Water conservancy project auxiliary decision-making method and system based on knowledge graph and large model

Abstract

The invention relates to a hydraulic engineering auxiliary decision-making system based on a large model algorithm, and the system is characterized in that firstly, various emergency problems possibly occurring in a hydraulic engineering are widely collected, and the problems are introduced into a professional knowledge graph of hydraulic engineering emergency disposal, so as to seek accurate solutions. Meanwhile, the model can record entity relations involved in the question answering process and necessary clarification questions in detail, and the entity relations and the necessary clarification questions are integrated into an optimal clarification path. And then, the optimal clarified path is used as a learning material of a large language model, and the model gradually masters professional knowledge in the water conservancy project emergency field through deep learning, so that the reply content is ensured to be closely connected with the water conservancy emergency reality. According to the method, missing entity information in the query is accurately captured and supplemented, so that the query clarification efficiency and the answer accuracy are improved. Decision questions and answers based on the knowledge graph and the large language model solve the problems of limited data of the knowledge graph and noise of the language model, and provide comprehensive and efficient answers for water conservancy management.

Description

Technical Field

[0001] The present invention relates to a water conservancy project decision-making assistance method and system based on knowledge graph and large model, which are applicable to the field of water conservancy project technology. Background Art

[0002] In the field of water conservancy projects, due to the emergence of diversified and heterogeneous information accumulated over a long period of time, workers face great challenges in quickly and accurately retrieving historical risk records and drawing on past emergency strategies, which undoubtedly weakens the practical application value of effective information. In order to meet this challenge, it is particularly important to build an efficient and intelligent question-answering system, which can provide workers with accurate information support and thus improve their efficiency in solving engineering safety problems.

[0003] At present, the question-answering system based on knowledge graph has successfully converted unstructured and semi-structured text data into structured data by integrating the relationship between entities into a clear triple form, which significantly improves the quality of data. This provides staff with clear and structured domain knowledge. However, since the system mainly serves vertical fields and is limited by the amount of text data, when the entities involved in the query statement exceed its preset range, the system will have a vocabulary out of bounds phenomenon. In addition, when the manager's question intention is not clear enough, the system still needs to rely on preset templates to clarify the question, such as "Are you asking about XX?" This mechanized interaction method often lacks flexibility.

[0004] With the rapid development of artificial intelligence technology, large language models (such as CHATGPT) have become a popular choice for solving specific task question answering. These models have been trained with massive data and have two significant advantages: one is that they contain rich prior knowledge; the other is that they have excellent semantic understanding and contextual learning capabilities, which enable them to perform well when dealing with general domain problems. However, when applied to vertical domain-related question answering tasks, large language models have not achieved the expected results due to noise and randomness issues. Summary of the invention

[0005] The technical problem to be solved by the present invention is: to provide a water conservancy project decision-making assistance method and system based on knowledge graph and large language model, to alleviate the shortcomings of water conservancy project knowledge graph and directly using large language model to answer water conservancy project questions. The present invention combines water conservancy project knowledge graph and large language model to make intelligent question and answer, using each other's advantages to make up for each other's shortcomings.

[0006] The technical solution adopted by the present invention is: a water conservancy project decision-making assistance method based on knowledge graph and large model, comprising the following steps: Step S1: Select a suitable general large model and a professional knowledge map in the field of water conservancy engineering emergency, and pre-learn the large model; Step S2: Collect various emergencies related to flood management and facility safety in water conservancy projects, including abnormal water level fluctuations, structural damage, and equipment operation failures, use the big model to analyze and identify the types and locations of various problem events, and clarify the unclear parts of the problems; Step S3: locate the analysis results of the event type and location of each type of problem in the water conservancy emergency response professional knowledge map and answer the questions, and record the entity relationships and clarification steps involved in the problem clarification in step S2 and the answering process in this step; Step S4: Combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path for each type of question and convert it into a vector representation; Step S5: The generated optimal clarification paths for various types of questions are used as further learning materials for the large model. By learning these materials, the large model's specialized knowledge mastery and reasoning ability in the field of water conservancy engineering emergencies is enhanced; Step S6: When a user asks a query, the big model is used to match and analyze the query with the pre-generated clarification path, determine the user's clarification needs, generate and integrate clarification information, and provide the final decision answer.

[0007] Further, step S2 includes the following steps: Step S201: Collect various emergencies related to flood management and facility safety in water conservancy projects, including abnormal water level fluctuations, structural damage, and equipment operation failures; Step S202: using a large model pre-trained with a professional knowledge graph in the field of water conservancy engineering emergency to identify whether there are unclear parts in the question regarding the event type and the location of occurrence; if there are unclear parts, the large model sets a clarification direction based on the entity relationships related to the unclear parts in the knowledge graph, and locates and presents relevant entity options; Step S203: Repeat steps S201 and S202 until the model can clearly analyze the type and location of the problem event.

[0008] Further, in step S3, the optimal clarification path includes: Initial first round of clarification directions and corresponding first round of entity options; When there are unclear parts in the question regarding the event type and location, the clarification directions and corresponding entity options for each round; Presentation of the final answer to the question after the unclear parts have been clarified.

[0009] Further, step S6 includes the following steps: Step S601: When a user asks a query, the big model converts the user's query into a vector, compares it with the existing optimal clarification paths, and selects the 20% clarification path data that best matches the query; Step S602: Based on the clarification path data of the matching results, the big model determines the user's clarification needs, generates corresponding clarification questions, and provides the user with multiple appropriate options for the user to select or enter answers by himself; Step S603: Integrate the generated clarifying question and the user's selected or input answer into the initial query, process it and generate a final answer.

[0010] According to a second aspect of the present invention, a water conservancy project decision support system based on a knowledge graph and a large model is provided, comprising: Optimal clarification unit: used to combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path, and use these paths as learning materials for the big model to learn, so as to improve the big model's knowledge mastery and reasoning ability on the professional knowledge map in the field of water conservancy engineering emergency; Clarification question generation unit: used to generate clarification questions based on the matching and analysis results of the large model when the user asks a query question, and identify the user's clarification needs; Alternative solution generation unit: used to generate appropriate options for users to choose from or enter their own answers based on clarification questions, thereby optimizing the understanding of user query intent; Query expansion unit: used to integrate the user's selected answer or input information into the initial query, expand and improve the query content, and provide more accurate data support for the generation of the final answer; Knowledge answering unit: used to locate problems and generate final decision answers based on the expanded query through the reasoning capabilities of the knowledge graph and large model.

[0011] Compared with the existing technology, the innovation of the present invention is that it does not rely on the extraction of existing entities in the query question, but realizes human-computer interaction through five intelligent units to accurately capture and supplement the missing entity information in the query question. This method not only improves the efficiency of query clarification, but also ensures the accuracy of the answer.

[0012] At the same time, the water conservancy project decision-making question and answer based on knowledge graph and large language model alleviates the problems of limited water conservancy knowledge graph data and rigid clarification sentences, and also alleviates the noise and randomness problems of large language models. It can provide comprehensive and efficient answers for water conservancy managers. The optimal clarification path and five AI units studied in this invention also greatly improve the efficiency of the clarification process and the correctness of the reply answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0014] This embodiment is a method and system for assisting decision-making in water conservancy projects based on knowledge graphs and large models. The following is a clear and complete description of the technical solutions in the embodiments of the present invention: The following is combined with Figure 1 The structure and working process of the present invention are specifically described in detail. A water conservancy project decision-making assistance method based on knowledge graph and large model includes the following steps: Step S1: Select a suitable general large model and a professional knowledge map in the field of water conservancy engineering emergency, and pre-learn the large model; Step S2: Collect various emergencies related to flood management and facility safety in water conservancy projects, including abnormal water level fluctuations, structural damage, and equipment operation failures, use the big model to analyze and identify the types and locations of various problem events, and clarify the unclear parts of the problems; Step S3: locate the analysis results of the event type and location of each type of problem in the water conservancy emergency response professional knowledge map and answer the questions, and record the entity relationships and clarification steps involved in the problem clarification in step S2 and the answering process in this step; Step S4: Combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path for each type of question and convert it into a vector representation; Step S5: The generated optimal clarification paths for various types of questions are used as further learning materials for the large model. By learning these materials, the large model's specialized knowledge mastery and reasoning ability in the field of water conservancy engineering emergencies is enhanced; Step S6: When a user asks a query, the big model is used to match and analyze the query with the pre-generated clarification path, determine the user's clarification needs, generate and integrate clarification information, and provide the final decision answer.

[0015] Enter at the same time Figure 1 As shown, the present invention provides a water conservancy project decision-making support system based on knowledge graph and large model, including: Optimal clarification unit: used to combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path, and use these paths as learning materials for the big model to learn, so as to improve the big model's knowledge mastery and reasoning ability on the professional knowledge map in the field of water conservancy engineering emergency; Clarification question generation unit: used to generate clarification questions based on the matching and analysis results of the large model when the user asks a query question, and identify the user's clarification needs; Alternative solution generation unit: used to generate appropriate options for users to choose from or enter their own answers based on clarification questions, thereby optimizing the understanding of user query intent; Query expansion unit: used to integrate the user's selected answer or input information into the initial query, expand and improve the query content, and provide more accurate data support for the generation of the final answer; Knowledge answering unit: used to locate problems and generate final decision answers based on the expanded query through the reasoning capabilities of the knowledge graph and large model.

[0016] The optimal clarification unit, clear question generation unit, alternative solution generation unit, query expansion unit and knowledge answering unit in the present invention gradually clarify the missing parts in the query question, and adopt a human-computer interaction method to better understand the manager's query object to determine the missing entity and add it to the query question for expansion, and then return the answer.

[0017] The present invention first allows the big model to learn the content of the knowledge graph, and then judges the missing parts in the user's question sentence according to the five AI units, and judges which direction is better for clarification, and then generates clarification questions, and generates corresponding options for the generated clarification questions according to the learned content. If the generated options do not meet the user's requirements, the user can also know how to input his or her own answers according to the generated option format to let the big model expand the query question, and then let the big model reply to the question. This approach allows the big model to learn domain knowledge first and then ask questions, which can not only guide users to answer questions better, but also makes the overall clarification process and human-computer interaction process closely revolve around the water conservancy field, and alleviates the noise and answer randomness problems caused by the big model when answering questions to the greatest extent.

[0018] The use of the auxiliary decision-making question-answering method of the present invention is as follows: after the manager inputs a water conservancy emergency problem, firstly, the matching mechanism is used to find the top 20% optimal clarification paths that are most relevant to the manager's problem, so that the large language model can clarify the problem through the matched optimal clarification path, which is helpful to improve the clarification efficiency, and then the problem is sent to the optimal clarification direction unit, and an optimal clarification direction is first determined according to the unspecified part in the initial query statement, and the clarification question generation unit then generates a related clarification question according to the obtained optimal clarification direction. Since the optimal clarification path has been learned in advance, the alternative solution generation unit can generate suitable candidate options for the clarification question and sort them according to the relevance to the initial query statement. After the manager makes a selection, the large language model understands the specific object of the unspecified part, and the query expansion unit then adds the above-obtained clarification question and the answer selected by the manager to the initial query statement for expansion, so that the semantic information is more specific. Finally, the knowledge answer unit gives the final answer according to the expanded query statement with clear semantic information and sorts it according to the relevance to the initial query statement, and puts the optimal answer in the first row to complete the entire question-answering process. In addition, in order to better play the functions of these five AI units, the present invention also sets corresponding task prompts for each unit. The overall process and the task prompts of each unit are shown in the attached figure. It is worth noting that the historical answers and historical answers contained in the attached figure are created for the manager's questions that contain multiple unspecified parts and require multiple rounds of questions and answers. When the manager's question still cannot be effectively answered after one round of questions and answers, the manager can choose to continue to clarify the questions and answers. At this time, the options selected by the manager in the first round and the answers given by the large language model will be stored in the system to facilitate the second round of clarification questions and answers.

[0019] Specifically in the field of water conservancy engineering emergency, when the user initially queries the problem: when a turbine fails, the process of the present invention is as follows: Best clarification aspect: Emergency type clarification question: Water level is too high → Alternative options: 1. Operation beyond flood limit, 2. Heavy rain, 3. Flood → User answer: Operation beyond flood limit → Extended sentence: The dam water level is 1m higher than the normal water storage level, what should I do? → Knowledge answer: 1. Lower the water level by opening the gate to release water, 2. Perform dam safety review calculations, 3. Determine whether the release has an impact on the downstream.

[0020] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A water conservancy project decision-making assistance method based on knowledge graph and large model, characterized in that: The following steps are involved: Step S1: Select a suitable general large model and a professional knowledge map in the field of water conservancy engineering emergency, and pre-learn the large model; Step S2: Collect various emergencies related to flood management and facility safety in water conservancy projects, including abnormal water level fluctuations, structural damage, and equipment operation failures, use the big model to analyze and identify the types and locations of various problem events, and clarify the unclear parts of the problems; Step S3: locate the analysis results of the event type and location of each type of problem in the water conservancy emergency response professional knowledge map and answer the questions, and record the entity relationships and clarification steps involved in the problem clarification in step S2 and the answering process in this step; Step S4: Combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path for each type of question and convert it into a vector representation; Step S5: The generated optimal clarification paths for various types of questions are used as further learning materials for the large model. By learning these materials, the large model's specialized knowledge mastery and reasoning ability in the field of water conservancy engineering emergencies is enhanced; Step S6: When a user asks a query, the big model is used to match and analyze the query with the pre-generated clarification path, determine the user's clarification needs, generate and integrate clarification information, and provide the final decision answer.

2. The water conservancy project decision-making assistance method based on a large model algorithm as claimed in claim 1, characterized in that: Step S2 includes the following steps: Step S201: Collect various emergencies related to flood management and facility safety in water conservancy projects, including abnormal water level fluctuations, structural damage, and equipment operation failures; Step S202: using a large model pre-trained with a professional knowledge graph in the field of water conservancy engineering emergency to identify whether there are unclear parts in the question regarding the event type and the location of occurrence; if there are unclear parts, the large model sets a clarification direction based on the entity relationships related to the unclear parts in the knowledge graph, and locates and presents relevant entity options; Step S203: Repeat steps S201 and S202 until the model can clearly analyze the type and location of the problem event.

3. The water conservancy project decision-making assistance method based on a large model algorithm as claimed in claim 1, characterized in that: In step S3, the optimal clarification path includes: Initial first round of clarification directions and corresponding first round of entity options; When there are unclear parts in the question regarding the event type and location, the clarification directions and corresponding entity options for each round; Presentation of final answer to the question after unclear parts have been clarified.

4. The water conservancy project decision-making assistance method based on a large model algorithm as claimed in claim 1, characterized in that: Step S6 includes the following steps: Step S601: When a user asks a query, the big model converts the user's query into a vector, compares it with the existing optimal clarification paths, and selects the 20% clarification path data that best matches the query; Step S602: Based on the clarification path data of the matching results, the big model determines the user's clarification needs, generates corresponding clarification questions, and provides the user with multiple appropriate options for the user to select or enter answers by himself; Step S603: Integrate the generated clarifying question and the user's selected or input answer into the initial query, process it and generate a final answer.

5. A water conservancy project decision support system based on knowledge graph and large model, characterized in that: include: Optimal clarification unit: used to combine various types of questions with the structured entity relationship paths involved in their answering process to form the optimal clarification path, and use these paths as learning materials for the big model to learn, so as to improve the big model's knowledge mastery and reasoning ability in the professional knowledge map in the field of water conservancy engineering emergency; Clarification question generation unit: used to generate clarification questions based on the matching and analysis results of the large model when the user asks a query question, and identify the user's clarification needs; Alternative solution generation unit: used to generate appropriate options for users to choose from or enter their own answers based on clarification questions, thereby optimizing the understanding of user query intent; Query expansion unit: used to integrate the user's selected answer or input information into the initial query, expand and improve the query content, and provide more accurate data support for the generation of the final answer; Knowledge answering unit: used to locate problems and generate final decision answers based on the expanded query through the reasoning capabilities of the knowledge graph and large model.

Images