Generalized wireless network fault decision generation method and system

Abstract

The invention provides a generalizable wireless network fault decision generation method and system, and belongs to the technical field of wireless network fault processing, and the method comprises the steps: obtaining fault phenomenon text information; and processing the obtained fault phenomenon text information by using a pre-trained generalization decision generation model to obtain a fault decision. According to the method, decision analysis is constructed by using related knowledge to reduce illusion, a domain decision analysis process quality verification mechanism is added, it is ensured that the analysis process fully contains domain knowledge related to input faults, then the quality of the analysis process is ensured, and the stability and reliability of the model are improved; fault phenomena are mapped into field basic problems, basic problem related knowledge is retrieved from a knowledge base to assist in generating the decision analysis process, the powerful text analysis processing capacity of a large model is utilized, a final decision is generated through the analysis process, and the generalization of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless network fault processing, and in particular to a generalizable wireless network fault decision-making generation method and system based on domain knowledge verification. Background Art

[0002] With the development of wireless network technology, online social networking and remote meetings have become the mainstream way of communication. People's higher requirements for wireless communication quality have brought about higher requirements for troubleshooting efficiency in the wireless network field. In order to solve problems such as poor signal quality, dropped calls, and low traffic in wireless communication, wireless network troubleshooting experts need to spend a lot of time and energy to analyze and solve problems. Therefore, the wireless network field urgently needs intelligent decision-making methods to improve troubleshooting efficiency.

[0003] There are three major challenges in domain decision generation: strong domain characteristics, decision hallucinations, and poor generalization of decision models. The technical means to solve the difficulty of strong domain characteristics is the injection of domain knowledge. Domain knowledge injection refers to the integration of professional knowledge and experience into the decision model. Existing integration methods include training fine-tuning and combining external knowledge sources. The purpose is to improve the intelligence level of the model in the professional field and the accuracy of decision-making. For the problem of decision hallucinations, the analysis process of fault decisions can be disassembled through the thinking chain technology to improve the reliability and interpretability of the analysis process. However, there is currently a lack of effective solutions to the challenge of poor generalization of decision models.

[0004] The Chinese invention patent application with publication number CN118036721A discloses a cognitive method based on a large language model, which can adaptively obtain the problem-related information corresponding to the problem text, and use the problem cognitive model and the problem-related information to plan and analyze the problem to generate a target cognitive result text for assisting users in making decisions. However, analyzing and planning the problem according to the prompt statement does play a vital role in improving the accuracy and reliability of the target text, but the refinement of the analysis process is not enough. This progressive approach will reduce the stability of the model. In addition, since the general large language model does not have professional domain knowledge, the target text generated by the model only by introducing relevant information cannot be guaranteed to contain sufficient domain knowledge, and its rationality in the field needs to be verified. In addition, the generation analysis process of this method is different from the model used to generate the target cognitive text. Considering that the large language model consumes computing resources very much, it is not convenient for actual implementation.

[0005] The Chinese invention patent with the authorization announcement number CN117785542B discloses an intelligent operation and maintenance method for a data center. The technical solution of the invention proposes to establish a mapping relationship based on the existing historical fault information through a preset reasoning algorithm, which does solve the fault problem to a certain extent. However, this method relies too much on historical data. For new and unprecedented fault problems, it is usually difficult to make effective decisions because it is impossible to find the corresponding mapping relationship from the historical data. This results in poor generalization ability of the technical solution. In addition, since this method requires a large amount of historical data as input, it has high requirements on the scale and quality of the data set, and the accuracy and efficiency of the model-generated fault decision are greatly affected by the data set. Therefore, this intelligent operation and maintenance method that relies on historical data is not ideal in practical applications. Summary of the invention

[0006] The object of the present invention is to provide a generalizable wireless network fault decision generation method and system based on domain knowledge verification to solve at least one technical problem existing in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a generalizable wireless network fault decision generation method, comprising:

[0009] Get fault phenomenon text information;

[0010] The acquired fault phenomenon text information is processed by using a pre-trained generalizable decision generation model to obtain a fault decision; the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic question; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through a word embedding model to obtain a detailed knowledge text; the generalized basic question is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the output text is used as the generated final decision.

[0011] Furthermore, the fault phenomenon is generalized, that is, the generalized prompts are used to guide the language model to map detailed fault problems into basic problems, including: serially splicing the fault phenomenon text Q uploaded by the user and the generalized prompt instruction 1 as a generalized prompt template, calling the locally deployed large language model ChatGLM3-6B, and the model output text as the generalized domain basic problem.

[0012] Furthermore, the fault phenomenon text uploaded by the user is vectorized and encoded through the word embedding model bge-large-zh, the cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated, and the three fragments with the highest cosine similarity are selected as relevant knowledge vectors. The relevant knowledge vectors are decoded through the word embedding model to obtain a detailed knowledge text; the generalized basic problem is processed in the same way to obtain a generalized knowledge text.

[0013] Furthermore, the detailed knowledge text and the generalized knowledge text are taken as known information knowledge, and the fault phenomenon uploaded by the user is taken as question Q, which is spliced ​​with the task instruction prompt 2 in text form and input into the large language model ChatGLM3-6B, and the model output is taken as the generated analysis process analysis.

[0014] Furthermore, through the similarity matching method, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library. Finally, the effectiveness of the analysis process is verified by calculating the analysis process and each related knowledge fragment k i The cosine similarity of , select the maximum value as the matching score s of the analysis process:

[0015] s=max(cosine_similarity(analysis,k i ))

[0016] A qualified decision quality threshold θ is set. If the matching score exceeds the qualified threshold, it means that the analysis process is effective, otherwise it is invalid.

[0017] Furthermore, the thinking chain prompt template, the verified effective analysis process, the uploaded fault phenomenon text and the task instruction prompt 3 are sequentially spliced ​​and input into the large language model ChatGLM3-6B, and the model outputs the text as the final decision generated.

[0018] In a second aspect, the present invention provides a generalizable wireless network fault decision making system, comprising:

[0019] An acquisition module is used to obtain text information of fault phenomena;

[0020] The processing module uses a pre-trained generalizable decision generation model to process the acquired fault phenomenon text information to obtain a fault decision; wherein, the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic problem; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through the word embedding model to obtain a detailed knowledge text; the generalized basic problem is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the output text is used as the generated final decision.

[0021] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the generalizable wireless network fault decision generation method based on domain knowledge verification as described in the first aspect is implemented.

[0022] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the generalizable wireless network fault decision generation method based on domain knowledge verification as described in the first aspect.

[0023] In a fifth aspect, the present invention provides an electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the generalizable wireless network fault decision generation method based on domain knowledge verification as described in the first aspect.

[0024] Terminology explanation:

[0025] Chain-of-Thought (CoT) is a technology that improves the performance of large language models in complex reasoning tasks. The core idea is to simulate the human thinking and reasoning process, chain together the steps to solve the problem, and gradually deduce the answer in the form of displayed text. Finally, the large model obtains the final answer based on the reasoning process. Chain-of-Thought technology enhances the interpretability of the model on the one hand, and improves the quality of the output on the other.

[0026] Retrieval-Augmented Generation (RAG) is a technology that combines retrieval and generation, using information from external data sources to assist text generation. Retrieval is to retrieve documents or information fragments related to the query from a large document collection, knowledge base or search engine. Generation is to splice the retrieved information with the original question and input it into a generative language model to obtain an answer or generate results. Retrieval-Augmented Generation can utilize a large amount of external knowledge to improve the model's understanding of the problem and improve the accuracy and richness of the answer.

[0027] The beneficial effects of the present invention are as follows: relevant knowledge is used to construct decision analysis to reduce its illusion, and a domain decision analysis process quality verification mechanism is added to ensure that the analysis process fully includes domain knowledge related to the input fault, thereby ensuring the quality of the analysis process and improving the stability and reliability of the model; the fault phenomenon is mapped to a basic problem in the domain, and the knowledge related to the basic problem is retrieved from the knowledge base to assist in generating the decision analysis process, and the powerful text analysis and processing capabilities of the large model are used to generate the final decision using the analysis process, thereby improving the generalization of the model.

[0028] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0030] Figure 1 This is a flow chart of a generalizable wireless network fault decision generation method based on domain knowledge verification according to an embodiment of the present invention.

[0031] Figure 2 This is a functional principle block diagram of the generalizable decision generation model based on thought chain technology described in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0033] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0034] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0035] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0036] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. Different embodiments or examples described in this specification and features of different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

[0037] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0038] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0039] The prior art generally assists in generating the decision analysis process by disassembling subtasks or asking questions step by step. However, the stability of this progressively generated analysis process is poor, and the quality cannot be guaranteed, which in turn affects the accuracy of the final decision. Therefore, it is necessary to further verify its rationality in the professional field. In addition, the prior art generally injects domain knowledge into the model by fine-tuning or retrieval enhancement generation. However, both methods rely on the content and quality of historical data, cannot solve new faults that do not appear in historical data, and have poor generalization. The present invention proposes a new domain intelligent decision generation method: first, based on the thinking chain prompt, the large model is guided to generate an analysis process as a basis for fault handling. Then, by retrieving the domain knowledge base, the large model is intervened in the decision analysis process generated, and a domain analysis process quality verification mechanism is proposed to ensure the reliability and stability of the analysis process. Among them, for the analysis process that has not passed the verification, the specific description of the fault problem is mapped to the basic problem of the domain, and the analysis process is corrected by using the retrieval enhancement generation technology until the verification is passed. Finally, based on the optimized analysis process, the large model is used to generate a generalized final decision. In addition to using relevant knowledge to construct decision analysis to reduce its illusion, the technical solution of the present invention also adds a domain decision analysis process quality verification mechanism to ensure that the analysis process fully contains domain knowledge related to the input fault, thereby ensuring the quality of the analysis process and improving the stability and reliability of the model. The technical solution of the present invention improves the generalization of the model by mapping the fault phenomenon to the basic problems of the domain, retrieving the knowledge related to the basic problems from the knowledge base to assist in generating the decision analysis process, and using the powerful text analysis and processing capabilities of the large model to generate the final decision through the analysis process.

[0040] Example 1

[0041] In this embodiment 1, a generalizable wireless network fault decision generation system is first provided, including: an acquisition module, which is used to acquire fault phenomenon text information. A processing module, which uses a pre-trained generalizable decision generation model to process the acquired fault phenomenon text information to obtain a fault decision; wherein, the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the text is output as the generalized domain basic problem; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through the word embedding model to obtain a detailed knowledge text; the generalized basic problem is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the text is output as the generated final decision.

[0042] like Figure 1As shown, in this embodiment 1, the above system is used to implement a generalizable wireless network fault decision generation method based on domain knowledge verification, including the following steps:

[0043] The first step is to upload the fault phenomenon and connect it with the thought chain prompt. The user uploads the fault phenomenon text to the model according to the reported wireless network fault phenomenon. The model automatically retrieves the thought chain prompt template 5shot_cot_prompt, which has a main structure of "fault-analysis process-decision" and contains 5 fixed specific examples, and connects it with the fault phenomenon input by the user to obtain the input text of the large language model.

[0044] The second step is to generalize the fault phenomenon, that is, to use generalized prompts to guide the language model to map detailed fault problems into basic problems. The fault phenomenon text Q uploaded by the user and the generalized prompt instruction 1 "Please refine the problem described in the text below in a phrase or a short sentence. Text: {Q}" are serially concatenated as a generalized prompt template, and the locally deployed large language model ChatGLM3-6B is called. The model outputs the text as the generalized domain basic problem.

[0045] The third step is to retrieve relevant domain knowledge. The fault phenomenon text uploaded by the user is vectorized and encoded through the word embedding model bge-large-zh. The cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated. The three segments with the highest cosine similarity are selected as relevant knowledge vectors. The relevant knowledge vectors are decoded through the word embedding model to obtain detailed knowledge text. The generalized basic problem is processed in the same way to obtain the generalized knowledge text.

[0046] Step 4: Generate the analysis process based on retrieval enhancement generation. The detailed knowledge text and generalized knowledge text obtained in step 3 are used as known information knowledge, and the fault phenomenon uploaded by the user is used as question Q. They are combined with task instruction prompt 2: "Instruction: Please analyze and reason the following questions step by step based on the known information. Please return the results in the form of #Analysis process: xxx. The analysis project contains multiple steps that need to be separated by line breaks. \nKnown information: {knowledge}\nQuestion: {Q}" according to task instruction prompt 2 in text form, input into the large language model ChatGLM3-6B, and the model output is used as the generated analysis process analysis.

[0047] The fifth step is to verify the quality of the analysis process. Through the similarity matching method, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library. Finally, the effectiveness of the analysis process is verified by calculating the analysis process analysis and each related knowledge fragment k i The cosine similarity of , select the maximum value as the matching score s of the analysis process:

[0048] s=max(cosine_similarity(analysis,k i ))

[0049] Set the decision quality qualified threshold θ. If the matching score exceeds the qualified threshold, it means that the analysis process is valid, otherwise it is invalid. For invalid analysis processes that fail to pass the verification, it is necessary to return to the fourth step and re-execute until the verification passes and a valid analysis process is obtained.

[0050] Step 6: Decision generation based on thought chain. The input text containing the thought chain prompt template obtained in the first step, the effective analysis process verified in the fifth step, and the task instruction prompt 3 "{5shot_cot_prompt}\n\nInstruction: Please summarize the final solution based on the problem and analysis process. Please return the result in the form of #Solution: xxx.\n#Analysis process: {analysis}\nQuestion: {Q}" are spliced ​​in the order of task instruction prompt 3, input into the large language model ChatGLM3-6B, and the model output text is used as the generated final decision.

[0051] In this embodiment, in order to address the problems of strong domain nature of fault problems and illusions in decision-making, a quality detection mechanism for the analysis process that integrates domain knowledge is set up. First, fault-related domain knowledge is retrieved from the local knowledge base, and then the quality of the analysis process is tested by calculating the cosine similarity between the analysis process and the domain knowledge. This ensures the stability and reliability of the analysis process while injecting domain knowledge at no cost.

[0052] In view of the difficulty of poor generalization of decision models, and considering that different fault problems have closed-set characteristics in the field, this embodiment improves the generalization of decision models by mapping specific fault problems into basic problems in the field, using the relevant knowledge of basic problems and specific fault problems, correcting the analysis process of fault decision-making, and finally using the corrected analysis process to generate the final decision. Among them, mapping specific fault problems into basic problems in the field includes: splicing the fault problem description text with the generalization instruction prompt "Please summarize the problem described in the above text in a phrase or a short sentence." as a generalization prompt, and calling the locally deployed large language model ChatGLM3-6B. Use the generalization prompt to guide the language model to map detailed fault problems into basic problems, and the model outputs text as the basic problem in the field.

[0053] Since the decisions directly generated by the general large language model have hallucination phenomena in the domain and poor stability of the decision analysis process, it cannot effectively solve domain problems. Therefore, this embodiment designs a quality verification mechanism that considers domain relevance for the generated analysis process. The mechanism consists of three parts: external knowledge source, retriever, and validity verification module.

[0054] First, a local domain knowledge vector library is constructed as an external knowledge source to provide a domain-specific reference for the model. Then, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library through the similarity matching method. Finally, the effectiveness of the analysis process is verified by calculating the analysis process and each relevant knowledge fragment k i The cosine similarity of , select the maximum value as the matching score s of the analysis process:

[0055] s=max(cosine_similarity(analysis,k i ))

[0056] Set the decision quality qualified threshold θ. If the matching score exceeds the qualified threshold, it means that the analysis process is valid, otherwise it is invalid. For the analysis process that has not passed the verification, it is necessary to iterate and optimize it repeatedly until it passes the verification. Among them, the steps of building a local domain knowledge vector library include: the first step is to collect unformatted domain documents: crawl the public wireless network fault handling files and the fault case files written by professional maintenance personnel into any format such as pdf, txt, docx, etc., and put them in the same folder. The second step is to load and read all documents: split the document content with the parameters of (chunk_size=200, chunk_overlap=50). The third step is to generate text vectors: the slice sequence obtained in the second step is fed into the word embedding model bge-large-zh to vectorize the slice. The fourth step is to build a local vector library: use Chroma in the langchain framework as the vector library, and add the knowledge slices that have been vectorized in the third step to the vector library. The fifth step is to store the knowledge vector library locally.

[0057] The above verification mechanism alleviates the hallucination phenomenon and ensures the stability and reliability of the analysis process.

[0058] Example 2

[0059] Traditional retrieval-enhanced generation technology retrieves relevant information from external knowledge sources based on input questions and uses the information to directly generate answers. However, this method performs poorly when dealing with decision-making tasks that require multi-step reasoning, and cannot handle faults that do not exist in the knowledge base, and does not have generalization. This embodiment combines the thought chain technology with the retrieval-enhanced generation technology to solve the above problems, and proposes a generalizable decision generation model based on the thought chain technology. The model structure is as follows: Figure 2 shown.

[0060] In this embodiment, the generalizable decision generation model based on the thought chain technology, in a specific application, the detailed process of the model working includes:

[0061] The first step is to upload the fault phenomenon and call the thinking chain prompt template. The user uploads the fault phenomenon text to the model according to the reported wireless network fault phenomenon. The model automatically calls the thinking chain prompt template 5shot_cot_prompt, which has a main structure of "fault-analysis process-decision-making" and contains 5 fixed specific examples.

[0062] The second step is to generalize the fault phenomenon, that is, to use generalized prompts to guide the language model to map detailed fault problems into basic problems. The fault phenomenon text Q uploaded by the user and the generalized prompt instruction 1 "Please refine the problem described in the text below in a phrase or a short sentence. Text: {Q}" are serially concatenated as a generalized prompt template, and the locally deployed large language model ChatGLM3-6B is called. The model outputs the text as the generalized domain basic problem.

[0063] The third step is to retrieve relevant domain knowledge. The fault phenomenon text uploaded by the user is vectorized and encoded through the word embedding model bge-large-zh. The cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated. The three segments with the highest cosine similarity are selected as relevant knowledge vectors. The relevant knowledge vectors are decoded through the word embedding model to obtain detailed knowledge text. The generalized basic problem is processed in the same way to obtain the generalized knowledge text.

[0064] Step 4: Generate the analysis process based on retrieval enhancement generation. The detailed knowledge text and generalized knowledge text obtained in step 3 are used as known information knowledge, and the fault phenomenon uploaded by the user is used as question Q. They are combined with task instruction prompt 2: "Instruction: Please analyze and reason the following questions step by step based on the known information. Please return the results in the form of #Analysis process: xxx. The analysis project contains multiple steps that need to be separated by line breaks. \nKnown information: {knowledge}\nQuestion: {Q}" according to task instruction prompt 2 in text form, input into the large language model ChatGLM3-6B, and the model output is used as the generated analysis process analysis.

[0065] The fifth step is to verify the quality of the analysis process. Through the similarity matching method, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library. Finally, the effectiveness of the analysis process is verified by calculating the analysis process analysis and each related knowledge fragment k i The cosine similarity of , select the maximum value as the matching score s of the analysis process:

[0066] s=max(cosine_similarity(analysis,k i ))

[0067] Set the decision quality qualified threshold θ. If the matching score exceeds the qualified threshold, it means that the analysis process is valid, otherwise it is invalid. For invalid analysis processes that fail to pass the verification, it is necessary to return to the fourth step and re-execute until the verification passes and obtains a valid analysis process.

[0068] Step 6: Decision generation based on thought chain. The thought chain prompt template obtained in the first step, the effective analysis process verified in the fifth step, the fault phenomenon text uploaded by the user and the task instruction prompt 3 "{5shot_cot_prompt}\n\nInstruction: Please summarize the final solution based on the problem and analysis process. Please return the result in the form of #Solution: xxx.\n#Analysis process: {analysis}\nQuestion: {Q}" are spliced ​​in the order of task instruction prompt 3, input into the large language model ChatGLM3-6B, and the model output text is used as the generated final decision.

[0069] In order to improve the performance of the model in executing decision-making generation tasks, this embodiment constructs a thinking chain prompt example based on the "fault-analysis process-decision" template, and prompts the large model to think and generate the analysis process and final decision as the basis through prompt learning.

[0070] In order to improve the generalization ability of the model, the difference between the general large language model and the domain knowledge is taken into consideration. According to the closed set characteristics of the fault phenomenon, this embodiment first uses the large model to map the fault phenomenon into the domain basic problem, so as to retrieve the characteristic knowledge that conforms to the original fault from the knowledge base. Then, based on the retrieval enhancement generation technology, the relevant knowledge of the fault phenomenon and the basic problem is retrieved from the knowledge base as additional input information of the model, to assist the large language model in generating the analysis process containing the domain characteristics.

[0071] In order to ensure the reliability of the analysis process, this embodiment uses a verification mechanism to verify the effectiveness of the analysis process and iteratively optimizes the analysis process. Finally, an effective analysis process including mapped domain knowledge is used to generate a generalized final decision using a large model.

[0072] Example 3

[0073] This embodiment 3 provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the generalizable wireless network fault decision generation method based on domain knowledge verification as described above is implemented. The method includes:

[0074] Get fault phenomenon text information;

[0075] The acquired fault phenomenon text information is processed by using a pre-trained generalizable decision generation model to obtain a fault decision; the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic question; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through a word embedding model to obtain a detailed knowledge text; the generalized basic question is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the output text is used as the generated final decision.

[0076] Example 4

[0077] This embodiment 4 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, the processor calls the program instructions to execute the above-mentioned generalizable wireless network fault decision generation method based on domain knowledge verification, the method comprising:

[0078] Get fault phenomenon text information;

[0079] The acquired fault phenomenon text information is processed by using a pre-trained generalizable decision generation model to obtain a fault decision; the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic question; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through a word embedding model to obtain a detailed knowledge text; the generalized basic question is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the output text is used as the generated final decision.

[0080] Example 5

[0081] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the above-mentioned generalizable wireless network fault decision generation method based on domain knowledge verification, the method comprising:

[0082] Get fault phenomenon text information;

[0083] The acquired fault phenomenon text information is processed by using a pre-trained generalizable decision generation model to obtain a fault decision; the fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic question; the fault phenomenon text is vectorized and encoded through a word embedding model, and the cosine similarity between the fault phenomenon vector and the knowledge slice in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through a word embedding model to obtain a detailed knowledge text; the generalized basic question is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information, input into the large language model, and the output text is used as the generated final decision.

[0084] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0086] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0088] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative work on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A generalizable wireless network fault decision generation method, characterized in that: include: Get fault phenomenon text information; Fault processing: Use the pre-trained generalizable decision generation model to process the acquired fault phenomenon text information to obtain a fault decision. The fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic problem. The fault phenomenon text is vectorized and encoded through the word embedding model, the cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated, the three fragments with the highest cosine similarity are selected as the relevant knowledge vectors, and the relevant knowledge vectors are decoded through the word embedding model to obtain the detailed knowledge text; the generalized basic problem is processed in the same way to obtain the generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information and input into the large language model, and the output text is used as the final decision generated.

2. The generalizable wireless network fault decision generation method according to claim 1, characterized in that: Fault phenomenon generalization, that is, using generalized prompts to guide the language model to map detailed fault problems into basic problems, including: serially splicing the fault phenomenon text Q uploaded by the user and the generalized prompt instruction 1 as a generalized prompt template, calling the locally deployed large language model ChatGLM3-6B, and the model output text as the generalized domain basic problem.

3. The generalizable wireless network fault decision generation method according to claim 1, characterized in that: The fault phenomenon text uploaded by the user is vectorized and encoded through the word embedding model bge-large-zh, and the cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated. The three fragments with the highest cosine similarity are selected as relevant knowledge vectors, and the relevant knowledge vectors are decoded through the word embedding model to obtain the detailed knowledge text; the generalized basic problem is processed in the same way to obtain the generalized knowledge text.

4. The generalizable wireless network fault decision generation method according to claim 3, characterized in that: The detailed knowledge text and the generalized knowledge text are taken as known information knowledge, the fault phenomenon uploaded by the user is taken as question Q, which is concatenated with the task instruction prompt 2 in text form and input into the large language model ChatGLM3-6B. The model output is used as the generated analysis process analysis.

5. The generalizable wireless network fault decision generation method according to claim 4, characterized in that: Through the similarity matching method, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library. Finally, the effectiveness of the analysis process is verified by calculating the analysis process and each related knowledge fragment k i The cosine similarity of , select the maximum value as the matching score s of the analysis process: s=max(cosine_similarity(analysis,k i )) A qualified decision quality threshold θ is set. If the matching score exceeds the qualified threshold, it means that the analysis process is effective, otherwise it is invalid.

6. The generalizable wireless network fault decision generation method according to claim 5, characterized in that: The thinking chain prompt template, the verified effective analysis process, the uploaded fault phenomenon text and the task instruction prompt 3 are sequentially spliced ​​and input into the large language model ChatGLM3-6B. The model outputs the text as the final decision generated.

7. A generalizable wireless network fault decision generation system, characterized in that: include: An acquisition module is used to obtain text information of fault phenomena; The processing module uses a pre-trained generalizable decision generation model to process the acquired fault phenomenon text information to obtain a fault decision. The fault phenomenon text Q and the generalized prompt instruction are serially spliced ​​as a generalized prompt template, and the locally deployed large language model is called for processing, and the output text is used as the generalized domain basic problem. The fault phenomenon text is vectorized and encoded through the word embedding model, the cosine similarity between the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated, the three fragments with the highest cosine similarity are selected as the relevant knowledge vectors, and the relevant knowledge vectors are decoded through the word embedding model to obtain the detailed knowledge text; the generalized basic problem is processed in the same way to obtain the generalized knowledge text; the detailed knowledge text and the generalized knowledge text are used as known information and input into the large language model, and the output text is used as the final decision generated.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the generalizable wireless network fault decision-making method according to any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the generalizable wireless network fault decision generation method according to any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the generalizable wireless network fault decision generation method as described in any one of claims 1-6.

Images