Generalizable wireless network fault decision generation method and system

Abstract

The present invention relates to the technical field of wireless network fault processing, and provides a generalizable wireless network fault decision generation method and system. The method comprises: acquiring fault phenomenon text information; and using a pre-trained generalizable decision generation model to process the acquired fault phenomenon text information to obtain a fault decision. In the present invention, relevant knowledge is used to construct decision analysis to reduce hallucinations, and a quality verification mechanism for a domain-specific decision analysis process is incorporated, thereby ensuring that the analysis process fully incorporates domain knowledge related to an input fault, ensuring the quality of the analysis process, and improving the stability and reliability of the model. A fault phenomenon is mapped to a domain-specific basic problem, knowledge related to the basic problem is retrieved from a knowledge base to assist in generating a decision analysis process, and the strong text analysis processing capability of a large model is used to generate a final decision on the basis of the analysis process, thereby improving the generalization capability of the model.

Description

A generalizable wireless network fault decision generation method and system TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless network fault processing, and particularly relates to a generalizable wireless network fault decision generation method and system based on domain knowledge verification. BACKGROUND

[0002] With the development of wireless network technology, online social interaction and remote conference have become the mainstream way of life communication for people. The increasing demand for wireless communication quality leads to the increasing demand for the efficiency of wireless network domain fault processing. In view of the problems such as poor signal quality, communication drop and low traffic in wireless communication, wireless network domain fault processing experts need to spend a lot of time and effort to analyze and solve problems. Therefore, the wireless network field urgently needs intelligent decision-making methods to improve the efficiency of fault processing.

[0003] There are three challenges in domain decision generation: strong domain, decision illusion and poor generalization of decision model. For the difficulty of strong domain, the technical means that can be solved is the injection of domain knowledge. Domain knowledge injection refers to integrating professional knowledge and experience into the decision model. The existing integration methods include training fine-tuning and combining external knowledge sources, and the purpose is to improve the intelligent level and accuracy of the decision model in the professional field. For the problem of decision illusion, the analysis process of fault decision can be improved by thinking chain technology to improve the reliability and explainability of the analysis process. However, for the challenge of poor generalization of the decision model, there is still a lack of effective solutions.

[0004] The Chinese invention patent application with publication number CN118036721A discloses a cognitive method based on a large language model, which can adaptively obtain problem text corresponding to problem related information, and use a problem cognitive model and the problem related information to analyze and plan the problem, so as to generate target cognitive result text for assisting users in making decisions. However, according to the prompt statement, the analysis and planning of the problem play a crucial role in improving the accuracy and reliability of the target text, but the detailed 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 with the help of related information cannot guarantee to contain sufficient domain knowledge, and it is necessary to verify its rationality in the domain. In addition, the analysis process generated by the method is different from the model used to generate the target cognitive text, and considering that the large language model consumes a lot of computing resources, it is not convenient for practical use.

[0005] The Chinese invention patent with the authorization announcement number CN117785542B discloses an intelligent operation and maintenance method for a data center. According to the existing historical fault information, the mapping relationship is established through the preset reasoning algorithm, which indeed solves the fault problem to a certain extent. However, this method relies too much on historical data. For new and never-before-seen fault problems, it is usually difficult to produce effective decisions because the corresponding mapping relationship cannot be found from the historical data. This leads to poor generalization ability of the technical solution. In addition, since this method requires a large amount of historical data as input, the size and quality of the data set are required to be high, and the accuracy and efficiency of the model in generating fault decisions are greatly affected by the data set. Therefore, the intelligent operation and maintenance method relying on historical data does not work well in practical applications. SUMMARY

[0006] The purpose of the present application 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 in the background art.

[0007] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

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

[0009] Obtaining fault phenomenon text information;

[0010] Processing the obtained fault phenomenon text information using a pre-trained generalizable decision generation model to obtain a fault decision; wherein the fault phenomenon text Q and the generalization prompt instruction are serially concatenated as a generalization prompt template, a locally deployed large language model is called to process, and the output text is used as the generalized domain basic question; the fault phenomenon text is vectorized and encoded by a word embedding model, the cosine similarity of 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, and the relevant knowledge vectors are decoded by the word embedding model to obtain detailed knowledge text; the generalized basic question is processed in the same way to obtain generalized knowledge text; the detailed knowledge text and the generalized knowledge text are input into the large language model as known information, and the output text is used as the generated final decision.

[0011] Further, the fault phenomenon generalization, i.e. using the generalization prompt to guide the language model to map the detailed fault question to the basic question, comprises: serially concatenating the user-uploaded fault phenomenon text Q and the generalization prompt instruction 1 as a generalization prompt template, calling a locally deployed large language model ChatGLM3-6B, and outputting the model text as the generalized domain basic question.

[0012] Further, the user-uploaded fault phenomenon text is vectorized and encoded by the word embedding model bge-large-zh, the cosine similarity of 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, and the relevant knowledge vectors are decoded by the word embedding model to obtain detailed knowledge text; the generalized basic question is processed in the same way to obtain generalized knowledge text.

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

[0014] Further, a plurality of knowledge segments related to the fault question are retrieved from the domain vector library by a similarity matching method, and finally the effectiveness of the analysis process is verified, the cosine similarity of the analysis process analysis and each relevant knowledge segment k i is calculated, and the maximum value is selected as the analysis process matching degree score s: s = max (cosine_similarity (analysis, k i ))

[0015] A decision quality qualified threshold value theta is set, and if the matching degree score exceeds the qualified threshold value, it indicates that the analysis process is effective, otherwise it indicates that it is invalid.

[0016] Further, the thought chain prompt template, the verified effective analysis process, the uploaded fault phenomenon text and the task instruction prompt 3 are sequentially spliced to input a large language model ChatGLM3-6B, and the model output text is taken as a generated final decision.

[0017] In a second aspect, the present application provides a generalizable wireless network fault decision generation system, comprising:

[0018] An acquisition module is configured to acquire fault phenomenon text information.

[0019] The processing module processes the obtained fault phenomenon text information by using a pre-trained generalizable decision generation model to obtain a fault decision; wherein, the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called to process, and the output text is taken as a generalized domain basic question; the fault phenomenon text is vectorized and encoded by a word embedding model, the cosine similarity of 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, and the detailed knowledge text is obtained after the relevant knowledge vectors are decoded by the word embedding model; 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 taken as known information, input into the large language model, and the output text is taken as the generated final decision.

[0020] In a third aspect, the present application provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implement the domain knowledge verification based generalizable wireless network fault decision generation method of the first aspect.

[0021] In a fourth aspect, the present application provides a computer device comprising a memory and a processor, the processor and the memory being in communication with each other, the memory storing program instructions executable by the processor, and the processor invoking the program instructions to execute the domain knowledge verification based generalizable wireless network fault decision generation method of the first aspect.

[0022] In a fifth aspect, the present application provides an electronic device comprising a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the instructions for implementing the domain knowledge verification based generalizable wireless network fault decision generation method of the first aspect.

[0023] Terminology explanation:

[0024] Chain-of-Thought (CoT) is a technology for improving the performance of large language models in performing complex reasoning tasks. The core idea is to simulate the human thinking and reasoning process, chain the steps of solving the problem, and gradually deduce the answer in the form of text, and finally get the final answer according to 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 hand.

[0025] Retrieval-Augmented Generation (RAG) is a technique that combines retrieval and generation, using information from external data sources to assist text generation. Retrieval is retrieving relevant documents or information fragments from a large document set, knowledge base or search engine. Generation is concatenating the retrieved information with the original question into a generative language model to obtain answers or generated results. Retrieval enhancement technology 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.

[0026] The present application has the following advantages: the related knowledge is used to construct the decision analysis to reduce its illusion, the field decision analysis process quality verification mechanism is added, the analysis process is ensured to fully contain the field knowledge related to the input fault, and the quality of the analysis process is ensured, and the stability and reliability of the model are improved; the fault phenomenon is mapped to a field basic problem, the basic problem related knowledge is retrieved from the knowledge base to assist the generation of the decision analysis process, and the powerful text analysis processing capability of the large model is used to generate the final decision by using the analysis process, and the generalization of the model is improved.

[0027] Additional aspects and advantages of the application will be described in the following description, which will become apparent from the description, or will be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0029] Fig. 1 is a flow chart of the wireless network fault decision generation method based on field knowledge verification according to the embodiment of the present application.

[0030] Fig. 2 is a functional principle block diagram of the generalizable decision generation model based on the thinking chain technology according to the embodiment of the present application. DETAILED DESCRIPTION

[0031] The prior art generally assists in generating a decision analysis process through disassembling sub-tasks or step-by-step questioning, but the analysis process generated in this progressive manner is poor in stability and quality cannot be guaranteed, thereby affecting the accuracy of the final decision, and therefore the rationality in the professional field needs to be further verified. In addition, the prior art generally injects domain knowledge into the model through fine-tuning or retrieval enhancement generation. However, both of these methods rely on the content and quality of historical data and cannot solve new faults that do not appear in historical data, and are poor in generalization. The present application proposes a new domain intelligent decision generation method: first, based on the thought chain prompt, guide the large model to generate an analysis process as the basis for fault handling. Then, by retrieving the domain knowledge base, intervene in the large model to generate a decision analysis process, and propose a domain analysis process quality verification mechanism to ensure the reliability and stability of the analysis process. Among them, for the analysis process that does not pass the verification, the specific description of the fault problem is mapped to the domain basic problem, and the retrieval enhancement generation technology is used to correct the analysis process until it passes the verification. Finally, based on the optimized analysis process, the large model is used to generate a final decision with generalization. In addition to using related knowledge to construct a decision analysis to reduce its illusion, the present application also adds a domain decision analysis process quality verification mechanism to ensure that the analysis process fully contains the 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 present application maps the fault phenomenon to the domain basic problem, retrieves the basic problem related knowledge from the knowledge base to assist in generating the decision analysis process, and uses the powerful text analysis and processing capabilities of the large model to generate the final decision, thereby improving the generalization of the model.

[0032] Embodiment 1

[0033] In this embodiment 1, first, a generalizable wireless network fault decision generation system is provided, including: an acquisition module for acquiring fault phenomenon text information. A processing module processes the acquired fault phenomenon text information using a pre-trained generalizable decision generation model to obtain a fault decision; wherein the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called to process, and the output text is used as the generalized domain basic problem; the fault phenomenon text is vectorized and encoded by a word embedding model, the cosine similarity of the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated, and the three segments with the highest cosine similarity are selected as the relevant knowledge vectors. The detailed knowledge text is obtained after decoding the relevant knowledge vectors by the word embedding model; the generalized basic problem is processed in the same way to obtain the generalization knowledge text; the detailed knowledge text and the generalization knowledge text are input into the large language model as known information, and the output text is used as the generated final decision.

[0034] As shown in FIG. 1, in this embodiment 1, a wireless network fault decision generation method based on domain knowledge verification is implemented by using the above system, which comprises the following steps:

[0035] Firstly, upload the fault phenomenon and splice with the thinking chain prompt. According to the reported wireless network fault phenomenon, the user uploads the fault phenomenon text to the model. The model automatically retrieves the "fault-analysis process-decision" as the main structure, which contains the fixed 5 specific thinking chain prompt templates 5shot_cot_prompt, and splices them with the user input fault phenomenon to obtain the input text of the large language model.

[0036] Secondly, fault phenomenon generalization, that is, using the generalization prompt to guide the language model to map the detailed fault problem to the basic problem. The user uploaded fault phenomenon text Q and generalization prompt instruction 1 "Please extract the problem described in the following text in a phrase or a short sentence. Text: {Q}" are spliced in series as a generalization prompt template, and the locally deployed large language model ChatGLM3-6B is called. The model output text is used as the generalized domain basic problem.

[0037] Thirdly, relevant domain knowledge retrieval. The user uploaded fault phenomenon text is vectorized and encoded by the word embedding model bge-large-zh, the cosine similarity of the fault phenomenon vector and the knowledge slices in the local knowledge vector library is calculated, and the three slices with the highest cosine similarity are selected as the relevant knowledge vector. The relevant knowledge vector is decoded by the word embedding model to obtain detailed knowledge text. The generalized basic problem is processed in the same way to obtain generalized knowledge text.

[0038] Fourthly, analysis process generation based on retrieval enhancement generation. The detailed knowledge text and generalized knowledge text obtained in the third step are used as known information knowledge, the user uploaded fault phenomenon is used as problem Q, and task instruction prompt 2 "instruction: please analyze and reason the following problem step by step according to the known information. Please return the result in the form of #analysis process: xxx, and the analysis process contains multiple steps and needs to be distinguished by line breaks. \nKnown information: {knowledge}\nQuestion: {Q}" is spliced in text form according to task instruction prompt 2, and input into the large language model ChatGLM3-6B. The model output is used as the generated analysis process analysis.

[0039] Fifthly, analysis process quality verification. A plurality of knowledge segments related to the fault problem are retrieved from the domain vector library by similarity matching method. Finally, the effectiveness of the analysis process is verified, and the analysis process analysis is calculated with each related knowledge segment k icosine similarity, and the maximum value is selected as the analysis process matching score s: s = max(cosine_similarity(analysis, k i ))

[0040] A decision quality threshold θ is set. If the matching score exceeds the threshold, the analysis process is valid, otherwise it is invalid. For invalid analysis processes that do not pass the verification, it is necessary to return to the fourth step and re-execute until the verification is passed to obtain a valid analysis process.

[0041] In the sixth step, the decision generation based on the thought chain. The input text containing the thought chain prompt template obtained in the first step, the valid analysis process verified in the fifth step, and the task instruction prompt 3 "{5shot_cot_prompt}\n\nInstruction: Please summarize the final solution according to 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, and the model output text is input into the large language model ChatGLM3-6B. The model output text is used as the generated final decision.

[0042] In this embodiment, for the problem of strong domain of fault problem and illusion of decision, a domain knowledge fusion analysis process quality detection mechanism is set. First, the fault related domain knowledge is retrieved from the local knowledge base, and then the cosine similarity between the analysis process and the domain knowledge is calculated to verify the quality of the analysis process. The domain knowledge is injected without cost, while the stability and reliability of the analysis process are guaranteed.

[0043] For the difficulty of poor generalization of the decision model, considering that different fault problems have closed set characteristics in the domain, this embodiment maps the specific fault problem to the domain basic problem, uses the related knowledge of the basic problem and the specific fault problem to correct the analysis process of the fault decision, and finally generates the final decision using the corrected analysis process, improving the generalization of the decision model. Among them, mapping the specific fault problem to the domain basic problem includes: splicing the fault problem description text and 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. The generalization prompt is used to guide the language model to map the detailed fault problem to the basic problem, and the model output text is used as the domain basic problem.

[0044] Because the decision generated directly by the general large language model has the illusion phenomenon in the domain and the problem of poor stability of the decision analysis process, it cannot effectively solve the domain problem. Therefore, this embodiment designs a quality verification mechanism considering the domain relevance for the generated analysis process. The mechanism includes three parts: an external knowledge source, a retriever, and an effectiveness verification module.

[0045] First, a local domain knowledge vector library is constructed as an external knowledge source to provide a strong reference for the model. Then, multiple knowledge fragments related to the fault problem are retrieved from the domain vector library through similarity matching. Finally, the effectiveness of the analysis process is verified by calculating the cosine similarity between the analysis and each relevant knowledge fragment k i , and selecting the maximum value as the analysis process matching score s: s = max(cosine_similarity(analysis, k i ))

[0046] A decision quality threshold θ is set, and if the matching score exceeds the threshold, the analysis process is effective, otherwise it is not. For the analysis process that does not pass the verification, it needs to be iterated and optimized until it passes the verification. The steps of constructing the local domain knowledge vector library include: first, collect non-formatted domain documents: crawl publicly available wireless network fault handling files and professional maintenance personnel written fault case files into pdf, txt, docx or any other format, and place them in the same folder. Second, load and read all documents: split the document content with parameters (chunk_size = 200, chunk_overlap = 50). Third, generate text vectors: send the slice sequence obtained in the second step to the word embedding model bge-large-zh to vectorize and encode the slices. Fourth, construct a local vector library: use Chroma in the langchain framework as a vector library and add the vectorized and encoded knowledge slices from the third step to the vector library. Fifth, store the knowledge vector library locally.

[0047] The above verification mechanism reduces the illusion phenomenon and ensures the stability and reliability of the analysis process.

[0048] Embodiment 2

[0049] Traditional retrieval augmented generation technology retrieves relevant information from external knowledge sources based on input questions and directly generates answers using the information. However, this approach performs poorly when dealing with multi-step reasoning decision tasks and cannot handle faults that do not exist in the knowledge base, lacking generalization. This embodiment combines the thinking chain technology with the retrieval augmented generation technology to solve the above problems and proposes a generalizable decision generation model based on the thinking chain technology, as shown in Figure 2.

[0050] In this embodiment, the generalizable decision generation model based on the thinking chain technology includes the following detailed processes in specific applications:

[0051] 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 "fault-analysis process-decision" as the main structure, which contains five specific thinking chain prompt templates 5shot_cot_prompt.

[0052] The second step is to generalize the fault phenomenon, that is, to use the generalization prompt to guide the language model to map the detailed fault problem to the basic problem. The user uploaded fault phenomenon text Q and generalization prompt instruction 1 "Please extract the problem described in the following text in a phrase or a short sentence. Text: {Q}" are concatenated in series as a generalization prompt template. The locally deployed large language model ChatGLM3-6B is called, and the model output text is used as the generalized domain basic problem.

[0053] The third step is to retrieve relevant domain knowledge. The user uploaded fault phenomenon text is vectorized and encoded by 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 slices with the highest cosine similarity are selected as the relevant knowledge vectors. The relevant knowledge vectors are decoded by the word embedding model to obtain detailed knowledge text. The generalized basic problem is processed in the same way to obtain generalized knowledge text.

[0054] The fourth step is to generate an analysis process based on the retrieved enhanced generation. The detailed knowledge text and generalized knowledge text obtained in the third step are used as known information knowledge. The user uploaded fault phenomenon is used as the problem Q. The task instruction prompt 2 "instruction: please analyze and reason the following problem step by step based on the known information. Please return the result in the form of #analysis process: xxx. The analysis process includes multiple steps and needs to be distinguished by line breaks. \n Known information: {knowledge}\n Question: {Q}" is concatenated in text form according to the task instruction prompt 2. The large language model ChatGLM3-6B is input, and the model output is used as the generated analysis process analysis.

[0055] The fifth step is to verify the quality of the analysis process. Through the method of similarity matching, multiple knowledge segments related to the fault problem are retrieved from the domain vector library. Finally, the effectiveness of the analysis process is verified. The cosine similarity between the analysis process analysis and each relevant knowledge segment k i is calculated, and the maximum value is selected as the analysis process matching degree score s: s = max (cosine_similarity (analysis, k i ))

[0056] A decision quality threshold θ is set, and a matching score exceeding the threshold indicates that the analysis process is valid, otherwise it is invalid. For invalid analysis processes that do not pass the verification, it is necessary to return to the fourth step and re-execute until the verification is passed to obtain a valid analysis process.

[0057] In the sixth step, decision generation based on the thought chain. The thought chain prompt template obtained in the first step, the valid analysis process verified in the fifth step, the user uploaded fault phenomenon text and task instruction prompt 3 "{5shot_cot_prompt}\n\nInstruction: please summarize the final solution according to the problem and analysis process. Please return the result in the form of #solution: xxx.\n#analysis process: {analysis}\nQuestion: {Q}." according to the order in task instruction prompt 3, input the large language model ChatGLM3-6B, and the model output text is the generated final decision.

[0058] To improve the performance of the model in executing the decision generation task, the embodiment constructs a thought chain prompt example based on the "fault-analysis process-decision" template, and promotes the large model to think and generate the analysis process and the final decision as the basis through prompt learning.

[0059] In order to improve the generalization ability of the model, and considering the difference between general large language models and domain knowledge. According to the closed set characteristics of the fault phenomenon, the embodiment first uses the large model to map the fault phenomenon into the domain basic problem, so as to retrieve the characteristic knowledge conforming to the original fault from the knowledge base. Then, based on the retrieval enhancement generation technology, the related knowledge of the fault phenomenon and the basic problem in the knowledge base is retrieved as additional input information of the model, to assist the large language model to generate the analysis process containing the domain characteristics.

[0060] In order to ensure the reliability of the analysis process, the embodiment uses a verification mechanism to verify the validity of the analysis process and iteratively optimize the analysis process. Finally, the valid analysis process containing the mapped domain knowledge is used to generate a final decision with generalization ability by using the large model.

[0061] Embodiment 3

[0062] The embodiment 3 provides a non-transitory computer readable storage medium for storing computer instructions, which are executed by a processor to implement the generalizable wireless network fault decision generation method based on domain knowledge verification as described above, and the method comprises:

[0063] Obtaining fault phenomenon text information;

[0064] The pre-trained generalizable decision generation model is used for processing the obtained fault phenomenon text information, and a fault decision is obtained; wherein, the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called for processing, and the output text is taken as a generalized domain basic question; the fault phenomenon text is vectorized and coded through a word embedding model, the cosine similarity of 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, and the detailed knowledge text is obtained after the relevant knowledge vectors are decoded through the word embedding model; the generalized basic question is processed in the same way to obtain a generalized knowledge text; and the detailed knowledge text and the generalized knowledge text are taken as known information, input into the large language model, and the output text is taken as the generated final decision.

[0065] Embodiment 4

[0066] The embodiment 4 provides a computer device, comprising a memory and a processor, the processor and the memory are in communication with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the generalizable wireless network fault decision generation method based on domain knowledge verification as described above, which comprises:

[0067] Obtaining fault phenomenon text information;

[0068] The pre-trained generalizable decision generation model is used for processing the obtained fault phenomenon text information, and a fault decision is obtained; wherein, the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called for processing, and the output text is taken as a generalized domain basic question; the fault phenomenon text is vectorized and coded through a word embedding model, the cosine similarity of 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, and the detailed knowledge text is obtained after the relevant knowledge vectors are decoded through the word embedding model; the generalized basic question is processed in the same way to obtain a generalized knowledge text; and the detailed knowledge text and the generalized knowledge text are taken as known information, input into the large language model, and the output text is taken as the generated final decision.

[0069] Embodiment 5

[0070] The embodiment 5 provides an electronic device, comprising a processor, a memory and a computer program; wherein, the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the instructions for realizing the generalizable wireless network fault decision generation method based on domain knowledge verification as described above, which comprises:

[0071] Obtaining fault phenomenon text information;

[0072] The obtained fault phenomenon text information is processed by using a pre-trained generalizable decision generation model to obtain a fault decision; wherein, the fault phenomenon text Q is serially spliced with a generalization prompt instruction as a generalization prompt template, a locally deployed large language model is called to process, and a text is output as a generalized domain basic question; the fault phenomenon text is vectorized and coded by a word embedding model, the cosine similarity of 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, and the relevant knowledge vectors are decoded by the word embedding model to obtain detailed knowledge text; the generalized basic question is processed in the same way to obtain generalized knowledge text; the detailed knowledge text and the generalized knowledge text are input into the large language model as known information, and a text is output as the generated final decision.

[0073] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can 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.

[0074] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the 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 produce a device that implements the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0075] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including instruction devices that implement the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0076] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks.

[0077] The above description of the specific embodiments of the present application is merely illustrative and is not intended to limit the scope of the application. Those skilled in the art should understand that various modifications and variations can be made to the disclosed technology without departing from the scope of the application.

Claims

1. A method of generating a generalizable wireless network failure decision, the method comprising: Comprise: Obtain fault phenomenon text information; Use the pre-trained generalizable decision generation model to process the obtained fault phenomenon text information, and obtain fault decision; wherein, the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called to process, and the output text is taken as the generalized domain basic question; The fault phenomenon text is vectorized and coded by a word embedding model, the cosine similarity of 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, and the detailed knowledge text is obtained after decoding the relevant knowledge vectors by the word embedding model; the generalized domain basic question is processed in the same way to obtain the generalized knowledge text; the detailed knowledge text and the generalized knowledge text are taken as known information, and input into the large language model, and the output text is taken as the generated final decision.

2. The method of claim 1, wherein, Fault phenomenon generalization, that is, using a generalization prompt to guide a language model to map a detailed fault question to a basic question, comprises: serially splicing the user-uploaded fault phenomenon text Q and the generalization prompt instruction 1 as a generalization prompt template, calling a locally deployed large language model ChatGLM3-6B, and taking the model output text as the generalized domain basic question.

3. The method of claim 1, wherein, The user-uploaded fault phenomenon text is vectorized and coded by a word embedding model bge-large-zh, the cosine similarity of 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, and the detailed knowledge text is obtained after decoding the relevant knowledge vectors by the word embedding model; the generalized domain basic question is processed in the same way to obtain the generalized knowledge text.

4. The method of claim 3, wherein, The detailed knowledge text and the generalized knowledge text are taken as known information knowledge, the user-uploaded fault phenomenon is taken as a question Q, and the task instruction prompt 2 is spliced in text form, input into the large language model ChatGLM3-6B, and the model output is taken as the generated analysis process analysis.

5. The method of claim 4, wherein, By the method of similarity matching, retrieve multiple knowledge fragments related to the fault problem from the domain vector library, and finally verify the effectiveness of the analysis process. By calculating the cosine similarity between the analysis process analysis and each relevant knowledge fragment k i , select the maximum value as the analysis process matching degree score s: s = max (cosine_similarity (analysis, k i )) A decision quality qualified threshold θ is set, and a matching degree score exceeding the qualified threshold indicates that the analysis process is valid, otherwise it is invalid.

6. The method of claim 5, wherein, The thought chain prompt template, the verified valid analysis process, the uploaded fault phenomenon text and the task instruction prompt 3 are sequentially spliced, input into the large language model ChatGLM3-6B, and the model output text is taken as the generated final decision.

7. A generalizable wireless network failure decision generation system, comprising: Comprise: An acquisition module for acquiring fault phenomenon text information; A processing module for processing the acquired fault phenomenon text information using a pre-trained generalizable decision generation model to obtain fault decision; wherein, the fault phenomenon text Q and the generalization prompt instruction are serially spliced as a generalization prompt template, a locally deployed large language model is called to process, and the output text is taken as the generalized domain basic question; The fault phenomenon text is vectorized and coded by a word embedding model, the cosine similarity of 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, and the detailed knowledge text is obtained after decoding the relevant knowledge vectors by the word embedding model; The generalized domain basic question is processed in the same way to obtain a generalized knowledge text; the detailed knowledge text and the generalized knowledge text are input into the large language model as known information, and the output text is used as a generated final decision.

8. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is used to store computer instructions, which are executed by a processor to implement the generable wireless network fault decision generation method of any one of claims 1-6.

9. A computer device, comprising: A device includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the generable wireless network fault decision generation method of any one of claims 1-6.

10. An electronic device, comprising: It includes: A processor, a memory, and a computer program; wherein the processor is connected with 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 to make the electronic device execute the instructions for implementing the generable wireless network fault decision generation method of any one of claims 1-6.

Images