparative learning training based on positive and negative sample pairs, and calculate the loss function value of the initial dispatcher preference model.

[0243] S3043. When the loss function value reaches the model convergence condition, the initial dispatcher preference model is determined as the dispatcher preference model to be determined.

[0244] S305. Calculate a joint loss function value based on the to-be-determined task generation model and the to-be-determined dispatcher preference model.

[0245] In this step, the calculation of the joint loss function value requires model optimization training based on the task generation model to be determined, the dispatcher preference model to be determined, the first sample training data set, and the second sample training data set.

[0246] For example, one possible implementation method for calculating the two loss function values ​​is:

[0247] S3051. Calculate the loss value of task generation based on the first sample training data set and the task generation model to be determined.

[0248] S3052. Calculate the loss value of preference matching based on the second sample training data set and the dispatcher preference model to be determined.

[0249] S3053. Perform a weighted summation on the loss value generated by the task and the loss value of the preference matching to calculate a joint loss function value.

[0250] S306. When the joint loss function value is lower than the preset loss function value, the task generation model to be determined is determined as the target task generation model, and the dispatcher preference model to be determined is determined as the target preference model.

[0251] In this step, when the joint loss function value is lower than the preset loss function value, the nitrogen concept model is received and the target task generation model and the target dispatcher preference model are output. When the joint loss function value is greater than or equal to the preset loss function value, the weighted calculation parameters of the joint loss function are adjusted and the model training is re-performed. The target task generation model and target dispatcher preference model generated in this step need to be deployed in the power grid system used to implement task scheduling and task generation, and support real-time task generation and optimization.

[0252] It should be noted that in addition to using the joint loss function to train the initial task generation model and the initial dispatcher preference model, the dispatcher preference model can also be used as a preference weighted calculation module, and the preference weighted mechanism of the preference weighted calculation module can be used to train the initial task generation model.

[0253] For example, the method of using the preference weighting mechanism to train the model can be:

[0254] f1. Generate a sample training dataset based on historical operation records.

[0255] In this step, the sample training data set includes: operation instructions, operation plans and corresponding dispatcher feedback information.

[0256] f2. Model training is performed based on the sample training data set. After each operation plan is generated by the initial task generation model, the weighted value of the preference feedback is calculated based on the dispatcher feedback information corresponding to the operation plan using the weight formula shown in Formula 3. Specifically, it is:

[0257]

[0258] Among them, F i is the feedback value of operation task i, which indicates the dispatcher’s satisfaction with the task. i is the execution time of operation task i, and long-time tasks are given lower weights because they may affect the real-time performance of the system. i is the priority or importance of operation task i. λ1, λ2, λ3 are the dispatcher preference weight coefficients, which are obtained through historical data learning. i is the weighted value of the preference feedback for the i-th operation task.

[0259] f3, the initial task generation model is adjusted according to the preference weight value, and the training process is optimized by adjusting the loss function; the loss function is shown in Formula 5:

[0260]

[0261] Among them, L pref Refers to the loss function value corresponding to preference learning, α refers to the standard loss function corresponding to the initial task generation model; W i is the weighted value of the preference feedback of the i-th operation task; Y i pred , Y i true are the predicted task steps and actual task steps of the i-th operation task, respectively.

[0262] In this embodiment, by jointly training the task generation model and the dispatcher preference model, the accuracy of operation plan generation is improved while ensuring the matching degree between the operation plan and the dispatcher information, thereby further improving the task generation efficiency.

[0263] Figure 4 Schematic diagram of the process of generating the grid task operation plan provided in this application Figure 3 ,like Figure 4 As shown, the method includes:

[0264] A1. Obtain input power grid task operation instructions.

[0265] A2. Use the task generation model to parse the operation instructions and obtain the operation plan corresponding to the operation instructions.

[0266] A3. Perform structured analysis on the operation plan and convert the operation plan output by the model into a readable structured plan.

[0267] A4. Use preset verification rules to perform logical verification on the structured solution.

[0268] A5. When the logical verification of the structural scheme fails, a corresponding correction scheme is generated.

[0269] A6. Modify the structured plan based on the revised plan to obtain a preliminary optimized plan.

[0270] A7. Input the preliminary optimization plan and the dispatcher information corresponding to the operation instruction into the dispatcher preference model for optimization processing to obtain the target optimization plan.

[0271] A8. Push the target optimization plan to the dispatcher and collect operational feedback information generated during the execution of the target optimization plan.

[0272] A9. Further optimize the task generation model and dispatcher preference model through incremental learning based on operation feedback information.

[0273] Figure 5 A schematic diagram of the structure of the device for generating the power grid task operation plan provided in this application is shown as follows: Figure 5 As shown, the device for generating a power grid task operation plan provided in this embodiment includes:

[0274] The acquisition module 501 is used to respond to the dispatcher's operation instruction, input the operation instruction into the pre-trained task generation model, and obtain a first operation plan corresponding to the operation instruction.

[0275] The first processing module 502 is configured to perform a logic check on the first operation plan based on a preset check rule to obtain a second operation plan.

[0276] The second processing module 503 is used to input the second operation plan and the dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation plan.

[0277] Among them, the pre-trained task generation model and dispatcher preference model are both trained based on the historical operation records of power grid tasks.

[0278] In a possible implementation, the first processing module 502 is further configured to:

[0279] Analyze the device operation sequence, operation task execution order, and operation task execution time in the first operation plan.

[0280] Based on the preset verification rules, the device operation sequence, the operation task execution order, and the logical conflicts and time conflicts in the operation task execution time are detected to obtain the verification results.

[0281] When the verification result indicates that the verification fails, the first operation plan is repaired based on the verification result to obtain a second operation plan.

[0282] In a possible implementation, the second processing module 503 is further configured to:

[0283] A solution feature is extracted based on the second operation solution.

[0284] Extract dispatcher features based on dispatcher information.

[0285] The scheme characteristics and dispatcher characteristics are input into the dispatcher preference model to obtain the optimization strategy for the second operation scheme.

[0286] The second operation plan is optimized based on the optimization strategy to obtain a third operation plan.

[0287] In a possible implementation, the apparatus further includes a third processing module 504, configured to:

[0288] The third operation plan is pushed to the dispatcher's corresponding device terminal, and the dispatcher is instructed to confirm and execute the plan.

[0289] Collect the dispatcher's operational feedback information on the third operation plan.

[0290] Based on the operation feedback information, the task generation model and the dispatcher preference model are incrementally learned online to obtain an updated task generation model and an updated dispatcher preference model.

[0291] In a possible implementation, the third processing module 504 is further configured to:

[0292] In response to the grid state sudden change request, the currently executed third operation plan is interrupted, and real-time operation state data corresponding to the grid task is acquired.

[0293] The unexecuted operation tasks in the third operation plan are adjusted based on the real-time operation status data to obtain a fourth operation plan.

[0294] In a possible implementation, the third processing module 504 is further configured to:

[0295] The operation feedback information is stored in the incremental database, and the size of the accumulated operation feedback information in the incremental database is calculated.

[0296] When the size of the accumulated operation feedback information in the incremental database reaches a preset number, online incremental learning is performed on the task generation model and the dispatcher preference model based on the accumulated operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, and the incremental database is cleared.

[0297] In a possible implementation, the acquisition module 501 is further configured to:

[0298] Obtain historical operation records of power grid tasks and construct the first sample training dataset.

[0299] The initial task generation model is fine-tuned based on the first sample training data set to obtain a task generation model to be determined.

[0300] The model parameters of the to-be-determined task generation model are fixed, and a second sample training data set is generated based on the to-be-determined task generation model and the first sample training data set.

[0301] The initial dispatcher preference model is compared and trained based on the second sample training data set to obtain the dispatcher preference model to be determined.

[0302] The joint loss function value is calculated based on the task generation model to be determined and the dispatcher preference model to be determined.

[0303] When the joint loss function value is lower than the preset loss function value, the task generation model to be determined is determined as the target task generation model, and the dispatcher preference model to be determined is determined as the target preference model.

[0304] The device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0305] Figure 6 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 6 As shown, the electronic device provided in this embodiment includes: at least one processor 601 and a memory 602. Optionally, the device further includes a communication component 603. The processor 601, the memory 602, and the communication component 603 are connected via a bus 604.

[0306] In a specific implementation process, at least one processor 601 executes computer-executable instructions stored in the memory 602 , so that at least one processor 601 executes the above-mentioned method or method for generating a power grid task operation plan.

[0307] The specific implementation process of the processor 601 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0308] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0309] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage.

[0310] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0311] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0312] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0313] The readable storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0314] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium may be an integral part of the processor. The processor and the readable storage medium may reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium may reside in a device as discrete components.

[0315] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, whether electrical, mechanical, or otherwise, through some interface.

[0316] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0317] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0318] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0319] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0320] Finally, it should be noted that those skilled in the art will readily identify other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. The present invention is not limited to the precise structure described above and illustrated in the accompanying drawings, and various modifications and variations may be made without departing from the scope thereof. The scope of the present invention is limited solely by the appended claims.

Claims

1. A method for generating a power grid task operation plan, characterized in that: include: In response to the dispatcher's operation instruction, the operation instruction is input into a pre-trained task generation model to obtain a first operation plan corresponding to the operation instruction; Performing a logic check on the first operation plan based on a preset verification rule to obtain a second operation plan; Inputting the second operation plan and the dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation plan; Among them, the pre-trained task generation model and dispatcher preference model are both trained based on the historical operation records of power grid tasks.

2. The method according to claim 1, characterized in that The performing a logic check on the first operation plan based on a preset check rule to obtain a second operation plan includes: Analyze the device operation sequence, operation task execution order, and operation task execution time in the first operation plan; Detecting the device operation sequence, the operation task execution order, and the logical conflicts and time conflicts in the operation task execution time based on the preset verification rules to obtain a verification result; When the verification result indicates that the verification fails, the first operation plan is repaired based on the verification result to obtain the second operation plan.

3. The method according to claim 2, characterized in that The second operation plan and the dispatcher information corresponding to the operation instruction are input into a pre-trained dispatcher preference model to obtain an optimized third operation plan, including: extracting scheme features based on the second operation scheme; Extracting dispatcher features based on the dispatcher information; Inputting the scheme characteristics and the dispatcher characteristics into the dispatcher preference model to obtain an optimization strategy for the second operation scheme; The second operation plan is optimized based on the optimization strategy to obtain the third operation plan.

4. The method according to claim 1, wherein After the optimized third operation plan, the method further includes: Pushing the third operation plan to the device terminal corresponding to the dispatcher, and instructing the dispatcher to confirm and execute the plan; Collecting operational feedback information of the dispatcher on the third operational plan; Based on the operation feedback information, online incremental learning is performed on the task generation model and the dispatcher preference model to obtain an updated task generation model and an updated dispatcher preference model.

5. The method according to claim 4, characterized in that The method further comprises: In response to the grid state mutation request, interrupting the currently executed third operation plan and acquiring real-time operation state data corresponding to the grid task; The unexecuted operation tasks in the third operation plan are adjusted based on the real-time operation status data to obtain a fourth operation plan.

6. The method according to claim 4, characterized in that The online incremental learning of the task generation model and the dispatcher preference model based on the operation feedback information to obtain an updated task generation model and an updated dispatcher preference model includes: Storing the operation feedback information in an incremental database, and calculating the size of the accumulated operation feedback information in the incremental database; When the size of the accumulated operation feedback information in the incremental database reaches a preset amount, online incremental learning is performed on the task generation model and the dispatcher preference model based on the accumulated operation feedback information to obtain an updated task generation model and an updated dispatcher preference model, and the incremental database is cleared.

7. The method according to any one of claims 1 to 6, characterized in that Before inputting the operation instruction into the pre-trained task generation model in response to the dispatcher's operation instruction to obtain the first operation plan corresponding to the operation instruction, the method further includes: Obtain historical operation records of power grid tasks and construct the first sample training data set; Fine-tune the initial task generation model based on the first sample training data set to obtain a task generation model to be determined; Fixing the model parameters of the to-be-determined task generation model, and generating a second sample training data set based on the to-be-determined task generation model and the first sample training data set; Performing comparative training on the initial dispatcher preference model based on the second sample training data set to obtain a dispatcher preference model to be determined; Calculating a joint loss function value based on the to-be-determined task generation model and the to-be-determined dispatcher preference model; When the joint loss function value is lower than the preset loss function value, the task generation model to be determined is determined as the target task generation model, and the dispatcher preference model to be determined is determined as the target preference model.

8. A device for generating a power grid task operation plan, characterized in that: include: an acquisition module, configured to respond to an operation instruction of the dispatcher, input the operation instruction into a pre-trained task generation model, and obtain a first operation plan corresponding to the operation instruction; A first processing module, configured to perform a logic check on the first operation plan based on a preset check rule to obtain a second operation plan; a second processing module, configured to input the second operation plan and the dispatcher information corresponding to the operation instruction into a pre-trained dispatcher preference model to obtain an optimized third operation plan; Among them, the pre-trained task generation model and dispatcher preference model are both trained based on the historical operation records of power grid tasks.

9. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

11. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 7 when executed by a processor.

Images