Engineering problem analysis and solution method and device based on large model and TRIZ theory, electronic equipment and medium
By combining large models with TRIZ theory and utilizing structured templates and LangChain technology, we have achieved efficient solutions to TRIZ problems, reduced the learning difficulty for users, improved the effectiveness and consistency of TRIZ methods, and solved the problem that traditional TRIZ theory relies on user capabilities.
Patent Information
- Application Number
- CN202410868312.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-01
- Publication Date
- 2024-11-01
AI Technical Summary
The application of traditional TRIZ theory requires users to have a deep knowledge of science and engineering, which leads to high learning pressure, and the results depend on the user's personal ability, making it difficult to guarantee effectiveness.
By using the fundamental knowledge of TRIZ theory as a knowledge source, a fine-tuning dataset is created using a large model and structured templates to train the large model. The knowledge source is stored in a knowledge base through LangChain. User input is received, processed, and answers are obtained. The similarity threshold and Top-k knowledge information of the large model are used to generate prompt text to improve the accuracy of the answers.
It reduces the learning difficulty of TRIZ theory for users, improves the effectiveness and consistency of TRIZ methods, reduces reliance on users' personal abilities, and enhances the efficiency and accuracy of TRIZ problem-solving.
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of information technology, and in particular to a method, device, electronic device and medium for analyzing and solving engineering problems based on large models and TRIZ theory. Background Art
[0002] The tools of TRIZ were developed for problem solving and generating innovative ideas. TRIZ is a comprehensive system with a variety of techniques that can be divided into several groups:
[0003] 1. System problem analysis tool
[0004] 2. Tools to prepare resources for solving problems or generating ideas
[0005] 3. Tools that enhance users’ imagination and creativity
[0006] 4. Tools for generating new ideas
[0007] 5. Tools for analyzing the evolution of technological systems
[0008] After technical personnel have a deeper understanding of TRIZ theory, they can apply this method to solve technical problems more efficiently.
[0009] Using TRIZ theory can effectively solve engineering problems, but the effectiveness of using TRIZ tools depends on the user's personal abilities, including their basic knowledge of science and engineering, problem analysis skills, research skills, and creativity. Even if users follow the same TRIZ guidelines, their results may vary depending on their personal abilities.
[0010] The traditional application model requires users to have a deep understanding of TRIZ theory and a rich reserve of knowledge in multiple disciplines. This puts users under great learning pressure when applying the theory. Summary of the invention
[0011] An object of embodiments of the present invention is to solve at least the above problems and / or disadvantages and to provide at least the advantages to be described below.
[0012] The embodiments of the present invention provide a method, a device, an electronic device and a storage medium for analyzing and solving engineering problems based on a large model and TRIZ theory.
[0013] To this end, the present invention provides the following technical solutions:
[0014] In the first aspect, a method for analyzing and solving engineering problems based on large models and TRIZ theory includes the following steps:
[0015] 1) Collect the basic knowledge related to TRIZ theory as knowledge source;
[0016] 2) inputting an instruction language model into the instruction language model through a structured template according to the collected knowledge source to produce an instruction fine-tuning dataset, and fine-tuning the large model through the instruction fine-tuning dataset;
[0017] 3) storing the knowledge source into the knowledge base of the large model, and performing relevant training on it according to the large model;
[0018] 4) Receive the content input by the user and process it into the first prompt text;
[0019] 5) Input the first prompt text into the large model trained in step 3) to obtain the answer.
[0020] Preferably, in the method for analyzing and solving engineering problems based on a large model and TRIZ theory, if the content input by the user belongs to the field of TRIZ knowledge question and answer, then in step 5), the knowledge base is searched for the first prompt text to obtain the similarity. If the similarity is greater than the similarity threshold, the answer is obtained by inputting the large model trained in step 3), and the process is terminated, 0<the similarity threshold≤1.
[0021] Preferably, in the method for analyzing and solving engineering problems based on large models and TRIZ theory, in step 2), the structured template is prompt;
[0022] In step 3), the knowledge source is stored in the knowledge base through LangChain, and the knowledge base is stored using the large model. Preferably, in the method for analyzing and solving engineering problems based on the large model and TRIZ theory, if the similarity is less than the similarity threshold, then step S6 is entered:
[0023] 6) Obtain a user doubt summary based on the first prompt text and the historical information text, obtain the Top-k knowledge information in the knowledge base based on the first prompt text and the user doubt summary, generate a second prompt text based on the user confusion summary, the Top-k knowledge information and a preset large model prompt template, and input the second prompt text into the large model trained in step 3) to obtain an answer.
[0024] Preferably, in the method for analyzing and solving engineering problems based on large models and TRIZ theory, if the problem input by the user is to seek an ideal solution, in step 4), the content input by the user processes the user's problem into a first prompt text, including:
[0025] If the length of the text of the question input by the user is insufficient, try to match the template of the relevant type, and apply the relevant template between the user input as the first prompt text. If the question is long enough, obtain the question type of TRIZ theory related questions based on the question input by the user, and generate the first prompt text based on the question and question type.
[0026] Preferably, in the method for analyzing and solving engineering problems based on large models and TRIZ theory, when the content input by the user belongs to the analysis of ideal solutions, the background of the ideal solution of the large model TRIA is imported in a structured manner and used as the background knowledge of the first prompt text;
[0027] Import the analysis steps of the ideal solution of the large model in a structured way and use them as the operation steps of the first prompt text, which is expressed as: Please apply the TRIZ ideal solution step by step and provide practical steps;
[0028] Import analysis examples of ideal solutions of large models in a structured way and use them as examples of first prompt text operations;
[0029] After enabling the command ideal solution, the user proceeds to the following steps:
[0030] Remind users to enter the system to be analyzed;
[0031] After the user enters the name of the system to be analyzed, it is combined with the first prompt text and submitted to the big model to get the answer;
[0032] The current analysis result is combined with the second prompt text to represent the new text and submitted to the big model to wait for the expanded answer.
[0033] Preferably, in the method for analyzing and solving engineering problems based on large models and TRIZ theory, if the problem input by the user is to seek an ideal solution, the basic knowledge related to TRIZ theory includes the following types: basic concepts of ideal solutions, concepts of TRIZ9 screen model analysis, concepts of functional analysis and component interaction analysis, concepts of root cause analysis, concepts of causal chain analysis, application of cutting and cutting rules, concepts of cutting and cutting rule application, application of contradiction matrix and innovation principle, concepts of technical shield and concepts of physical contradiction;
[0034] In step 3), the large model is trained according to different types of basic knowledge related to the TRIZ theory.
[0035] Preferably, the method for analyzing and solving engineering problems based on large models and TRIZ theory further includes the following steps:
[0036] 7) Based on the user's evaluation of the returned results, feedback reinforcement training is performed on the large model.
[0037] Second, a device for analyzing and solving engineering problems based on large models and TRIZ theory, including:
[0038] A module that collects basic knowledge related to TRIZ theory as a knowledge source;
[0039] A module for producing an instruction fine-tuning dataset by inputting an instruction language model through a structured template according to the collected knowledge source, and performing fine-tuning training on the large model through the instruction fine-tuning dataset;
[0040] A module for storing the knowledge source in a structured manner into the knowledge base of the large model and performing relevant training on it according to the large model;
[0041] A module that receives user input and processes it into the first prompt text;
[0042] The module that inputs the first prompt text into the trained large model to obtain the answer.
[0043] Optionally, if the content input by the user belongs to the field of TRIZ knowledge question and answer, then in step 5), the knowledge base is searched for the first prompt text to obtain the similarity. If the similarity is greater than the similarity threshold, the answer is input into the trained large model to terminate the process, 0<the similarity threshold≤1.
[0044] Optionally, the structured template is prompt;
[0045] The knowledge source is stored in the knowledge base through LangChain, and is stored in the knowledge base using a large model.
[0046] Optionally, if the similarity is less than a similarity threshold, a user doubt summary is obtained based on the first prompt text and the historical information text, Top-k knowledge information in the knowledge base is obtained based on the first prompt text and the user doubt summary, a second prompt text is generated based on the user confusion summary, the Top-k knowledge information and a preset large model prompt template, and the second prompt text is input into the trained large model to obtain the answer.
[0047] Optionally, if the question input by the user is to seek an ideal solution, in step 4), the content input by the user processes the user's question into a first prompt text, including:
[0048] If the length of the text of the question input by the user is insufficient, try to match the template of the relevant type, and apply the relevant template between the user input as the first prompt text. If the question is long enough, obtain the question type of TRIZ theory related questions based on the question input by the user, and generate the first prompt text based on the question and question type.
[0049] Optionally, when the content input by the user belongs to the ideal solution analysis aspect, the background of the ideal solution of the large model TRIA is imported in a structured manner and used as the background knowledge of the first prompt text;
[0050] Import the analysis steps of the ideal solution of the large model in a structured way and use them as the operation steps of the first prompt text, which is expressed as: Please apply the TRIZ ideal solution step by step and provide practical steps;
[0051] Import analysis examples of ideal solutions of large models in a structured way and use them as examples of first prompt text operations;
[0052] After enabling the command ideal solution, the user proceeds to the following steps:
[0053] Remind users to enter the system to be analyzed;
[0054] After the user enters the name of the system to be analyzed, it is combined with the first prompt text and submitted to the big model to get the answer;
[0055] The current analysis result is combined with the second prompt text to represent the new text and submitted to the big model to wait for the expanded answer.
[0056] Optionally, when the problem input by the user is to seek an ideal solution, the basic knowledge related to TRIZ theory includes the following types: basic concept of ideal solution, concept of TRIZ9 screen model analysis, concept of function analysis and component interaction analysis, concept of root cause analysis, concept of cause-effect chain analysis, application of cutting and cutting rules, concept of cutting and cutting rules application, application of contradiction matrix and innovation principle, concept of technical shield and concept of physical contradiction;
[0057] In step 3), the large model is trained according to different types of basic knowledge related to the TRIZ theory.
[0058] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes any one of the methods described.
[0059] According to a fourth aspect, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, any of the methods described above is implemented.
[0060] The embodiments of the present invention have at least the following beneficial effects:
[0061] The present invention can improve the effectiveness of learning and using the TRIZ method, and is used to reduce the problem that users need to have a deeper understanding of the theory of TRIZ before applying it in practice.
[0062] Other advantages, objectives and features of the embodiments of the present invention will be reflected in part through the following description, and in part will be understood by those skilled in the art through study and practice of the embodiments of the present invention. DETAILED DESCRIPTION
[0063] The embodiments of the present invention are further described in detail below in conjunction with the embodiments so that those skilled in the art can implement the embodiments according to the description.
[0064] The present invention provides a method for analyzing and solving engineering problems based on a large model and TRIZ theory, which is characterized by comprising the following steps:
[0065] In the first aspect, a method for analyzing and solving engineering problems based on large models and TRIZ theory includes the following steps:
[0066] 1) Collect the basic knowledge related to TRIZ theory as knowledge source;
[0067] 2) inputting an instruction language model into the instruction language model through a structured template according to the collected knowledge source to produce an instruction fine-tuning dataset, and fine-tuning the large model through the instruction fine-tuning dataset;
[0068] 3) storing the knowledge source into the knowledge base of the large model, and performing relevant training on it according to the large model;
[0069] 4) Receive the content input by the user and process it into the first prompt text;
[0070] 5) Input the first prompt text into the large model trained in step 3) to obtain the answer.
[0071] In one embodiment of the present invention, preferably, if the content input by the user belongs to the aspect of TRIZ knowledge question and answer, then in step 5), the knowledge base is searched for the first prompt text to obtain the similarity. If the similarity is greater than the similarity threshold, the answer is input into the large model trained in step 3) to terminate the process, 0<the similarity threshold≤1.
[0072] In one embodiment of the present invention, preferably, in step 2), the structured template is prompt;
[0073] In step 3), the knowledge source is stored in the knowledge base through LangChain, and the knowledge base is stored using the large model. Preferably, in the method for analyzing and solving engineering problems based on the large model and TRIZ theory, if the similarity is less than the similarity threshold, then step S6 is entered:
[0074] 6) Obtain a user doubt summary based on the first prompt text and the historical information text, obtain the Top-k knowledge information in the knowledge base based on the first prompt text and the user doubt summary, generate a second prompt text based on the user confusion summary, the Top-k knowledge information and a preset large model prompt template, and input the second prompt text into the large model trained in step 3) to obtain an answer.
[0075] In one embodiment of the present invention, as a preference, when the question input by the user is to seek an ideal solution, in step 4), the content input by the user processes the question of the user into a first prompt text, including:
[0076] If the text of the question input by the user is not long enough (e.g. less than 5 characters, and its meaning cannot be fully understood), then try to match the template of the relevant type, and apply the relevant template between the user input and supplement it as the first prompt text. If the question is long enough, then according to the question input by the user, obtain the question type of TRIZ theory related questions, and generate the first prompt text according to the question and question type.
[0077] In one embodiment of the present invention, preferably, when the content input by the user belongs to the ideal solution analysis, the background of the ideal solution of the large model TRIA is imported in a structured manner and used as the background knowledge of the first prompt text;
[0078] Import the analysis steps of the ideal solution of the large model in a structured way and use them as the operation steps of the first prompt text, which is expressed as: Please apply the TRIZ ideal solution step by step and provide practical steps;
[0079] Import analysis examples of ideal solutions of large models in a structured way and use them as examples of first prompt text operations;
[0080] After enabling the command ideal solution, the user proceeds to the following steps:
[0081] Remind users to enter the system to be analyzed;
[0082] After the user enters the name of the system to be analyzed, it is combined with the first prompt text and submitted to the big model to get the answer;
[0083] The current analysis result is combined with the second prompt text to represent the new text and submitted to the big model to wait for the expanded answer.
[0084] In one embodiment of the present invention, as a preference, when the problem input by the user is to seek an ideal solution, the basic knowledge related to the TRIZ theory includes the following types: basic concepts of ideal solutions, concepts of TRIZ9 screen model analysis, concepts of functional analysis and component interaction analysis, concepts of root cause analysis, concepts of causal chain analysis, application of cutting and cutting rules, concepts of cutting and cutting rule application, application of contradiction matrix and innovation principle, concepts of technical shield and concepts of physical contradiction;
[0085] In step 3), the large model is trained according to different types of basic knowledge related to the TRIZ theory.
[0086] In one embodiment of the present invention, as a preferred embodiment, the method for analyzing and solving engineering problems based on large models and TRIZ theory further includes the following steps:
[0087] 7) Based on the user's evaluation of the returned results, feedback reinforcement training is performed on the large model.
[0088] In a second aspect, the present invention provides a device for analyzing and solving engineering problems based on a large model and TRIZ theory, comprising:
[0089] A module that collects basic knowledge related to TRIZ theory as a knowledge source;
[0090] A module for producing an instruction fine-tuning dataset by inputting an instruction language model through a structured template according to the collected knowledge source, and performing fine-tuning training on the large model through the instruction fine-tuning dataset;
[0091] A module for storing the knowledge source in a structured manner into the knowledge base of the large model and performing relevant training on it according to the large model;
[0092] A module that receives user input and processes it into the first prompt text;
[0093] The module that inputs the first prompt text into the trained large model to obtain the answer.
[0094] In one embodiment of the present invention, a method for analyzing and solving engineering problems based on large models and TRIZ theory, preferably based on the TRIZ knowledge question and answer aspect of the large model, comprises the steps of:
[0095] S1. Obtain and clean the basic knowledge related to TRIZ theory. Cleaning includes data structured processing, deduplication, and other abnormal data processing. The theoretical knowledge related to TRIZ is obtained from the TRIZ-related theoretical textbooks compiled by the applicant, which contain various concepts, terms, examples, solutions and other information.
[0096] S2. Create a fine-tuning instruction dataset based on the relevant knowledge of TRIZ theory, and use the fine-tuning instruction dataset to fine-tune the large model.
[0097] Step S2 specifically includes:
[0098] S21, design prompt, input TRIZ theoretical knowledge information into the instruction language model through prompt to obtain a fine-tuned knowledge data set. In this embodiment, prompt is a structured template and its set, and the language model is an existing model.
[0099] S22. Use model fine-tuning tools or methods such as LoRA, PEFT, Freeze, etc. to fine-tune the model.
[0100] S3. Store TRIZ knowledge information into the database through LangChain, and use the big model to perform embedding training on the knowledge information.
[0101] S4. Receive the user input question and process the user input question into the first prompt text; specifically, obtain the type of the user input question, if the question length is sufficient, try to match the relevant template, apply the relevant template between the user input supplement as the first prompt text, for example, if the user inputs "contradiction matrix", it will eventually be converted into "help me find the concept and application scenario of contradiction matrix". If the question length is sufficient, obtain the relevant question type according to the user input question, and generate the first prompt according to the question and question type.
[0102] S5. Perform a LangChain search on the first prompt text to obtain the similarity. If the similarity is greater than the first similarity threshold, the answer is obtained and the process ends; if the similarity is less than the first similarity threshold, proceed to step S6.
[0103] S6. According to the first prompt text and context history information, combine and obtain other Topk knowledge information in LangChain, apply the corresponding template to generate a new prompt, and input the prompt into the big model, and return the obtained result to the user as output.
[0104] S7. Based on the user’s evaluation of the returned results, the large model is trained with feedback reinforcement.
[0105] In one embodiment of the present invention, a method for analyzing and solving engineering problems based on a large model and TRIZ theory, preferably, based on the analysis of generating an ideal solution based on a large model, comprises the steps of:
[0106] S1. Import the background of the ideal solution of the large model TRIZ in a structured way and use it as the background knowledge of the first prompt. Its text is expressed as:
[0107] [Background] The ideal technical system in TRIZ is a system that provides the main functions without and does not have any harmful side effects.
[0108] [Example 1] An ideal lawn mower cuts grass to the desired length without and does not have any harmful side effects.
[0109] [Example 2] An ideal car key can open and close the car, start and stop the car, without or with any harmful side effects.
[0110] [Directive] to develop an ideal technical framework for {system}”
[0111] S2. Import the analysis steps of the ideal solution of the large model in a structured manner and use it as the operation step of the first prompt. Its text is expressed as:
[0112] Please apply the TRIZ ideal solution step by step:
[0113] [Step 1] Ask the user to describe the technical system you are interested in. (This can be an existing system or a hypothetical system, (wait for user input)
[0114] [STEP 2] Next, apply four key strategies to improve the desirability of this technology system.
[0115] Strategy 1: Reduce or eliminate harmful effects and undesirable functions.
[0116] Strategy 2: Add useful features and benefits.
[0117] Strategy 3: Maintain or reduce the resources required to achieve useful functions.
[0118] Strategy 4: Add new useful features to the system.
[0119] For each of these 4 strategies, provide specific suggestions on how to develop or improve the technology system to enhance its desirability.
[0120] Finally, for each recommended direction of development, identify any key challenges or contradictions that may need to be addressed.”
[0121] S3. Import an analysis example of an ideal solution of a large model in a structured way and use it as a first prompt operation example.
[0122] S4. After the user activates the command ideal solution, proceed to the following steps:
[0123] S41. Remind the user to enter the system to be analyzed.
[0124] S42. After the user enters the name of the system to be analyzed, it is combined with the first prompt and submitted to the large model to obtain the answer.
[0125] S43. Combine the current analysis result with the second prompt, expressed as “Are there other systems that can achieve the same or similar useful main functions?” and submit it to the big model to wait for an expanded answer.
[0126] Structured here means that the background knowledge is assembled in a table-like or JSON format, with column headers and content, which is convenient for computers or AI to understand accurately.
[0127] For example, a method for analyzing and solving engineering problems based on large models and TRIZ theory adopts the following technical solutions:
[0128] S1. Collect the basic knowledge related to TRIZ theory as knowledge source.
[0129] S2. Create an instruction fine-tuning dataset based on the collected TRIZ theoretical knowledge, and use the instruction fine-tuning dataset to fine-tune the large model.
[0130] S3. Use LangChain to store TRIZ knowledge information into the knowledge base and perform relevant training based on the big model.
[0131] S4. Receive the content input by the user and process the user's TRIZ theory related question as the first prompt.
[0132] S5. Perform a LangChain search on the first prompt text to obtain similarity; if the similarity is greater than the first similarity threshold, obtain the answer and end the process; if the similarity is less than the first similarity threshold, proceed to step S6, 0<the similarity threshold≤1:
[0133] S6. Obtain a user confusion summary based on the first prompt text and the historical information text, obtain Topk knowledge information in LangChain based on the first prompt text and the user confusion summary, generate a third prompt text based on the user confusion summary, Topk knowledge information and a preset large model prompt template, and input the third prompt text into the large model to obtain an answer.
[0134] For example, based on the large model to generate the ideal solution analysis type, the following technical solutions are adopted:
[0135] S1. Preset the basic concept of ideal solution in the large model as the background knowledge of the first prompt.
[0136] S2. Preset the general analysis steps of the ideal solution in the large model as the first prompt operation steps
[0137] S3. Preset practical application examples of ideal solutions and expected results in the large model as examples of the first prompt.
[0138] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0139] For example, based on the large model, the current state or status of the technical system analyzed by TRIZ 9 screen model is as follows:
[0140] S1. Preset the TRIZ9 screen model analysis concept in the big model as the background knowledge for the first prompt.
[0141] S2. Preset the general analysis steps of TRIZ9 screen in the large model as the operation steps of the first prompt
[0142] S3. Preset the actual application example of TRIZ9 screen and the expected result in the large model as an example of the first prompt.
[0143] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0144] For example, based on the type of function analysis and component interaction analysis of the large model, the following technical solutions are adopted:
[0145] S1. Preset the concepts of functional analysis and component interaction analysis in the large model as the background knowledge of the first prompt.
[0146] S2. Preset the general analysis steps of functional analysis and component interaction analysis in the large model as the operation steps of the first prompt
[0147] S3. Preset practical application examples of functional analysis and component interaction analysis in the large model and the expected results as examples of the first prompt.
[0148] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0149] For example, based on the type of root cause analysis (RCA) of the large model, the following technical solutions are adopted:
[0150] S1. Preset the concept of root cause analysis (RCA) in the big model as the background knowledge of the first prompt.
[0151] S2. Preset the general analysis steps of root cause analysis (RCA) in the large model as the first prompt operation steps
[0152] S3. Preset the actual application example of root cause analysis (RCA) and the expected results in the large model as an example of the first prompt.
[0153] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0154] For example, a type of causal chain analysis (CECA) based on a large model is provided, using the following technical solution:
[0155] S1. Preset the concept of Causal Chain Analysis (CECA) in the big model as the background knowledge of the first prompt.
[0156] S2. Preset the general analysis steps of Causal Chain Analysis (CECA) in the large model as the first prompt operation steps
[0157] S3. Preset the actual application example of Causal Chain Analysis (CECA) and the expected results in the large model as an example of the first prompt.
[0158] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0159] For example, based on the type of cropping of the large model and the application of cropping rules, the following technical solutions are adopted:
[0160] S1. Preset the concepts of cutting and cutting rules application in the large model as the background knowledge of the first prompt.
[0161] S2. Preset the general analysis steps of cutting and cutting rules application in the large model as the first prompt operation steps
[0162] S3. Preset cropping in the large model and apply actual application examples of cropping rules and expected results as examples of the first prompt.
[0163] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0164] For example, based on the type of cropping of the large model and the application of cropping rules, the following technical solutions are adopted:
[0165] S1. Preset the concepts of cutting and cutting rules application in the large model as the background knowledge of the first prompt.
[0166] S2. Preset the general analysis steps of cutting and cutting rules application in the large model as the first prompt operation steps
[0167] S3. Preset cropping in the large model and apply actual application examples of cropping rules and expected results as examples of the first prompt.
[0168] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0169] For example, based on the 40 innovative principles of the big model and the types of applications of the contradiction matrix, the following technical solutions are adopted:
[0170] S1. Preset the concepts of innovation principle and contradiction matrix in the big model as the background knowledge for the first prompt.
[0171] S2. Preset the general analysis steps of applying the contradiction matrix and innovation principle in the large model as the operation steps of the first prompt
[0172] S3. Preset practical application examples of innovation principles and contradiction matrix applications and expected results in the big model as examples of the first prompt.
[0173] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0174] For example, based on the analysis and solution of large model technical contradictions, the following technical solutions are adopted:
[0175] S1. The concept of presetting the technical shield in the large model serves as the background knowledge for the first prompt.
[0176] S2. Preset the general analysis steps of technical contradictions in the large model as the first prompt operation steps
[0177] S3. Preset practical application examples of technical contradiction analysis and resolution and expected results in the large model as examples of the first prompt.
[0178] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0179] For example, based on the type of large model physical contradiction analysis and resolution, the following technical solutions are adopted:
[0180] S1. Preset the concept of physical contradiction in the large model as the background knowledge for the first prompt.
[0181] S2. Preset the general analysis steps of physical contradictions in the large model as the first prompt operation steps
[0182] S3. Preset practical application examples of physical contradiction analysis and resolution and expected results in the large model as examples of the first prompt.
[0183] S4. When the user uses this method, the user inputs the large model into the system according to the first prompt and the technical system to be analyzed entered by the user to obtain the answer.
[0184] An embodiment of the present invention also provides an electronic device. The electronic device includes: at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the method described. Specifically, the above-mentioned memory and processor are connected together via a bus, and can be a general memory and processor, which is not specifically limited here. When the processor runs the computer program stored in the memory, it can perform the various operations and functions described in the embodiments of the present invention.
[0185] The embodiment of the present invention further provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented. The specific implementation can be found in the method embodiment, which will not be described in detail here.
[0186] Although the embodiments of the present invention are disclosed above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the embodiments of the present invention. For those familiar with the art, additional modifications can be easily realized. Therefore, without departing from the general concept defined by the claims and equivalent scope, the embodiments of the present invention are not limited to the specific details and the embodiments shown and described here.
Claims
1. A method for analyzing and solving engineering problems based on large models and TRIZ theory, characterized in that: The steps include: 1) Collect the basic knowledge related to TRIZ theory as knowledge source; 2) inputting an instruction language model into the instruction language model through a structured template according to the collected knowledge source to produce an instruction fine-tuning dataset, and fine-tuning the large model through the instruction fine-tuning dataset; 3) storing the knowledge source into the knowledge base of the large model, and performing relevant training on it according to the large model; 4) Receive the content input by the user and process it into the first prompt text; 5) Input the first prompt text into the large model trained in step 3) to obtain the answer.
2. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 1, characterized in that: If the content input by the user belongs to the TRIZ knowledge question and answer aspect, then in step 5), the knowledge base is searched for the first prompt text to obtain the similarity. If the similarity is greater than the similarity threshold, the answer is input into the trained large model to end the process, 0<the similarity threshold≤1.
3. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 2, characterized in that: In step 2), the structured template is prompt; In step 3), the knowledge source is stored in the knowledge base through LangChain, and the knowledge source stored in the knowledge base is trained by embedding using a large model.
4. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 2, characterized in that: If the similarity is less than the similarity threshold, proceed to step S6: 6) Obtain a user doubt summary based on the first prompt text and the historical information text, obtain the Top-k knowledge information in the knowledge base based on the first prompt text and the user doubt summary, generate a second prompt text based on the user confusion summary, the Top-k knowledge information and a preset large model prompt template, and input the second prompt text into the trained large model to obtain an answer.
5. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 1, characterized in that: If the question input by the user is to seek an ideal solution, in step 4), the content input by the user processes the user's question into a first prompt text, including: If the length of the text of the question input by the user is insufficient, try to match the template of the relevant type, and apply the relevant template between the user input as the first prompt text. If the question is long enough, obtain the question type of TRIZ theory related questions based on the question input by the user, and generate the first prompt text based on the question and question type.
6. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 5, characterized in that: When the content input by the user belongs to the ideal solution analysis, the background of the ideal solution of the large model TRIA is imported in a structured manner and used as the background knowledge of the first prompt text; Import the analysis steps of the ideal solution of the large model in a structured way and use them as the operation steps of the first prompt text, which is expressed as: Please apply the TRIZ ideal solution step by step and provide practical steps; Import analysis examples of ideal solutions of large models in a structured way and use them as examples of first prompt text operations; After enabling the command ideal solution, the user proceeds to the following steps: Remind users to enter the system to be analyzed; After the user enters the name of the system to be analyzed, it is combined with the first prompt text and submitted to the big model to get the answer; The current analysis result is combined with the second prompt text to represent the new text and submitted to the big model to wait for the expanded answer.
7. The method for analyzing and solving engineering problems based on large models and TRIZ theory as claimed in claim 1, characterized in that: If the problem input by the user is to seek an ideal solution, the basic knowledge related to TRIZ theory includes the following types: basic concept of ideal solution, concept of TRIZ9 screen model analysis, concept of function analysis and component interaction analysis, concept of root cause analysis, concept of cause-effect chain analysis, application of cutting and cutting rules, concept of cutting and cutting rules application, application of contradiction matrix and innovation principle, concept of technical shield and concept of physical contradiction; In step 3), the large model is trained according to different types of basic knowledge related to the TRIZ theory.
8. A device for analyzing and solving engineering problems based on large models and TRIZ theory, characterized in that: include: A module that collects basic knowledge related to TRIZ theory as a knowledge source; A module for producing an instruction fine-tuning dataset by inputting an instruction language model through a structured template according to the collected knowledge source, and performing fine-tuning training on a large model through the instruction fine-tuning dataset; A module for storing the knowledge source in a structured manner into the knowledge base of the large model and performing relevant training on it according to the large model; A module that receives user input and processes it into the first prompt text; The first prompt text is input into the large model trained in step 3) to obtain the answer module.
9. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor performs the method according to any one of claims 1 to 7.
10. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Innovation knowledge base construction tool, electronic device and computer readable medium
CN112199558A
Auxiliary invention device, electronic equipment and computer readable medium
CN112632335A
Search question and answer method based on large model
CN117609444A
System and method for anticipating advances in technology
US11934968B1
Cited By
Solution generation method and device based on TRIZ and large language model
CN120822601A
