Scheduling hall intelligent explanation system and method based on big language model
By constructing an intelligent explanation system based on a large language model, the problems of inconsistency and lack of interactivity in traditional dispatch hall explanation methods have been solved, enabling efficient and personalized explanation of power dispatch knowledge and reducing labor costs.
Patent Information
- Application Number
- CN202510924224.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-04
- Publication Date
- 2025-11-25
AI Technical Summary
Traditional methods of explanation in the dispatch hall lack consistency and comprehensiveness. Human explanations rely on individual knowledge, while electronic explanation devices lack interactivity and cannot provide accurate and in-depth explanations.
A smart explanation system based on a large language model is constructed, including a data acquisition and preprocessing module, a large language model core module, an interactive interface module, and a display and feedback module. This system enables automated generation of explanation content and personalized services, and uses the large language model to understand and generate professional text answers.
It improves the accuracy and comprehensiveness of explanations, enables personalized services, reduces labor costs, and can efficiently meet the needs of visitors around the clock.
Smart Images
Figure CN121008764A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of power system automation, artificial intelligence application and multimedia display, in particular to a dispatch hall intelligent interpretation system and method based on a large language model, which is suitable for visit reception interpretation in power station, power grid dispatching center and other scenes. BACKGROUND
[0002] In the visit reception process of the dispatching (central control) hall, the traditional interpretation method has many limitations. On the one hand, manual interpretation depends on the individual knowledge reserve and expression ability of the interpreter, and it is difficult to ensure the consistency and comprehensiveness of each interpretation. For complex power dispatching knowledge, such as three lines of defense, primary frequency modulation, and cooperative operation mechanism of different power equipment, the interpreter may not be able to provide accurate and in-depth interpretation for visitors due to memory bias or insufficient understanding. On the other hand, traditional electronic interpretation equipment, such as pre-recorded voice guide or fixed content electronic display board, lacks flexibility and interactivity, and cannot provide targeted answers according to the questions and concerns of visitors.
[0003] With the rapid development of artificial intelligence technology, large language models have shown strong natural language processing capabilities and can understand complex semantics and generate reasonable text responses. Applying them to dispatch hall interpretation has great potential value, but there are currently no mature related applications. SUMMARY
[0004] In view of the problems of lack of consistency and comprehensiveness in manual interpretation and lack of interactivity in traditional electronic interpretation equipment in the visit reception of the power dispatching hall, the present application proposes a dispatch hall intelligent interpretation system and method based on a large language model, which builds a system architecture including a data collection and preprocessing module, a large language model core module, an interactive interface module, and a display and feedback module, and designs an initialization, interaction, and feedback optimization method process, realizes automatic generation of interpretation content, accurate and professional output, and personalized services according to the needs of visitors, significantly improves the visit reception interpretation effect, and helps visitors better understand the power dispatching related knowledge and operation mechanism.
[0005] In order to realize the above technical features, the purpose of the present application is realized as follows: a dispatch hall intelligent interpretation system based on a large language model, comprising: a data collection and preprocessing module for collecting power and reservoir dispatching related data and performing cleaning, labeling and structural processing on the data; a large language model core module using an optimized and fine-tuned large language model, which learns based on the preprocessed data and has the ability to understand power and reservoir dispatching professional problems and generate interpretation text; An interactive interface module provides voice interaction and text interaction modes, receives visitor questions and displays answers generated by the large language model to the visitors; A display and feedback module integrates display explanation content with the dispatching hall display device and collects visitor feedback information for optimizing the system.
[0006] Preferably, the data collected by the data collection and preprocessing module includes real-time operation data of power plants, historical scheduling data, power equipment parameters, power industry regulations and policies, and popular science knowledge.
[0007] Preferably, the large language model core module is based on the Transformer architecture and learns and understands scheduling knowledge by pre-training and fine-tuning on a large amount of power and reservoir scheduling field data.
[0008] Preferably, the voice interaction part of the interactive interface module integrates a speech recognition engine and a speech synthesis engine to realize the conversion of voice and text and the playback of answer voice.
[0009] Preferably, the text interaction part of the interactive interface module realizes visitor text question input and answer text display through a touch query terminal.
[0010] Preferably, the display and feedback module displays charts and animation visual content on the display device to assist in explanation, and collects visitor evaluation, new questions and suggestion feedback information through feedback buttons and input boxes.
[0011] In another aspect of the application, a dispatching hall intelligent explanation method based on a large language model is provided, which is implemented by using the explanation system and includes the following steps: Initialization stage: start the system, the data collection and preprocessing module collects and updates data and inputs the large language model for loading and preheating, and the interactive interface module initializes the speech recognition, speech synthesis engine and text interaction interface; Interaction stage: the interaction stage includes a three-level processing flow: Semantic analysis layer: key terms in the question are extracted and converted to keywords using named entity recognition, and the question type is determined by an intent classifier; Knowledge retrieval layer: relevant knowledge points are quickly located from the water conservancy and power knowledge graph according to the question keywords, and the model reasoning range is narrowed; Answer generation layer: the model generates a structured answer in combination with real-time data and historical cases; Feedback optimization stage: the display and feedback module collects visitor feedback information, which is transmitted back to the large language model core module and the data collection and preprocessing module for fine-tuning and optimization of the model and improvement of data collection.
[0012] Preferably, the key terms of the semantic analysis layer include: hydropower station, active power; The problem types of the semantic analysis layer include: device principle class, operation mechanism class or fault handling class; The power knowledge graph of the knowledge retrieval layer contains at least 100,000 entity relationships; The real-time data of the answer generation layer includes the current power station load; The historical cases of the answer generation layer include similar fault handling records; The generation of the structured answer of the answer generation layer includes text, chart reference identification.
[0013] Preferably, in the interaction stage, if the visitor asks questions by voice, the voice recognition engine first converts the voice into text and then transmits it to the large language model core module; if the question is asked by text, it is directly transmitted to the large language model core module.
[0014] Preferably, in the feedback optimization stage, the data acquisition and preprocessing module collects more abundant data according to the feedback information, and the large language model core module adjusts and optimizes the model parameters according to the new data and feedback, to improve the accuracy and adaptability of subsequent explanation.
[0015] The present application has the following advantages: 1. Improve the accuracy and comprehensiveness of the explanation: the large language model learns based on massive water conservancy and power data, and can accurately and comprehensively answer the questions of the visitors, avoiding the knowledge bias and omissions that may occur in artificial explanation. For complex dispatching knowledge, the model can provide a more complete knowledge system from multiple angles. For example, when explaining the power system stability analysis, the model can elaborate the principles, influencing factors and corresponding dispatching control measures of various stability types (such as power angle stability, voltage stability and frequency stability), while artificial explanation may not be so comprehensive due to individual knowledge limitations.
[0016] 2. Realize personalized explanation: according to the questions and feedback of the visitors, the system can flexibly adjust the explanation content and method to realize personalized service. Different visitors have different levels of understanding and interest points of water conservancy and power knowledge, and the system can provide the most suitable explanation for each visitor according to their specific circumstances. For example, for visitors with power professional background, the system can provide more in-depth technical details and cutting-edge research results; for ordinary visitors, the system uses simple and easy-to-understand language and vivid metaphors to explain.
[0017] 3. Reduce labor costs: reduce the dependence on a large number of professional interpreters, reduce the cost of human training and management. The system can provide interpretation services for visitors all year round, and ensure the stable quality of each interpretation, which is not affected by factors such as personnel emotions, fatigue, etc. During the peak of the visit, it can also easily cope with the needs of a large number of visitors to ask questions, improve reception efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0018] The present application will be further described below in conjunction with the accompanying drawings and examples.
[0019] Fig. 1 A system architecture schematic diagram of a scheduling hall intelligent interpretation system and method based on a large language model is provided for the present application.
[0020] Fig. 2 A method flowchart of a scheduling hall intelligent interpretation system and method based on a large language model is provided for the present application.
[0021] Fig. 3 A display interface effect schematic diagram of a scheduling hall intelligent interpretation system and method based on a large language model is provided for the present application. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.
[0023] Embodiment 1: Participation Figs. 1-3 The present embodiment provides a specific process of system building of a scheduling (central control) hall intelligent interpretation system based on a large language model: 1. Data acquisition and preprocessing: Obtain real-time operation data from the power plant dispatching automation system, including active, reactive, voltage, etc. of the power plant, and collect them through data interface at a certain time interval, such as every minute; Extract power and reservoir scheduling data in the past few years from the historical database, including daily load curve, power generation plan execution, fault record, etc. Collect equipment manuals, technical parameters, etc. provided by power plant equipment manufacturers and organize them into structured data.
[0024] Collect authoritative regulations and policies, academic research papers, and popular science articles in the water conservancy and power industry. Use natural language processing techniques to clean up the text data, remove duplicates, errors, and irrelevant content, and then label them, such as classifying regulations and policies according to different topics (such as power safety, dispatching management, etc.).
[0025] 2. Large language model selection and fine-tuning: Select open-source large language models that perform well in natural language processing, such as GPT-NeoX.
[0026] Fine-tune the model using the collected data in the power and reservoir dispatching fields. Organize the labeled data into a format suitable for model training, such as grouping questions and corresponding answers. During fine-tuning, set appropriate training parameters such as learning rate, training rounds, etc. Through multiple rounds of training, the model better adapts to the understanding and generation of power dispatching knowledge. For example, after multiple rounds of fine-tuning, the model can accurately understand professional questions such as "What are the functions of reactive power compensation devices in the power system?" and generate detailed and accurate answers such as "The main functions of reactive power compensation devices include improving power factor, reducing line loss; stabilizing voltage, improving power quality; improving the stability and transmission capacity of the power system, etc. Common reactive power compensation devices include shunt capacitors, static var compensators, and synchronous phase modulators, which have their own advantages in different power scenarios……".
[0027] 3. Interactive interface development: Voice interaction part, integrate open-source speech recognition engine (such as Kaldi) and speech synthesis engine (such as Festival). In the dispatching (central control) hall, multiple high-quality microphones are arranged to collect visitors' voice questions. The microphone is connected to the speech recognition engine to convert the collected voice into text in real time. The speech synthesis engine synthesizes natural and smooth voice according to the answer text generated by the large language model and plays it through the loudspeaker to the visitors.
[0028] Text interaction part, set multiple touch query terminals on the visitor route in the dispatching (central control) hall. Visitors can input text questions on the terminal. The touch query terminal is connected to the large language model core module through the network, sends the question to the model, and receives the answer text generated by the model, then displays it on the terminal screen.
[0029] 4. Integration of display and feedback modules: Integrate with the large screen in the dispatching (central control) hall. When explaining content related to the operating state of the power system that needs to be visualized, the system can display relevant charts and animations on the large screen in synchronization. For example, when explaining the trend of power load changes, the large screen will dynamically display the load curve changes in the past week.
[0030] A feedback button is set on the touch query terminal and voice interaction device, and visitors can evaluate the interpretation content by clicking the button (such as satisfied, general, and dissatisfied), and can also input new questions or suggestions as feedback. These feedback information is transmitted back to the large language model core module and the data collection and preprocessing module through the network, for subsequent optimization.
[0031] Embodiment 2: Based on the system construction method in embodiment 1, a large language model-based scheduling (central control) hall intelligent interpretation system is finally constructed, which specifically includes: A data collection and preprocessing module is used to collect and clean, label, and structure process related data such as power and reservoir scheduling; A large language model core module adopts an optimized and fine-tuned large language model, which learns based on preprocessed data and has the ability to understand professional problems such as power and reservoir scheduling and generate interpretation text; An interactive interface module provides voice interaction and text interaction methods, receives visitor questions, and displays the answers generated by the large language model to the visitors; A display and feedback module integrates with the scheduling (central control) hall display device to display interpretation content and collects visitor feedback information for optimizing the system.
[0032] Further, the data collected by the data collection and preprocessing module includes real-time operation data of power stations, historical scheduling data, power equipment parameters, power industry regulations and policies, and popular science knowledge.
[0033] Further, the large language model core module is based on the Transformer architecture, which learns and understands scheduling knowledge by pre-training and fine-tuning on a large amount of power and reservoir scheduling domain data.
[0034] Further, the voice interaction part of the interactive interface module integrates a speech recognition engine and a speech synthesis engine to realize the conversion of voice and text and the playback of answer voice; the text interaction part realizes the input of visitor text questions and the display of answer text through the touch query terminal.
[0035] Further, the display and feedback module displays visual content such as charts and animations on the display device to assist interpretation, and collects feedback information such as visitor evaluation, new questions, and suggestions through feedback buttons and input boxes.
[0036] Embodiment 3: Based on the interpretation system in embodiment 2, a large language model-based scheduling hall intelligent interpretation method is provided, including the following steps: Initialization phase: start the system, the data acquisition and preprocessing module collects and updates data and inputs the large language model for loading and preheating, and the interactive interface module initializes the speech recognition, speech synthesis engine and text interaction interface; Interaction phase: the interaction phase includes a three-level processing flow: Semantic analysis layer: use named entity recognition (NER) to extract key terms in the question (such as "hydropower station" and "active power"), and use an intent classifier to determine the question type (device principle class / operation mechanism class / fault handling class); Knowledge retrieval layer: according to the key words of the question, quickly locate the relevant knowledge points from the water conservancy and power knowledge graph (containing more than 100,000 entity relationships), and narrow the model reasoning range; Answer generation layer: the model combines real-time data (such as the current power station load) and historical cases (such as similar fault handling records) to generate structured answers (including text and chart reference identifiers); Feedback optimization phase: the display and feedback module collects visitor feedback information, transmits it back to the large language model core module and the data acquisition and preprocessing module, and fine-tunes and optimizes the model and improves data collection.
[0037] Further, in the interaction phase, if the visitor asks questions by voice, the speech recognition engine first converts the voice into text and then transmits it to the large language model core module; if the question is asked by text, it is directly transmitted to the large language model core module.
[0038] Further, in the feedback optimization phase, the data acquisition and preprocessing module collects more rich data according to the feedback information, and the large language model core module adjusts and optimizes the model parameters according to the new data and feedback, to improve the accuracy and adaptability of subsequent explanations.
[0039] Embodiment 4: This embodiment provides a specific use process of a large language model-based dispatch (central control) hall intelligent explanation system: 1. When the visitor enters the dispatch (central control) hall and walks to the area where the interactive device is set, he or she can choose to ask questions by voice or text. For example, a visitor asks by voice "What tasks do hydropower station power dispatchers mainly handle in their daily work?"; 2. The speech recognition engine converts the voice into text "What tasks do hydropower station power dispatchers mainly handle in their daily work?" and transmits the text to the large language model core module; 3. The large language model core module receives the question and searches and reasons in its fine-tuned knowledge system. It generates the following answer text from the learned knowledge related to the work responsibilities of a hydropower station dispatcher: "The daily work of a hydropower station power dispatcher mainly includes real-time monitoring of the operation status of the hydropower station, closely monitoring parameters such as water turbine speed, water head, active power, reactive power, voltage, current, etc., to ensure safe and stable operation of the hydropower station; according to the power demand of the power grid, the inflow of the reservoir and the power generation plan, reasonably arrange the power generation output of the hydropower station, scientifically formulate the power generation dispatching plan, coordinate the allocation of power resources between the hydropower station and other power stations; when the hydropower station or the power grid fails, quickly determine the nature and location of the fault, command the maintenance personnel to carry out fault troubleshooting and repair work, minimize the power outage time, and ensure power supply; at the same time, it also needs to communicate and coordinate with the power trading center, upstream and downstream power stations, etc., participate in power market transactions, ensure the smooth completion of power transactions, and maximize the power generation efficiency of the hydropower station"; 4. The generated answer text is transmitted back to the interactive interface module, and the speech synthesis engine synthesizes the text into speech and plays it to the visitor, while the touch query terminal (if the visitor is asking questions by voice, the nearby touch query terminal can also display the answer text synchronously); 5. After listening to the answer, if the visitor is satisfied, he can continue to visit; if he has further questions, he can ask again. If the visitor is not satisfied with the answer, he can select "not satisfied" through the feedback button and input feedback opinions such as "the answer is too general, and I hope to introduce the fault judgment method in detail". These feedback information is transmitted back to the system, and the data collection and preprocessing module collects more detailed information related to the power fault judgment method according to the feedback opinions, and the large language model core module is fine-tuned and optimized again according to the new information, so as to provide more detailed and accurate content when answering similar questions in the future.
[0040] Through the above system building and running process, the dispatching (central control) hall intelligent explanation system and method based on large language model of the present application can effectively provide high-quality and personalized explanation services for visitors, and improve the visit and reception level of the dispatching (central control) hall.
[0041] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.
Claims
1. A large language model-based dispatch hall intelligent interpretation system, characterized in that, The method comprises the following steps: A data collection and preprocessing module is used to collect power and reservoir scheduling related data, and to clean, label and structure the data; A large language model core module is used to learn based on the preprocessed data, and has the ability to understand power and reservoir scheduling professional problems and generate explanation texts; An interactive interface module is provided to receive visitor questions and display the answers generated by the large language model to the visitors; A display and feedback module is integrated with the scheduling hall display device to display the explanation content and collect visitor feedback information for optimizing the system.
2. The dispatch hall intelligent interpretation system based on a large language model according to claim 1, characterized in that, The data collected by the data collection and preprocessing module includes real-time operation data of power stations, historical scheduling data, power equipment parameters, power industry regulations and policies, and popular science knowledge.
3. The dispatch hall intelligent interpretation system based on a large language model according to claim 1, characterized in that, The large language model core module is based on the Transformer architecture, and learns and understands scheduling knowledge through pre-training and fine-tuning on a large amount of power and reservoir scheduling field data.
4. The dispatch hall intelligent interpretation system based on a large language model according to claim 1, characterized in that, The voice interaction part of the interactive interface module integrates a speech recognition engine and a speech synthesis engine to realize the conversion of voice and text and the playback of answering voice.
5. The dispatch hall intelligent interpretation system based on a large language model according to claim 1, characterized in that, The text interaction part of the interactive interface module realizes visitor text question input and answer text display through a touch query terminal.
6. The dispatch hall intelligent interpretation system based on a large language model according to claim 1, characterized in that, The display and feedback module displays charts and animation visual content on the display device to assist in explanation, and collects visitor evaluation, new questions and suggestion feedback information through feedback buttons and input boxes.
7. A large language model-based dispatch hall intelligent interpretation method, characterized in that, The method is implemented by using the explanation system of any one of claims 1-6, comprising the following steps: Initialization stage: start the system, the data collection and preprocessing module collects and updates the data and inputs the large language model for loading and preheating, and the interactive interface module initializes the speech recognition, speech synthesis engine and text interaction interface; Interactive stage: the interactive stage includes three processing procedures: Semantic analysis layer: use named entity recognition to extract key terms in the question and convert them to keywords, and use an intent classifier to determine the question type; Knowledge retrieval layer: quickly locate related knowledge points from the water conservancy and power knowledge graph according to the question keywords, and narrow the model reasoning range; Answer generation layer: the model generates a structured answer combining real-time data and historical cases; Feedback optimization stage: the display and feedback module collects visitor feedback information, which is transmitted back to the large language model core module and the data collection and preprocessing module for fine-tuning and optimization of the model and improvement of data collection.
8. The method according to claim 7, wherein, The key terms of the semantic analysis layer include: hydropower station, active power; The question types of the semantic analysis layer include: device principle type, operation mechanism type or fault handling type; The power knowledge graph of the knowledge retrieval layer contains at least 100,000 entity relationships; The real-time data of the answer generation layer includes the current power station load; The historical cases of the answer generation layer include similar fault handling records; The generated structured answer of the answer generation layer includes text and chart reference identifiers.
9. The method according to claim 7, wherein, In the interaction stage, if the visitor asks questions by voice, the voice recognition engine first converts the voice into text and then transmits it to the large language model core module; if the question is asked by text, it is directly transmitted to the large language model core module.
10. The method of claim 7, wherein the method further comprises: In the feedback optimization stage, the data acquisition and preprocessing module collects more abundant data according to the feedback information, and the large language model core module adjusts and optimizes the model parameters according to the new data and feedback, improving the accuracy and adaptability of subsequent explanations.