Power distribution network load transfer strategy intelligent recommendation method, intelligent control method and device

Abstract

The invention discloses a power distribution network load transfer strategy intelligent recommendation method, intelligent control method and device, and the method comprises the steps: enabling a large model to receive an input request of a user, enabling the large model to recognize a scene of the input request, and extracting an intention scene; performing entity identification on the input request; extracting parameter information of the transfer constraint and the transfer target, obtaining complete parameter information through multiple rounds of dialogues, converting the complete parameter information into mechanism model request parameters, sending a request to the mechanism model, and receiving multiple candidate transfer strategies and evaluation indexes returned by the mechanism model; and reading a historical wheeling strategy and a strategy constraint standard through the RAG to determine a group of evaluation rules, evaluating each candidate wheeling strategy by adopting a thinking chain mode, screening out the candidate wheeling strategies meeting conditions, sorting the candidate wheeling strategies, determining an optimal wheeling strategy, and outputting the optimal wheeling strategy. The method can be suitable for a complex large-scale power distribution system to quickly and accurately generate the power distribution network load transfer strategy.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to an intelligent recommendation method, an intelligent control method and a device for a load transfer strategy of a distribution network. Background Art

[0002] Under the current distribution network management business model of the power system, dispatching business personnel usually need to comprehensively consider the topology, measurement and source-load prediction results, rely on expert experience to arrange the day-ahead plan, analyze the operating status trend, and then formulate load transfer strategy plans for planned maintenance, heavy overload, fault anomalies and other events. This process often requires multiple rounds of verification and evaluation summary. However, this process will have many problems such as cumbersome multi-source information collection, strategy plans need to rely on experience, long ticket writing time, and difficult knowledge induction and refinement. The degree of intelligence and execution efficiency are low. With the rapid increase in the scale of multi-loads in the new distribution network and its gradual development towards activeness, the complexity of the source-load uncertainty and complex and changeable topology faced by the distribution network dispatching and operation business will be further increased. The traditional manual preparation of load transfer strategy plans is difficult to adapt to the current real-time and accuracy requirements.

[0003] In order to realize the intelligent load transfer, some practitioners have proposed to use traditional optimization algorithms, deep reinforcement learning or large language models (LLM) to assist in the intelligent generation of distribution network load transfer strategy solutions. However, the above methods are only applicable to simple scenarios. In the operation mode of complex large-scale distribution systems, the accuracy and reliability are not high, which cannot meet the actual needs and have many limitations in practical application. For example, traditional optimization algorithms perform well in small-scale power systems, but they are prone to explosive growth in dimensions in large-scale complex power systems, resulting in the problem of non-convergence; deep reinforcement learning algorithms are prone to large solution spaces, too many calculations, and are prone to falling into local optimal or non-convergence situations. In the existing technology, the large language model is a general model that lacks professional knowledge in the power field, and therefore has limited understanding and analysis capabilities for complex power scenarios. Even if a large amount of text data such as the working data of the power system and the power system regulation is used, the transfer strategy solution generated by relying solely on the knowledge base and contextual logical reasoning of the large model is often unreliable. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a distribution network load transfer strategy intelligent recommendation method, intelligent control method and device with simple implementation method, low cost, high accuracy and reliability, which can be applied to quickly and accurately generate distribution network load transfer strategies under complex large-scale distribution system operation modes, and realize intelligent recommendation of load transfer strategies.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is: A method for intelligently recommending load transfer strategies in a distribution network, comprising the following steps: Request input: The large model receives the user's input request, which includes the distribution network load transfer scenario information, power grid equipment entity information, and transfer constraint and transfer target parameter information required by the user; Intention scenario recognition: The large model recognizes the distribution network load transfer scenario required by the received input request and extracts the user's intention scenario; Power grid entity recognition: The large model performs entity recognition on the received input request and extracts the entity name of the power grid equipment; Dynamic call of mechanism model: The big model extracts the parameter information of the transfer constraints and transfer targets from the input request, and obtains the complete parameter information of the transfer constraints and transfer targets through multiple rounds of dialogue. The extracted intention scenario, grid equipment entity name and obtained parameter information are converted into mechanism model request parameters and a request is sent to the mechanism model to call the mechanism model to calculate the transfer strategy. The big model receives multiple candidate transfer strategies and evaluation index data of each candidate transfer strategy returned by the mechanism model. Evaluation of candidate power transfer strategies: RAG is used to read historical power transfer strategies and strategy constraint standards to determine a set of evaluation rules. A chain of thought approach is used to evaluate each candidate power transfer strategy under each evaluation rule. The candidate power transfer strategies that meet the preset conditions are screened out and ranked according to the evaluation results. The optimal power transfer strategy is determined based on the ranking results and output.

[0006] Furthermore, the intended scenarios include any one or more of a main transformer heavy overload scenario, a feeder heavy overload scenario, a fault or maintenance scenario, a power supply task reading scenario, a power supply transfer plan scenario, and a power supply task optimization scenario.

[0007] Furthermore, the load transfer scenario of the distribution network required to be applied to the received input request is identified, and the user's intended scenario is extracted, including: Training the intent recognition model: pre-training the intent recognition model using an intent recognition data set, wherein the intent recognition data set includes multiple types of conversations corresponding to different distribution network load transfer scenarios; Candidate intent recognition: input the current input request into the trained intent recognition model for inference, and derive multiple candidate intent scenarios; Retrieval enhancement: providing the identified candidate intent scenarios as examples in the prompt words to the LLM; Model reasoning: embed retrieval enhancement into the prompt word template to obtain the prompt word, input the request to construct a complete prompt word, input the complete prompt word into the large model for reasoning, and obtain the final intention scenario.

[0008] Further, the performing entity recognition on the received input request and extracting the entity name of the power grid equipment includes: Constructing a knowledge base of power grid equipment entities: using a database containing information on various types of equipment entities in the distribution network to construct a power grid entity SQL vector knowledge base and a RAG vector database in the large model, wherein the RAG vector database is a textual knowledge base of power grid entities; Entity name query: extract the input entity name from the input request, generate an SQL statement based on the power grid entity SQL vector knowledge base, and query the entity information containing the input entity name based on the generated SQL statement. If no matching item is found, recall the results whose vector similarity exceeds the preset threshold in the RAG vector database as context, use the large model to determine the matching entity name, and output the queried entity name.

[0009] Furthermore, it also includes regularly learning entity name rewriting rules from the sample library. When an entity name is queried, the entity name is rewritten according to the learned entity name rewriting rules to be output as the final queried entity name; when no match is found and the use of the large model to determine the matching entity name fails, the large model is used to prompt the retrieval failure and guide the re-verification of the input entity name.

[0010] Furthermore, the thinking chain method is used to evaluate each candidate power transfer strategy under each evaluation rule, and the candidate power transfer strategies that meet the preset conditions are screened out and sorted according to the evaluation results. Based on the sorting results, the optimal power transfer strategy is determined and output including: Determination and adjustment of evaluation rules: Read historical transfer strategies and strategy constraint standards through RAG to determine a set of evaluation rules, and adjust the order and weight of each evaluation rule according to the indicators to be concerned, to form an evaluation rule list; Traversing the evaluation rule screening scheme: The large model uses prompt words to define the thinking process based on the thinking chain method, removes each evaluation rule from the evaluation rule list in turn, and uses the removed evaluation rules to evaluate each candidate power transfer strategy until the evaluation results of all candidate power transfer strategies under all evaluation rules are obtained, and the candidate power transfer strategies that meet the preset conditions are screened out and sorted according to the evaluation results; Intelligent recommendation result output: The large model outputs the recommended results of the power transfer strategy based on the prompt word template. The recommended results include the optimal power transfer strategy and the thinking process.

[0011] Furthermore, the method further includes a prompt word optimization step, wherein the prompt word optimization step includes: Conversation history analysis: Collect historical conversations in which current prompts interact with the big model in the distribution network load transfer scenario, and analyze the collected historical conversations to identify the directions in which prompts need to be optimized; Optimization plan generation: Generate multiple optimization plans based on the analysis results of historical conversations. The optimization plans include one or more of adding new instructions, modifying current instructions, and adjusting the order of instructions. The final optimization plan is selected from various optimization plans. Optimization plan integration: The final selected optimization plan is integrated into the original prompt to form an integrated optimization plan; The conversation history analysis, the optimization solution generation and the optimization solution integration are continuously performed until a preset iteration condition is reached to obtain an optimized prompt word.

[0012] A method for intelligently controlling a load transfer strategy in a distribution network, comprising the following steps: Determine the optimal supply transfer strategy according to the above method; A plurality of potential task scheduling plans are generated according to the tasks to be executed in the optimal transfer strategy, and an action tree is constructed according to each potential task scheduling plan. The final task scheduling plan is decided according to the action tree to control the execution of tasks according to the decided task scheduling plan. If two task scheduling plans in the action tree have the same tasks, they are aggregated into one branch, and different tasks form different branches.

[0013] Furthermore, the steps of generating a plurality of potential task scheduling plans according to the tasks to be executed in the optimal transfer strategy, constructing an action tree according to each potential task scheduling plan, and determining the final task scheduling plan according to the action tree to control the execution of tasks according to the determined task scheduling plan include: Task scheduling sampling: Obtain various types of task scheduling to form a preliminary task scheduling execution plan, and use the big model to generate multiple potential task scheduling plans based on the global information of the task and the preliminary task scheduling execution plan; Action tree construction: Take the load transfer entry task as the root node of the tree, and compare the potential task scheduling plans in sequence starting from the first task. When two task scheduling plans have the same tasks, they are aggregated into one branch, and different tasks form different branches, until all task scheduling plans are built into the tree to form a complete action tree; Tree-based decision: select a node from any branch in the tree for execution in turn. If the execution result of the executed node is valid, proceed to the next node in turn. If the task of the executed node fails, it is marked as an invalid node and traced back to the nearest valid node. Then select other branches for execution until the tasks of all nodes on the target branch are successfully executed. The task scheduling plan of the target branch is determined as the final decision.

[0014] A load transfer intelligent device for a distribution network comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0015] Compared with the prior art, the advantages of the present invention are: 1. The present invention uses a large model to analyze the user's input request, identify the intended scenario, the entity name of the power grid equipment, and the parameter information of the transfer constraints and transfer targets, convert them into mechanism model request parameters to dynamically call the mechanism model, and then evaluate the candidate transfer strategies fed back by the mechanism model. It can determine the optimal transfer strategy by combining the mechanism model and the large model, realize the intelligent generation of distribution network load transfer strategies under complex scenarios, improve the automation and intelligence of the distribution network load transfer process, and at the same time improve the accuracy and reliability of load transfer strategy generation.

[0016] 2. After determining the optimal transfer strategy, the present invention further adopts a task dynamic planning method based on Tree-Planner. By first generating a variety of potential task scheduling plans according to the tasks to be executed in the optimal transfer strategy, and then constructing an action tree, the task scheduling plan is decided based on the action tree to control the execution of tasks. This can effectively improve the automation and intelligence of the distribution network load transfer process, while improving the flexibility and efficiency of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the implementation flow of the intelligent recommendation method for load transfer strategy in distribution network embodiment 1 of the present invention.

[0018] Figure 2 It is a schematic diagram of the implementation process of intention scene recognition in Example 1 of the present invention.

[0019] Figure 3 It is a schematic diagram of the implementation process of power grid equipment entity identification in Example 1 of the present invention.

[0020] Figure 4 It is a schematic diagram of the implementation process of the dynamic call of the mechanism model in Example 1 of the present invention.

[0021] Figure 5 It is a schematic diagram of the implementation process of the candidate supply transfer strategy indicator evaluation in Example 1 of the present invention.

[0022] Figure 6 It is a schematic diagram of the implementation process of the prompt word optimization in Example 1 of the present invention.

[0023] Figure 7 It is a schematic diagram of the implementation flow of the intelligent control method for load transfer in the distribution network in Example 2 of the present invention.

[0024] Figure 8 It is a schematic diagram of the implementation process of dynamic task planning in Example 2 of the present invention. DETAILED DESCRIPTION

[0025] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0026] For ease of understanding, the relevant technical background involved in this embodiment is first introduced by way of example.

[0027] Big model: The big model refers to the professional big model (L1) in the power field that is strengthened by adding power forecast training to the open source big model (L0) using Tranformer technology. By introducing professional corpus in the power field for training based on the open source big language model, the model's understanding of power-specific scenarios can be improved, providing support for smart load transfer decisions.

[0028] Prompt words: Prompt words are to package user input into a fixed context and send it to the big model. This can improve the big model's ability to express professional knowledge, reduce ambiguity and misunderstanding, and can even be used to specify the big model to achieve specific tasks.

[0029] RAG: RAG (Retrieval-Augmented Generation) technology is an artificial intelligence method that combines information retrieval and text generation. It retrieves relevant information from a large amount of data and then uses the information to assist in generating more accurate and rich answers. The core of RAG technology lies in three steps: retrieval, utilization, and generation. First, the model retrieves relevant documents or information fragments based on the given question or task; then, analyzes and extracts key information from the retrieval results; finally, combines the information and context to generate fluent and relevant text output.

[0030] Chain of Thought (CoT): Chain of Thought CoT technology is a prompting strategy used to improve the performance of large models in complex reasoning tasks. It imitates the human thinking process by showing intermediate reasoning steps. This technology allows the model to break down complex problems into multiple sub-problems and reason step by step to finally come up with a solution. The core of the Chain of Thought CoT technology is to provide visibility of the reasoning process, improve the interpretability of model decisions, and help the model perform complex logical reasoning. In the development and application stage, based on the Chain of Thought technology, the thinking process can be set through prompt words, allowing the large model to think in the set way.

[0031] Agent: A task process implemented by technologies such as large models, RAG, prompt words, and thought chains. Another agent can be nested in an agent process. The agents used for nesting are called sub-processes or sub-agents to build multi-agents.

[0032] Task orchestration: The process of arranging and scheduling multiple interrelated or dependent tasks in an orderly manner to ensure that they are completed in the predetermined order.

[0033] Load transfer: Load transfer is a power supply strategy. When a certain area or user in the power system cannot be powered normally due to a fault, maintenance or other reasons, the power grid system will transfer power to other areas or users to ensure the stability and continuity of the overall power supply. Load transfer includes two situations: one is to temporarily switch the power supply to other power sources for users who should be powered by a certain power grid; the other is to transfer part of the load to other transmission lines or substations to reduce the original power supply pressure.

[0034] Tasks in load transfer scenarios usually involve multi-level decision-making and analysis processes, such as topology identification, overload warning, load transfer plan generation, etc. Such tasks have complex logical chains, and thinking chain technology can decompose complex tasks into executable logical steps, which helps the load transfer agent to gradually complete the task objectives, improve execution efficiency and decision-making quality, and combine with intention recognition, entity recognition, dynamic call of mechanism model and transfer plan indicator evaluation, it can greatly enhance the application ability of the agent in complex power dispatching scenarios, quickly generate accurate and reliable optimal load transfer strategies, and improve the reliability of load transfer strategy generation.

[0035] The present invention introduces a large model (L1) for the power field after training with power corpus based on an open source large language model (L0), based on multi-agent and thinking chain technology, combined with the characteristics of load transfer scenarios such as planned maintenance, heavy overload, and fault handling of distribution networks, and analyzes the user's input request using a large model to identify the intended scenario, the entity name of the power grid equipment, and the parameter information of the transfer constraints and transfer targets, and uses the above information to convert it into a mechanism model request parameter to dynamically call the mechanism model, and then evaluates each candidate transfer strategy fed back by the mechanism model, and can determine the optimal transfer strategy by combining the mechanism model and the large model, realize the intelligent generation of distribution network load transfer strategies under complex scenarios, improve the automation and intelligence of the distribution network load transfer process, and at the same time improve the accuracy and reliability of load transfer strategy generation.

[0036] The present invention will be further described below in conjunction with specific embodiments.

[0037] Embodiment 1: like Figure 1As shown, the steps of the intelligent recommendation method for load transfer strategy of distribution network in this embodiment include: Step S101. Request input: The large model receives an input request from a user, where the input request includes the distribution network load transfer scenario information, grid equipment entity information, and transfer constraints and transfer target parameter information required by the user.

[0038] The user can send an input request to the large model through a program or external input. The input request contains the distribution network load transfer scenario information, the power grid equipment entity information, and the parameter information of the transfer constraints and transfer targets, etc. Other types of information can also be added according to actual needs. The distribution network load transfer scenario is the scenario where load transfer is required, including the main transformer heavy overload scenario process, feeder heavy overload scenario, fault or maintenance scenario, reading power protection task scenario, power protection transfer plan scenario, power protection task optimization scenario, opening diagram task and other scenarios. The power grid equipment entity information is the information of the equipment entity involved in the required application scenario. For example, if it is a main transformer heavy overload scenario, the equipment entity is the main transformer entity. If it is a feeder heavy overload scenario, the equipment entity is the target feeder and the fault feeder. If it is a fault or maintenance scenario, the equipment entity is the target feeder. The parameter information of the transfer constraints and transfer targets is the constraints that the transfer strategy needs to satisfy and the parameters involved in the targets that the transfer strategy needs to achieve. For example, the transfer constraints include information such as the current load rate, the upper limit of the load rate, the number of transfers, etc. The transfer target parameters include parameters corresponding to targets such as minimum switch action, minimum transfer load, line load balance, and main transformer load balance.

[0039] Taking the input request as: "The first loop of the machine high-speed transmission line is overloaded, please provide a transfer plan" as an example, the subsequent large model first identifies the intention of feeder overload and the need for transfer, and determines it as a feeder overload scenario; further, the large model begins to identify the target feeder and the faulty feeder (the faulty feeder cannot be used to receive the transferred load), and identifies the target feeder as the first loop of the machine high-speed transmission line, and there is no faulty feeder; further, the large model begins to identify the transfer target and transfer constraints. Since the input request does not contain the above information, the large model uses the preset scenario parameters given by the prompt word as the default value. This process supports multiple rounds of dialogue. When the user supplements the input: "The maximum load rate is 83%, please regenerate the transfer plan", the large model will analyze it in combination with the historical input "The first loop of the machine high-speed transmission line is overloaded, please provide a transfer plan". In the stage of identifying the transfer target and transfer constraint, the parameter identified is 83%, and other parameters remain unchanged by default.

[0040] Step S102. Intention scenario identification: The large model identifies the distribution network load transfer scenario required for the received input request and extracts the user's intention scenario.

[0041] Intent recognition is to use a large model to analyze user input, identify the user's interactive intent, format it and output it, so that the intelligent agent can call the corresponding sub-agent to respond to the user's needs. This embodiment uses a large model to analyze the user's input request to infer the category of distribution network load transfer scenario required by the user. All intention recognition scenarios for distribution network load transfer include but are not limited to main transformer overload scenario processes, feeder overload scenarios, fault or maintenance scenarios, reading power supply task scenarios, power supply transfer plan scenarios, power supply task optimization scenarios, opening diagram tasks, etc., and you can also configure a general dialogue task process for replying when none of the above scenarios match.

[0042] There are many intention scenarios and the similarity between scenarios is high. The recognition accuracy of the traditional intention recognition method based on keyword matching is low. This embodiment realizes the recognition of intention scenarios by using a two-level intention recognition structure method that combines a fine-tuned small model with a retrieval enhancement mechanism. First, a single-round intention recognition small model is trained using a single-round data set. Then, a retrieval enhancement mechanism is used to construct prompts by retrieving single-round examples with semantic similarities to multi-round test samples, thereby promoting context learning. Finally, the prompts are used to guide LLM to reason to identify the intention in the conversation, which can effectively improve the accuracy of intelligent agent intention recognition. This embodiment realizes intention scenario recognition by combining a fine-tuned small model with a retrieval enhancement mechanism, which can deeply understand the terminology and tasks of the power industry, enabling the system to autonomously identify various tasks, thereby reducing manual intervention and improving the degree of automation.

[0043] As an optional implementation, Figure 2 As shown, the large model can identify the intention scenario of the distribution network load transfer scenario required by the received input request by taking the following steps: Step S121. Training the intention recognition model: pre-training the intention recognition model using an intention recognition dataset, where the intention recognition dataset contains multiple types of conversations corresponding to different distribution network load transfer scenarios.

[0044] Specifically, the intent recognition model can be constructed by combining a simple label attention model and a HiTiN model. The HiTiN model is a hierarchical text classification model trained on an intent recognition dataset. The label attention model only considers the semantic information of user utterances and label-query attention information, ignoring the hierarchical tree structure in intent recognition. The tree isomorphic network classification structure in HiTiN is integrated into the label attention model to improve the performance of the model. The intent recognition dataset contains all possible conversations in the load transfer scenario of the distribution network and is manually annotated to cover various types of daily dispatcher interaction behaviors.

[0045] Step S122. Candidate intent recognition: Input the current input request into the trained intent recognition model for reasoning, and derive multiple candidate intent scenarios.

[0046] Specifically, the most recent user input and historical conversation records can be combined, and inference can be performed through the single-round intention recognition model in step S121. The inference vector can be probabilized using the softmax function to obtain the probability vector of the intention. Finally, the top three intentions with the highest scores are taken as candidate intentions according to the scores.

[0047] Step S123. Retrieval enhancement: Provide the identified candidate intent scenarios as examples in the prompt words to the LLM.

[0048] Specifically, the candidate intents identified in step S122 can be used as examples in the prompt words to provide a decision basis for the LLM and define a standardized output format, where the prompt words are used to guide and constrain the output of the large model, and the standardized output format converts natural language into intent labels.

[0049] Step S124. Model reasoning: embed the prompt word obtained by retrieval enhancement in the prompt word template, input the request to construct a complete prompt word, input the complete prompt word into the large model for reasoning, and obtain the final intention scenario.

[0050] Specifically, a complete prompt word can first be constructed based on the prompt word template embedded in the prompt word of step S123, the historical conversation record and the most recent user input, and then the complete prompt word is input into the professional large model (L1) combined with the power field for reasoning to identify the intention of the user input and obtain the final user intention.

[0051] Through the above steps, this embodiment can identify the distribution network load transfer scenario required for the input request based on the secondary intent recognition structure to obtain the intent scenario, first use the intent recognition model to preliminarily identify the candidate intent, and then accurately identify the intent based on the large model RAG technology, and combine the interactive multi-round confirmation process to ensure the accuracy of the entity name, which can effectively improve the accuracy and efficiency of intent recognition, thereby enhancing the intent understanding ability of natural language problems in the retrieval-based interactive system in the distribution network load transfer.

[0052] It is understandable that the specific input data of the large model in the above steps can also be configured according to actual needs, such as adding more types of data other than historical conversations as input data to further improve the accuracy of intent recognition, or streamlining some input data to further improve recognition efficiency.

[0053] Step S103. Power grid entity identification: The large model performs entity identification on the received input request and extracts the entity name of the power grid equipment.

[0054] Grid entity recognition is to identify the entity name of grid equipment. In the grid entity recognition task, the grid entity name is not only lengthy, but users tend to use abbreviations, and users may enter non-standard names (aliases, etc.), which makes it difficult to fully match the actual name, thereby affecting the efficiency and accuracy of recognition. This embodiment can effectively improve the accuracy of entity recognition by combining SQL knowledge base recall with text knowledge base recall.

[0055] As an optional implementation, the SQL knowledge base recall is combined with the text knowledge base recall to perform entity recognition on the received input request to obtain the entity name of the power grid equipment. The following steps can be used: Step S131. Constructing a power grid equipment entity knowledge base: using a database containing entity information of various types of distribution network equipment to construct a power grid entity SQL vector knowledge base and a RAG vector database in a large model. The RAG vector database is a power grid entity text knowledge base.

[0056] Specifically, Figure 3As shown, a power grid entity ledger database can be first constructed and connected to the large model platform. By using the vectorized database Milvus, the data table is converted into a point in a high-dimensional vector space so that the subsequent SQL generation model can be efficiently called. In addition, the database content is exported and a vector knowledge base is constructed on the large model platform to form a power grid entity SQL vector knowledge base and a power grid entity text knowledge base (RAG vector database). The power grid entity SQL vector knowledge base is a vector database formed by converting the power grid equipment entity database information into a mathematical vector form. Specifically, this step is performed by using a pre-trained encoder to encode the table name in the power grid equipment entity database, the field name in the table, and the corresponding annotation information of the two and adding them to the vector database. In the execution link, the large model extracts keywords according to the user output, encodes the keywords, and queries them in the vector database, recalls k data with high vector similarity, and then the large model generates a sql statement that can be used to query the power grid entity based on the recalled data, combined with the dependency relationship between the table name and the field name and the user's input. After the sql statement is generated, it is then in the database Execute this statement and obtain the returned database query results, and finally summarize the entity information queried by polishing the output of the large model; the power grid entity text knowledge base is a knowledge base containing texts such as power grid entity names. Specifically, this step first organizes the names and IDs of all entities in the database and exports them as PDF documents, then segments the PDF documents using a text segmentation tool, and then uses an encoder to encode the entity names and IDs and add them to the vector database. In the execution phase, the large model extracts keywords based on user input, encodes the keywords and user input, and then queries them in the vector database, recalls k data with high similarity, and then summarizes the entity information queried by polishing the output of the large model based on the recalled data.

[0057] Step S132. Entity name query: extract the input entity name from the input request, generate an SQL statement based on the power grid entity SQL vector knowledge base, and query the entity information containing the input entity name based on the generated SQL statement. If no matching item is found, recall the results whose vector similarity exceeds the preset threshold in the RAG vector database as context, use the large model to determine the matching entity name, and output the queried entity name.

[0058] Specifically, Figure 3 As shown in the figure, when the user inputs a task, the large model can first extract the entity name input by the user, and then generate an SQL statement based on the SQL vector knowledge base to query the entity information containing the user input; if the query has no matching items, the results with high vector similarity are recalled from the RAG vector database as context, and the large model determines the entity name that is truly fully matched based on the prompt word, and finally returns the queried entity name and entity-related attribute information.

[0059] This embodiment uses the entity recognition that combines the above-mentioned SQL knowledge base recall method and text knowledge base recall method to effectively improve the recognition accuracy of entities such as equipment names in the power system. The streaming task process is used to asynchronously execute task blocks, which can also improve the response speed and increase the efficiency of system resource utilization.

[0060] When all matching methods fail, such as Figure 3 As shown, this embodiment also configures a fallback answer process: when all matching methods fail, the big model will prompt the user that the search has failed and guide the user to re-verify the input entity name. In addition, a feedback mechanism can be integrated to allow users to report recognition errors. When users correct entities to the big model, user input and context data are saved, which can be used to locate process errors and improve the subsequent entity name rewriting module.

[0061] As an optional implementation, in order to further improve the recall success rate, it also includes regularly learning entity name rewriting rules from the sample library, such as Figure 3 As shown, when an entity name is queried, the entity name is rewritten according to the learned entity name rewriting rules to be output as the final queried entity name; when no match is found and the large model fails to match the entity name, the large model prompts the retrieval failure and guides the re-verification of the input entity name.

[0062] Specifically, the entity name rewriting module can be configured, and the module uses the Bert sequence prediction model using natural language processing technology, and uses a scheduled task to regularly learn and apply rewriting rules from the sample library to rewrite the abbreviation or alias entered by the user into a formal entity name, and then accurately identify the power grid entity name through the above steps. By making full use of the sequence transcription ability of the Bert model and the summarization ability of the large model, the pre-trained Bert model is used to realize the conversion ability of entity aliases and entity abbreviations to formal names. At the same time, with the help of the large model to summarize the entity objects queried based on the formal name, combined with user input to achieve accurate identification of the entity, it can further improve the accuracy of entity recognition and realize the fuzzy search ability of entity aliases and entity abbreviations.

[0063] Step S104. Dynamically call the mechanism model: The large model extracts the parameter information of the transfer constraints and transfer targets from the input request, and obtains the complete parameter information of the transfer constraints and transfer targets through multiple rounds of dialogue, converts the extracted intention scenario, grid equipment entity name and the obtained parameter information into the mechanism model request parameters and sends a request to the mechanism model to call the mechanism model to calculate the transfer strategy, and receives multiple candidate transfer strategies returned by the mechanism model and the evaluation index data of each candidate transfer strategy.

[0064] The dynamic invocation mechanism model can give the big model the ability to flexibly call related mechanism models according to the user's execution intention, accurately extract key parameters to construct the request body, and finally calculate the distribution network's transfer plan. The dynamic invocation mechanism model combines multiple rounds of dialogue to allow the big model to continuously deepen its understanding during the interaction with the user, thereby more accurately grasping the user's intentions and needs.

[0065] As an optional implementation, the dynamic invocation of the mechanism model may be performed in the following steps: Step S141: scene recognition and entity extraction.

[0066] Specifically, Figure 4 As shown in the figure, the big model first conducts an in-depth analysis of the user's input, extracts the transfer constraints and transfer target parameters, and identifies the transfer scenario to determine the scenario category to which it belongs. The transfer constraints include information such as the current load rate, the upper limit of the load rate, and the number of transfers. The transfer target parameters include parameters corresponding to the minimum switch action, the minimum transfer load, the line load balance, and the main transformer load balance. The transfer scenarios include main transformer heavy overload, feeder heavy overload, or fault maintenance. In the process of scene recognition, by designing the big model prompt words, the big model can extract and recognize the power grid equipment entities, parameter entity information, and scene categories in the business scenarios and dialogues from the user input. When it is identified as a main transformer heavy overload scenario, the target main transformer name in the input request is extracted, and then the target main transformer entity information is recalled. If it is a feeder heavy overload scenario, the target feeder name and the faulty feeder name in the input request are extracted, and then the target feeder entity and the faulty feeder entity information are recalled. If it is a fault or maintenance scenario, the target feeder name in the input request is extracted, and then the target feeder entity information is recalled. The recalled entity information is combined with the transfer constraints and transfer target parameters and converted into parameters of the mechanism model, which are sent to the mechanism model.

[0067] It can also be configured that when the user's intention is unclear and causes the scene recognition failure or entity extraction failure, the actual error message can be embedded through a prompt word template containing error examples, so that the large model can ask questions about the intention and feedback the error message, and clarify the user's intention after the user updates the input, ensuring the correct identification and extraction of the main transformer or feeder entity that needs to be transferred.

[0068] Step S142: Parameter extraction and context understanding.

[0069] Specifically, in user input, the parameters of the transfer constraints and transfer targets are often not given all at once. The large model can implement multiple rounds of dialogue by introducing historical dialogue information into prompt words, gradually guiding users to provide more information, or using contextual information and default values ​​to infer missing parameters when users fail to provide complete information.

[0070] Step S143: Request body assembly and sending.

[0071] Specifically, after successfully extracting all necessary parameters, the big model can summarize the descriptive text of the parameters required by the output mechanism model, and then convert the descriptive text output by the big model into mechanism model request parameters through text parsing tools, and then initiate a request to the mechanism model through the request tool to calculate the specific transfer plan, and finally obtain the transfer plan list returned by the mechanism model, which includes evaluation index data of the transfer plan, etc.

[0072] Through the above method, this embodiment can fully utilize the mechanism model to dynamically call and accurately obtain multiple candidate power transfer strategies and evaluation index data of each candidate power transfer strategy.

[0073] Step S105. Evaluation of candidate power transfer strategies: Read historical power transfer strategies and strategy constraint standards through RAG to determine a set of evaluation rules, use the thinking chain method to evaluate each candidate power transfer strategy under each evaluation rule, screen out the candidate power transfer strategies that meet the preset conditions and sort them according to the evaluation results, determine the optimal power transfer strategy based on the sorting results and output it.

[0074] The prompt word technology based on L1 and transfer scheme evaluation can improve the load transfer agent's ability to evaluate the scheme, realize the optimal sorting of candidate transfer schemes, and provide an interpretable explanation in combination with input data, mechanism model parameters, and evaluation index system. At the same time, using RAG technology to refer to historical schemes and procedural system standards can provide corresponding optimization strategy guidance.

[0075] The core of the transfer plan indicator evaluation is to receive the list of strategic plans output by the mechanism model (which contains detailed evaluation indicators for each strategic plan), and then integrate key data sources such as plan indicators, user input, historical plans, and strategic constraint standards (such as procedural system standards, etc.), analyze and determine the best plan through the intelligent processing of the big model, and finally generate a summary document containing the best plan, evaluation reasons, related optimization strategies and presentations.

[0076] As an optional implementation, Figure 5 As shown, the following steps can be used to rank the performance of each candidate power transfer strategy according to each candidate power transfer strategy and a preset evaluation rule: Step S151. Determine and adjust the evaluation rules: Read the historical transfer strategy and strategy constraint standards through RAG to determine a set of evaluation rules, and adjust the order and weight of each evaluation rule according to the indicators that need to be paid attention to to form an evaluation rule list.

[0077] Specifically, the RAG technology can be used to read historical transfer strategies and policy constraint standards (such as procedures and system standards) to determine a set of evaluation rules, and then adjust the order and weight of the evaluation rules according to the evaluation requirements and key indicators input by the user. Finally, the large model summarizes and outputs a list object composed of evaluation rules. The above historical transfer strategies can be RAG recall transfer plan evaluation related documents, SQL recall line historical transfer plans, etc.

[0078] Step S152. Traverse the evaluation rule screening scheme: The large model uses prompt words to define the thinking process based on the thinking chain method, removes each evaluation rule from the evaluation rule list in turn, and uses the removed evaluation rules to evaluate each candidate power transfer strategy until the evaluation results of all candidate power transfer strategies under all evaluation rules are obtained, and the candidate power transfer strategies that meet the preset conditions are screened out and sorted according to the evaluation results.

[0079] Specifically, the big model can use the prompt word technology in combination with the adjusted evaluation rules to evaluate the solutions one by one. First, the big model takes out the rules from the rule list in step S151 in order, and then the big model screens out the solutions that meet the conditions according to the rules. The big model uses the thinking chain technology to improve the thinking ability of the big model by defining the thinking process through prompt words, so as to output the solutions that meet the conditions while outputting the thinking process to ensure the accurate understanding and execution of the rules. The process of taking rules and thinking is executed in a loop until all rules are taken out or the optimal solution has been selected, and finally the optimal solution or the optimal solution set is screened out. The optimal solution can be the best solution with the highest score.

[0080] Step S1531. Intelligent recommendation result output: The large model outputs the recommended results of the supply transfer strategy according to the prompt word template. The recommended results include the optimal supply transfer strategy and the thinking process.

[0081] Specifically, after the selection of the solutions is completed, the best solution ID, the thinking process of each step, and the relevant regulations and system texts obtained through the RAG tool query are embedded in the prompt word template through the design of the prompt word template. Finally, the large model generates a logical summary text based on the filled prompt words. The above summary text contains the best solution, thinking process, reference regulations, reference historical solution information, etc. The specific output information can be configured according to actual needs.

[0082] This embodiment combines historical power transfer strategy solutions with user input to form a set of evaluation rules from a large model, and then uses the thinking chain technology to implement solution evaluation. It can evaluate dynamically generated power transfer strategy solutions based on evaluation indicators, so that personalized power transfer strategy solution recommendations and evaluation services can be provided.

[0083] By providing a set of input integrated instruction sets, prompt word engineering can help the large model understand human intentions, guide the large model to generate expected responses, and ensure the relevance, coherence, and accuracy of the responses without fine-tuning model parameters. In the load transfer scenario of the distribution network, optimizing prompt words can help the large model more accurately identify task objectives, optimize decision-making plans, and improve the ability to select the best plan. As an optional implementation, it also includes a prompt word optimization step S106 to optimize the prompt words. Figure 6 As shown, the prompt word optimization step S106 may adopt the following steps: Step S161. Conversation history analysis: Collect historical conversations in which the current prompt interacts with the large model in the distribution network load transfer scenario, and analyze the collected historical conversations to identify directions in which the prompts need to be optimized.

[0084] Specifically, we can first collect the conversation history of the current prompt interacting with the large model (L1) in the distribution network load transfer scenario, which records the performance of the large model (L1) in the power field, such as success and failure. Then, we use the prompt optimization fine-tuning model to analyze the historical conversations, identify the main reasons for the poor performance in the historical conversations, find out the optimization ideas, and get the direction in which the current prompt needs to be optimized.

[0085] Step S162. Optimization scheme generation: Generate multiple optimization schemes based on the analysis results of historical conversations, and select the final optimization scheme from various optimization schemes.

[0086] Specifically, a variety of possible optimization solutions can be generated based on the analysis results of step S161, including adding new instructions, modifying existing instructions, adjusting the order of instructions, etc., and then the system evaluates each optimization solution and selects the one that is most likely to solve the current problem as the final optimization method. Specifically, the chain of thought (COT) technology can be used to gradually solve the optimization solution, including: analyzing each step of the original prompt to determine where the new instruction is added to the original prompt, or whether the instruction at the existing position needs to be replaced; analyzing the relationship between the existing instructions and the new instructions to ensure that the prompt remains consistent after the instruction is updated; retaining key elements to ensure that the meaning of the prompt word remains unchanged.

[0087] Step S163. Optimization scheme integration: integrating the selected final optimization scheme into the original prompt to form an integrated optimization scheme.

[0088] Specifically, the most effective optimization solution selected in step S162 can be incorporated into the original prompt, such as modifying the content of the original prompt, adding new steps to the original prompt, adjusting the order of steps in the original prompt, etc. The original prompt may include instructions such as large model identity definition, user input introduction, dialogue context introduction (optional), other information introduction (optional), rule definition, thinking process (optional), output example (optional), output format definition (optional), etc. Furthermore, the key elements in the original prompt can be retained in the consideration of the new instruction to ensure that the optimized prompt still meets the basic requirements of the task.

[0089] Step S164. Continue to execute step S161 of conversation history analysis, step S162 of optimization solution generation, and step S163 of optimization solution integration until the preset iteration condition is reached to obtain the optimized prompt word.

[0090] Specifically, the process of steps S161 to S163 is continuously repeated, and the quality of the prompt words can be improved in each iteration. During the iteration process, the performance improvement after each iteration is monitored, and the iteration is stopped when the performance improvement is no longer significant or when the iteration requirements are met. In the early stage of the iteration, it may be inclined to adopt the method of adding new instructions, but as the iteration proceeds, more attention may be paid to the refinement of the prompt words. As the number of iterations increases, the quality of the prompt words will gradually improve until a stable state is reached.

[0091] Preferably, RePrompt can be used to implement the above prompt word optimization to help the large model (L1) in the power field to improve and optimize the existing prompt words based on the existing prompt words when processing tasks, while following the principle of minimum modification and ensuring that the structure and content of the original prompt words are retained as much as possible. RePrompt is a prompt optimization method based on interactive action generation. RePrompt combines the elements of interactive generation and feedback loop, and continuously refines and optimizes prompt words through interaction with feedback providers (ReAct). When this embodiment uses RePrompt, it includes analyzing historical conversations, generating optimization plans, integrating instructions, and iterative optimization to gradually improve the quality of prompts, that is, collecting chat history through RePrompt, and then analyzing user feedback, identifying focus and common problems related to specific tasks, for optimizing subsequent generated prompts, and then avoiding overfitting of prompts due to inappropriate data through batch summarization.

[0092] By adopting the above method, this embodiment can combine large model analysis and historical conversation records to achieve semi-automatic prompt word optimization.

[0093] This embodiment constructs the above-mentioned intelligent entity and utilizes the large model of the power field, load transfer task scheduling tools, load transfer mechanism model, intention recognition, entity recognition and indicator evaluation to achieve intelligent decision-making and optimization plan recommendation for load transfer in the distribution network, thereby improving the load transfer response efficiency and accuracy of the power system in maintenance, heavy overload, fault and other scenarios.

[0094] The intelligent recommendation device for load transfer strategy of distribution network in this embodiment includes: A request input module is used for the large model to receive a user's input request, wherein the input request includes the distribution network load transfer scenario information, power grid equipment entity information, and transfer constraint and transfer target parameter information required by the user; The intention scenario recognition module is used for the large model to identify the distribution network load transfer scenario required for the received input request and extract the user's intention scenario; The power grid entity recognition module is used for the large model to perform entity recognition on the received input request and extract the entity name of the power grid equipment; The mechanism model dynamic call module is used for the large model to extract the parameter information of the transfer constraints and transfer targets from the input request, and obtain the complete parameter information of the transfer constraints and transfer targets through multiple rounds of dialogue, convert the extracted intention scenario, the entity name of the power grid equipment and the obtained parameter information into the mechanism model request parameters and send the request to the mechanism model to call the mechanism model to calculate the transfer strategy, and receive the multiple candidate transfer strategies returned by the mechanism model and the evaluation index data of each candidate transfer strategy; The candidate power transfer strategy evaluation module is used to read the historical power transfer strategies and strategy constraint standards through RAG to determine a set of evaluation rules, and use the thinking chain method to evaluate each candidate power transfer strategy under each evaluation rule, screen out the candidate power transfer strategies that meet the preset conditions and sort them according to the evaluation results, and determine the optimal power transfer strategy based on the sorting results and output it.

[0095] The distribution network load transfer strategy intelligent recommendation device of this embodiment corresponds to the above-mentioned distribution network load transfer strategy intelligent recommendation method, and will not be described in detail here.

[0096] Embodiment 2: like Figure 7 As shown, the steps of the intelligent control method for load transfer strategy of the distribution network in this embodiment include: Step S201. Determine the optimal transfer strategy according to the method in Example 1, that is, determine the optimal transfer strategy after sequentially performing intention recognition, entity recognition, dynamic calling of mechanism model, transfer scheme indicator evaluation, and semi-automatic optimization of prompt words (which can be configured according to actual needs); Step S202. Dynamic task planning: Generate multiple potential task scheduling plans based on the tasks to be executed in the optimal transfer strategy, and construct an action tree based on each potential task scheduling plan. Determine the final task scheduling plan based on the action tree to control the execution of tasks according to the decided task scheduling plan. If two task scheduling plans in the action tree have the same tasks, they are aggregated into one branch, and different tasks form different branches.

[0097] The task process is planned in advance through a priori task planning, and the task is executed according to the planned process based on the result of intention recognition. However, the load transfer scenario of the distribution network is complex and changeable, and a priori task planning is difficult to meet the actual application scenario. After determining the optimal transfer strategy, this embodiment adopts a task dynamic planning method based on Tree-Planner (tree planner). By first generating a variety of potential task scheduling plans according to the tasks to be executed in the optimal transfer strategy, and then constructing an action tree, the task scheduling plan is decided based on the action tree to control the execution of tasks. This can effectively improve the automation and intelligence of the load transfer process of the distribution network, while improving the flexibility and efficiency of task execution.

[0098] As an optional implementation, Figure 8 As shown, step S202 of the task dynamic method based on Tree-Planner can adopt the following steps: Step S221. Task scheduling sampling: Obtain various types of task scheduling to form a preliminary task scheduling execution plan, and use the big model to generate multiple potential task scheduling plans based on the global information of the task and the preliminary task scheduling execution plan.

[0099] Specifically, firstly, all possible task execution plans are compiled to form a preliminary task scheduling execution plan, and then the power industry big model is used to generate a series of potential task scheduling plans based on the global information of the task and the preliminary task scheduling plan. The potential task scheduling plan is generated by the big model based on common sense and logical reasoning according to prompts.

[0100] Step S222. Action tree construction: Take the entry task of load transfer as the root node of the tree, and compare the potential task scheduling plans in sequence starting from the first task. When two task scheduling plans have the same tasks, they are aggregated into one branch, and different tasks form different branches, until all task scheduling plans are built into the tree to form a complete action tree.

[0101] Specifically, the load transfer entry task can be used as the root node of the tree, and then all potential task scheduling plans generated in step S221 can be compared in sequence starting from the first task. When two task scheduling plans have the same tasks, they are aggregated into one branch, and different task parts form different branches until all task scheduling plans are constructed into a complete tree. Since some of the potential task scheduling plans are repeated and invalid, they cannot be used directly and need to be further screened in subsequent steps. The above steps can avoid repeated execution of the same task and improve the efficiency of task execution.

[0102] Step S223. Tree-based decision: select a node from any branch in the tree for execution in turn. If the execution result of the executed node is valid, proceed to the next node in turn. If the task of the executed node fails to execute, it is marked as an invalid node and traced back to the nearest valid node, and then other branches are selected for execution until the tasks of all nodes on the target branch are successfully executed, and the task scheduling plan of the target branch is determined as the final decision.

[0103] Specifically, based on the constructed action tree, when executing instructions, first select the first node of any branch in the tree for execution; if the execution result of the node is valid, then execute the subsequent nodes in sequence; if a node task fails to execute, then mark the node as an invalid node, and trace back to the nearest valid node, and then select other branches for execution. Continue according to the above process until the tasks of all nodes on a branch are successfully executed, and the final decision will be determined as the task scheduling plan of the branch, that is, the final task scheduling plan is decided.

[0104] The above method of this embodiment, by adopting three stages of task scheduling sampling, task tree construction, and task tree execution, can flexibly realize the dynamic calling of task processes, improve the automation and intelligence of the distribution network load transfer process, and improve the flexibility and efficiency of task execution.

[0105] In a specific application embodiment, by configuring a task orchestration tool built for load transfer intelligent entities, model calling, tool calling, task identification and execution of mechanism models such as topology identification, heavy overload warning, and load transfer can be automatically completed, thereby achieving full-process automated control of load transfer in the distribution network and greatly improving control efficiency.

[0106] The present invention realizes intelligent recommendation of distribution network load transfer strategy by integrating intention recognition, entity recognition, dynamic calling of mechanism model and index evaluation. It can intelligently recognize the user's intention scenario and the entity name of the power grid equipment, and then realize dynamic calling of the mechanism model through dynamic parameter adjustment based on the results of intention recognition and entity recognition. It can autonomously call the corresponding mechanism model to ensure the flexibility and real-time performance of the system. Then, based on the prompt words of L1 and the evaluation of the transfer strategy scheme, it improves the ability of the load transfer intelligent body to evaluate the scheme, and realizes the optimal ranking of the candidate transfer schemes. Ultimately, it can effectively reduce the decision-making time for distribution network fault handling (by more than 70%), improve the overall transfer strategy availability (by more than 90%), and significantly improve the distribution network load transfer efficiency.

[0107] This embodiment further provides a distribution network load transfer intelligent device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the methods of the above-mentioned embodiments 1 and 2.

[0108] It is understandable that the above method of this embodiment can be executed by a single device, such as a computer or server, etc., and can also be applied to a distributed scenario and completed by multiple devices in cooperation with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the above method of this embodiment, and multiple devices interact to complete the above method. The processor can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., for executing related programs to implement the above method of this embodiment. The memory can be implemented in the form of a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device. The memory can store an operating system and other applications. When the above method of this embodiment is implemented by software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0109] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0110] Those skilled in the art should understand that the above-mentioned embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0111] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for intelligently recommending load transfer strategies in a distribution network, characterized in that the steps include: Request input: The large model receives the user's input request, which includes the distribution network load transfer scenario information, power grid equipment entity information, and transfer constraint and transfer target parameter information required by the user; Intention scenario recognition: The large model recognizes the distribution network load transfer scenario required by the received input request and extracts the user's intention scenario; Power grid entity recognition: The large model performs entity recognition on the received input request and extracts the entity name of the power grid equipment; Dynamic call of mechanism model: The big model extracts the parameter information of the transfer constraints and transfer targets from the input request, and obtains the complete parameter information of the transfer constraints and transfer targets through multiple rounds of dialogue. The extracted intention scenario, grid equipment entity name and obtained parameter information are converted into mechanism model request parameters and a request is sent to the mechanism model to call the mechanism model to calculate the transfer strategy. The big model receives multiple candidate transfer strategies and evaluation index data of each candidate transfer strategy returned by the mechanism model. Evaluation of candidate power transfer strategies: RAG is used to read historical power transfer strategies and strategy constraint standards to determine a set of evaluation rules. A chain of thought approach is used to evaluate each candidate power transfer strategy under each evaluation rule. The candidate power transfer strategies that meet the preset conditions are screened out and ranked according to the evaluation results. The optimal power transfer strategy is determined based on the ranking results and output.

2. The method for intelligently recommending load transfer strategies for distribution network according to claim 1, characterized in that: The intended scenarios include any one or more of a main transformer heavy overload scenario, a feeder heavy overload scenario, a fault or maintenance scenario, a power supply task reading scenario, a power supply transfer plan scenario, and a power supply task optimization scenario.

3. The method for intelligently recommending load transfer strategies for distribution network according to claim 1, characterized in that: The step of identifying the load transfer scenario of the distribution network required for the received input request and extracting the user's intended scenario includes: Training the intent recognition model: pre-training the intent recognition model using an intent recognition data set, wherein the intent recognition data set includes multiple types of conversations corresponding to different distribution network load transfer scenarios; Candidate intent recognition: input the current input request into the trained intent recognition model for inference, and derive multiple candidate intent scenarios; Retrieval enhancement: providing the identified candidate intent scenarios as examples in the prompt words to the LLM; Model reasoning: embed retrieval enhancement into the prompt word template to obtain the prompt word, input the request to construct a complete prompt word, input the complete prompt word into the large model for reasoning, and obtain the final intention scenario.

4. The method for intelligently recommending load transfer strategies for distribution network according to claim 1, characterized in that: The performing entity recognition on the received input request and extracting the entity name of the power grid equipment includes: Constructing a knowledge base of power grid equipment entities: using a database containing information on various types of equipment entities in the distribution network to construct a power grid entity SQL vector knowledge base and a RAG vector database in the large model, wherein the RAG vector database is a textual knowledge base of power grid entities; Entity name query: extract the input entity name from the input request, generate an SQL statement based on the power grid entity SQL vector knowledge base, and query the entity information containing the input entity name based on the generated SQL statement. If no matching item is found, recall the results whose vector similarity exceeds the preset threshold in the RAG vector database as context, use the large model to determine the matching entity name, and output the queried entity name.

5. The method for intelligently recommending load transfer strategies for distribution network according to claim 4, characterized in that: It also includes regularly learning entity name rewriting rules from the sample library. When an entity name is queried, the entity name is rewritten according to the learned entity name rewriting rules to be output as the final queried entity name; when no match is found and the large model fails to match the entity name, the large model prompts the retrieval failure and guides the re-verification of the input entity name.

6. The method for intelligently recommending load transfer strategies for distribution network according to any one of claims 1 to 5, characterized in that: The method of adopting the thinking chain method is to evaluate each candidate power transfer strategy under each evaluation rule, screen out the candidate power transfer strategies that meet the preset conditions and sort them according to the evaluation results, determine the optimal power transfer strategy based on the sorting results and output the following: Determination and adjustment of evaluation rules: Read historical transfer strategies and strategy constraint standards through RAG to determine a set of evaluation rules, and adjust the order and weight of each evaluation rule according to the indicators to be concerned, to form an evaluation rule list; Traversing the evaluation rule screening scheme: The large model uses prompt words to define the thinking process based on the thinking chain method, removes each evaluation rule from the evaluation rule list in turn, and uses the removed evaluation rules to evaluate each candidate power transfer strategy until the evaluation results of all candidate power transfer strategies under all evaluation rules are obtained, and the candidate power transfer strategies that meet the preset conditions are screened out and sorted according to the evaluation results; Intelligent recommendation result output: The large model outputs the recommended results of the power transfer strategy based on the prompt word template. The recommended results include the optimal power transfer strategy and the thinking process.

7. The method for intelligently recommending load transfer strategies in a distribution network according to any one of claims 1 to 5, characterized in that: The method further includes a prompt word optimization step, wherein the prompt word optimization step includes: Conversation history analysis: Collect historical conversations in which current prompts interact with the big model in the distribution network load transfer scenario, and analyze the collected historical conversations to identify the directions in which prompts need to be optimized; Optimization plan generation: Generate multiple optimization plans based on the analysis results of historical conversations. The optimization plans include one or more of adding new instructions, modifying current instructions, and adjusting the order of instructions. The final optimization plan is selected from various optimization plans. Optimization plan integration: The final selected optimization plan is integrated into the original prompt to form an integrated optimization plan; The conversation history analysis, the optimization solution generation and the optimization solution integration are continuously performed until a preset iteration condition is reached to obtain an optimized prompt word.

8. A method for intelligent control of load transfer strategy in a distribution network, characterized in that the steps include: Determine the optimal transfer strategy according to the method described in any one of claims 1 to 7; A plurality of potential task scheduling plans are generated according to the tasks to be executed in the optimal transfer strategy, and an action tree is constructed according to each potential task scheduling plan. The final task scheduling plan is decided according to the action tree to control the execution of tasks according to the decided task scheduling plan. If two task scheduling plans in the action tree have the same tasks, they are aggregated into one branch, and different tasks form different branches.

9. The intelligent control method for load transfer strategy of distribution network according to claim 8 is characterized in that: The steps of generating a plurality of potential task scheduling plans according to the tasks to be executed in the optimal transfer strategy, constructing an action tree according to each potential task scheduling plan, and determining the final task scheduling plan according to the action tree to control the execution of tasks according to the determined task scheduling plan include: Task scheduling sampling: Obtain various types of task scheduling to form a preliminary task scheduling execution plan, and use the big model to generate multiple potential task scheduling plans based on the global information of the task and the preliminary task scheduling execution plan; Action tree construction: Take the load transfer entry task as the root node of the tree, and compare the potential task scheduling plans in sequence starting from the first task. When two task scheduling plans have the same tasks, they are aggregated into one branch, and different tasks form different branches, until all task scheduling plans are built into the tree to form a complete action tree; Tree-based decision: select a node from any branch in the tree for execution in turn. If the execution result of the executed node is valid, proceed to the next node in turn. If the task of the executed node fails, it is marked as an invalid node and traced back to the nearest valid node. Then select other branches for execution until the tasks of all nodes on the target branch are successfully executed. The task scheduling plan of the target branch is determined as the final decision.

10. A load transfer intelligent device for a distribution network, comprising a processor and a memory, wherein the memory is used to store a computer program, characterized in that: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 9.

Images