Intelligent question answering method, device and equipment for lithium battery equipment fault diagnosis and medium

Abstract

The invention relates to an intelligent question answering method and device for lithium battery equipment fault diagnosis, equipment and a medium, and the method comprises the steps: obtaining user question data, carrying out the entity recognition of the user question data, and obtaining entity query information; candidate problems are screened in a preset fault knowledge base based on the entity query information, a candidate problem set is obtained, and the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data; performing multi-similarity calculation on each candidate question in the candidate question set and the user question data; screening target candidate questions based on multiple similarity calculation results; and querying answers of the target candidate questions, and feeding back the answers. In the whole process, preliminary screening based on entity query is carried out based on user question data, then more accurate target candidate questions are obtained through calculation by adopting multiple similarities, finally answers of the target candidate questions are queried and fed back, and intelligent question answering of accurate lithium battery equipment fault diagnosis can be achieved.

Description

Technical Field

[0001] The present application relates to the field of intelligent question-answering technology, and in particular to an intelligent question-answering method, device, computer equipment, storage medium and computer program product for lithium battery equipment fault diagnosis. Background Art

[0002] With the rapid development of global science and technology, the demand for traditional energy and environmental issues have become imminent challenges. Countries attach great importance to and develop renewable energy, and lithium batteries have become a key force in adjusting the energy structure. However, the complexity of lithium battery technology and equipment has made it challenging for business personnel to acquire knowledge and hindered technical personnel in operating equipment.

[0003] At present, question answering systems (QA) already exist in other fields to realize relevant knowledge question answering. Specifically, the question answering system is an advanced form of information retrieval, which can accurately understand the questions asked by users in natural language, and return concise and accurate matching answers by searching corpora, knowledge graphs or question-answering knowledge bases. In the current field of natural language processing in artificial intelligence, question answering systems have attracted much attention. Frequently Asked Question (FAQ) is a subtask of the question answering system, which aims to collect and organize frequently asked questions and standard answers from users, and construct a FAQ dataset. When a user asks a new question, the system will retrieve the most similar questions in the FAQ dataset and return the corresponding answers to the user.

[0004] Although there are question-and-answer systems available to answer relevant knowledge, this technology is generally used in scenarios such as online shopping and self-service. There is currently no accurate intelligent question-and-answer solution for lithium battery equipment fault diagnosis scenarios. Summary of the invention

[0005] Based on this, it is necessary to provide an accurate intelligent question-and-answer method, device, computer equipment, storage medium and computer program product for lithium battery equipment fault diagnosis in response to the above technical problems.

[0006] In a first aspect, the present application provides an intelligent question-answering method for lithium battery equipment fault diagnosis. The method comprises:

[0007] Acquire user question data, perform entity recognition on the user question data, and obtain entity query information;

[0008] Based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0009] Performing multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0010] Based on the results of multiple similarity calculations, target candidate questions are screened;

[0011] Query the answer to the target candidate question and provide feedback.

[0012] In one embodiment, before screening candidate questions in a preset fault knowledge base based on the entity query information to obtain a candidate question set, the method further includes:

[0013] Obtaining lithium battery equipment failure operation and maintenance data, wherein the lithium battery equipment failure operation and maintenance data includes a lithium battery equipment manual, a lithium battery equipment instruction manual, and operation and maintenance record data;

[0014] Performing knowledge extraction on the lithium battery equipment fault operation and maintenance data to obtain knowledge extraction data;

[0015] Extracting fault fields from the knowledge extraction data based on preset key fields, wherein the preset key fields include fault code, equipment, component, fault phenomenon, cause, confirmation method, treatment measures, and material spare parts;

[0016] Using a pre-trained language model to convert the fault field into a vector;

[0017] The fault field and the vector are associated and stored to generate a preset fault knowledge base.

[0018] In one embodiment, the method further comprises:

[0019] Receive alarm information sent by lithium battery equipment;

[0020] Parsing the alarm information and extracting the alarm content carried in the alarm information;

[0021] Based on the alarm content, fault handling measures are matched in the preset fault knowledge base to obtain fault handling measure data.

[0022] In one embodiment, matching fault handling measures in a preset fault knowledge base based on the alarm content to obtain fault handling measures data includes:

[0023] According to the alarm content, a rough sorting based on BM25 similarity is performed in the preset fault knowledge base to obtain an initial similar fault description result;

[0024] The initial similar fault description results are sorted for a second time using edit distance similarity and Jaccard similarity, and a fault description result that satisfies a preset edit distance similarity threshold and a preset Jaccard similarity threshold is searched in the initial similar fault description results to obtain a target fault description result;

[0025] The preset fault knowledge base is searched for fault handling measure data corresponding to the target fault description result.

[0026] In one embodiment, performing multiple similarity calculations on each candidate question in the candidate question set and the user question data includes:

[0027] Calculate the BM25 similarity and cosine similarity of each candidate question in the candidate question set and the user question data;

[0028] Based on the BM25 similarity calculation result, each candidate question is arranged in descending order of similarity, and based on the cosine similarity calculation, the descending queue is adjusted to obtain the initial candidate question sorting queue.

[0029] In one embodiment, the method of arranging each candidate question in descending order of similarity based on the BM25 similarity calculation result and adjusting the descending order queue based on the cosine similarity calculation to obtain the initial candidate question sorting queue includes:

[0030] Get the minimum threshold and maximum recall number K of BM25 similarity, as well as the minimum threshold and maximum threshold of cosine similarity;

[0031] Based on the BM25 similarity calculation result, recall K candidate questions whose BM25 similarity is greater than a minimum threshold of BM25 similarity from the candidate questions;

[0032] Based on the cosine similarity calculation result, recall from the candidate questions all questions whose cosine similarity is greater than a minimum threshold of the cosine similarity and are not repeated with the K candidate questions;

[0033] Starting from position K+1, if the cosine similarity corresponding to the current candidate question is not less than the maximum threshold of the cosine similarity, the current candidate question is ranked first to obtain the initial candidate question ranking queue.

[0034] In one embodiment, screening target candidate questions based on multiple similarity calculation results includes:

[0035] Based on the multiple similarity calculation results, the first candidate question is obtained;

[0036] Calculate the cosine similarity between the first candidate question and the user question data;

[0037] If the cosine similarity is not greater than a preset cosine similarity threshold, calculating the Jaccard similarity between the first candidate question and the user question data;

[0038] If the Jaccard similarity is not greater than a preset Jaccard similarity threshold, calculating the edit distance similarity between the first candidate question and the user question data;

[0039] If the edit distance similarity is not greater than a preset edit distance similarity threshold, M candidate questions with relatively high cosine similarities are screened from the multiple similarity calculation results to obtain a target candidate question.

[0040] In a second aspect, the present application also provides an intelligent question-answering device for lithium battery equipment fault diagnosis. The device comprises:

[0041] An entity recognition module is used to obtain user question data, perform entity recognition on the user question data, and obtain entity query information;

[0042] A problem initial screening module, used to screen candidate problems in a preset fault knowledge base based on the entity query information to obtain a candidate problem set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0043] A similarity calculation module, used to perform multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0044] The question secondary screening module is used to screen target candidate questions based on multiple similarity calculation results;

[0045] The query feedback module is used to query the answers to the target candidate questions and provide feedback.

[0046] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0047] Acquire user question data, perform entity recognition on the user question data, and obtain entity query information;

[0048] Based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0049] Performing multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0050] Based on the results of multiple similarity calculations, target candidate questions are screened;

[0051] Query the answer to the target candidate question and provide feedback.

[0052] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0053] Acquire user question data, perform entity recognition on the user question data, and obtain entity query information;

[0054] Based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0055] Performing multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0056] Based on the results of multiple similarity calculations, target candidate questions are screened;

[0057] Query the answer to the target candidate question and provide feedback.

[0058] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0059] Acquire user question data, perform entity recognition on the user question data, and obtain entity query information;

[0060] Based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0061] Performing multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0062] Based on the results of multiple similarity calculations, target candidate questions are screened;

[0063] Query the answer to the target candidate question and provide feedback.

[0064] The intelligent question-and-answer method, device, computer device, storage medium and computer program product for lithium battery equipment fault diagnosis obtains user question data, performs entity recognition on the user question data, and obtains entity query information; based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, and the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data; multiple similarity calculations are performed on each candidate question in the candidate question set and the user question data; based on the multiple similarity calculation results, the target candidate questions are screened; the answers to the target candidate questions are queried and fed back. In the whole process, an initial screening based on entity query is performed based on the user question data, and then multiple similarity calculations are used to obtain more accurate target candidate questions. Finally, the answers to the target candidate questions are queried and fed back, which can realize accurate intelligent question-and-answer for lithium battery equipment fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A diagram of an application environment of an intelligent question-answering method for lithium battery device fault diagnosis in one embodiment;

[0066] Figure 2 A schematic diagram of a flow chart of an intelligent question-answering method for lithium battery device fault diagnosis in one embodiment;

[0067] Figure 3 Construct a process diagram for the preset fault knowledge base;

[0068] Figure 4 A schematic diagram of the equipment fault alarm information matching process;

[0069] Figure 5 A schematic diagram of a flow chart of an intelligent question-answering method for lithium battery device fault diagnosis in another embodiment;

[0070] Figure 6 This is a schematic diagram of the secondary screening process for the problem;

[0071] Figure 7 It is a schematic diagram of the overall process of an intelligent question-answering method for lithium battery equipment fault diagnosis in an application example;

[0072] Figure 8 A structural block diagram of an intelligent question-and-answer device for lithium battery device fault diagnosis in one embodiment;

[0073] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0075] The intelligent question-answering method for lithium battery device fault diagnosis provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The user operates on the terminal 102 (such as a smart phone), and the user terminal 102 sends a lithium battery device fault diagnosis request to the server 104. The server 104 responds to the lithium battery device fault diagnosis request, extracts the user question data carried therein, and performs entity recognition on the user question data to obtain entity query information; based on the entity query information, the candidate questions are screened in the preset fault knowledge base to obtain a candidate question set, and the preset fault knowledge base is constructed based on the lithium battery device fault operation and maintenance data; multiple similarity calculations are performed on each candidate question in the candidate question set and the user question data; based on the multiple similarity calculation results, the target candidate questions are screened; the answers to the target candidate questions are queried and fed back to the terminal 102. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable device may be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0076] In one embodiment, Figure 2 As shown in the figure, an intelligent question-answering method for lithium battery equipment fault diagnosis is provided. Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0077] S100: Obtain user question data, perform entity recognition on the user question data, and obtain entity query information.

[0078] The server 104 receives the fault description or query question input by the user through natural language. It can be text, voice (converted to text through voice recognition) or other forms of data. The received user question data is preprocessed, including removing irrelevant characters (such as punctuation marks, special symbols), stop word filtering, word form restoration, etc., to improve the accuracy of subsequent processing. The named entity recognition (NER) algorithm in the natural language processing (NLP) technology is used to identify entities related to lithium battery equipment failures from user questions, such as equipment models, fault phenomena, component names, etc. These entities will be used as keywords or conditions for the query. Based on the identified entities, a query statement or query condition for querying the preset fault knowledge base is constructed to obtain entity query information.

[0079] Specifically, entity recognition refers to extracting entities such as equipment and components from user questions, and then limiting the search scope to a specific set. First, the equipment names and component names are organized into an entity vocabulary. Then, the key equipment and component entities are quickly extracted using a keyword matching method based on AC automata. At the same time, typo correction and entity disambiguation are introduced to improve the effect of entity recognition.

[0080] S200: Screening candidate questions in a preset fault knowledge base based on entity query information to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data.

[0081] There is a lot of data in the fault knowledge base, but only a small part of the data is related to the user's questions, so the data needs to be screened. The specific approach is: based on the equipment and component entities obtained in the entity recognition stage, fuzzy retrieval is used to find data related to these entities to form the sub-data set required for similarity calculation, that is, to obtain a candidate question set, which contains multiple candidate questions. Specifically, the preset fault knowledge base should contain multi-source data such as historical fault cases, maintenance experience, equipment manuals, etc., and be structured for fast retrieval. Questions that match the entity query information are screened out from the fault knowledge base as candidate questions to form a candidate question set.

[0082] S300: performing multiple similarity calculations on each candidate question in the candidate question set and the user question data.

[0083] Select a suitable similarity calculation model, such as similarity based on TF-IDF, similarity based on pre-trained models such as BERT, etc. According to specific needs, multiple similarity calculation methods can be used in combination to obtain a more comprehensive evaluation. For each candidate question in the candidate question set, multiple similarities are calculated with the user question data. Comprehensively consider the surface similarity of the text (such as keyword matching) and the deep semantic similarity (such as sentence structure, contextual meaning, etc.). Specifically, the above-mentioned multiple similarity calculations can include at least two of cosine similarity, BM25 similarity, Jaccard similarity and edit distance similarity. Furthermore, it can be at least one similarity calculation combination of cosine similarity, similarity with BM25, Jaccard similarity and edit distance similarity.

[0084] The above processing and the characteristics and technical advantages of the above different similarity calculations will be further explained in detail below.

[0085] Calculate the similarity between the user's question and the candidate questions, and return the question or questions with the highest similarity as the candidate questions to the user. The above multiple similarity calculations include at least two of cosine similarity, BM25 similarity, Jaccard similarity, and edit distance similarity. By integrating multiple similarity calculations, the optimal ranking of candidate questions is achieved.

[0086] Cosine similarity measures the difference between two strings by using the cosine value of the angle between two vectors in the vector space. i and the string C j When calculating the cosine similarity of , the Sentence-Transformer pre-trained language model is used to obtain the embedding representation vector of the user query, and then the cosine similarity formula is used to calculate:

[0087]

[0088] Among them, S cos (C i , C j ) represents the string C i and the string C j The cosine similarity of and is the string C i and the string C j The embedding representation vector of and is the magnitude of the vector.

[0089] BM25 is an improved algorithm based on IF-IDF, which takes into account word frequency and document frequency and introduces some adjustment parameters to improve the quality of information retrieval. The calculation formula is:

[0090]

[0091] Score(D, Q) represents the matching score between query Q and document D. First, query Q is segmented to obtain the score of each word q i . Calculate word q i Inverse document frequency Where N is the total number of documents (constant), n(q i ) is a string containing the word q i The number of documents. i , D) represents word q i The frequency of occurrence in document D. k 1 is a positive coefficient that controls the word frequency saturation. b is usually set to 0.75 to control the impact of document length on the score. avgdl is the average length of all documents. |D| is the length of document D.

[0092] Jaccard similarity evaluates the similarity of two sets based on the ratio of the size of the intersection to the size of the union of the two sets. It does not consider the order of the elements in the set, but only focuses on the existence of the elements. In text similarity calculation, a string can be regarded as a set of words, so the similarity of the word set can measure the similarity of the string. The calculation formula is:

[0093]

[0094] Among them, |A| represents the number of elements in set A, |B| represents the number of elements in set B, |A∩B| represents the number of elements in the intersection of A and B, and |A∪B| represents the number of elements in the union of A and B.

[0095] Edit distance is the distance between the string C i Convert to string C j The minimum number of operations required, including insertion, deletion and replacement. The formula for calculating the edit distance similarity is:

[0096]

[0097] Where LD is the string C i To string C j The minimum number of operations required for the conversion, length(C j ) and length(C j ) represent the string C i and the string C j Length, s LD (C i , C j ) represents the string C i and the string C j The edit distance similarity.

[0098] Since each similarity has different importance and similarity calculation requires a certain amount of time, the four similarities are calculated in a serial order instead of calculating the four similarities for all questions at once. Cosine similarity is the most important because the sentence-transformer pre-trained language model can describe text similarity more accurately from a semantic level. Using cosine similarity, you can quickly select candidate questions with the highest semantic similarity. The core of BM25 similarity, Jaccard similarity, and edit distance similarity is the degree of character overlap, which complements cosine similarity.

[0099] S400: Screening target candidate questions based on multiple similarity calculation results.

[0100] According to the multiple similarity calculation results, the questions in the candidate question set are sorted, and the target candidate questions are screened out based on the sorting results. Specifically, when screening the target candidate questions, multiple similarity calculations and preset thresholds can be used again for screening.

[0101] S500: Query the answer to the target candidate question and provide feedback.

[0102] Search the preset fault knowledge base for answers to target candidate questions. Sort the answers to ensure the accuracy and readability of the information. Feedback the sorted answers to users in a user-friendly way, which can be in the form of text, charts, voice, etc.

[0103] The intelligent question-and-answer method for lithium battery equipment fault diagnosis described above obtains user question data, performs entity recognition on the user question data, and obtains entity query information; based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, and the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data; multiple similarity calculations are performed on each candidate question in the candidate question set and the user question data; based on the multiple similarity calculation results, the target candidate questions are screened; the answers to the target candidate questions are queried and fed back. During the entire process, an initial screening based on entity query is performed based on the user question data, and then multiple similarity calculations are used to obtain more accurate target candidate questions. Finally, the answers to the target candidate questions are queried and fed back, which can achieve accurate intelligent question-and-answer for lithium battery equipment fault diagnosis.

[0104] In one embodiment, before screening candidate questions in a preset fault knowledge base based on the entity query information to obtain a candidate question set, the method further includes:

[0105] Obtain lithium battery equipment failure operation and maintenance data, which includes lithium battery equipment manuals, lithium battery equipment instructions and operation and maintenance record data; perform knowledge extraction on the lithium battery equipment failure operation and maintenance data to obtain knowledge extraction data; extract fault fields from the knowledge extraction data based on preset key fields, which include fault code, equipment, components, fault phenomenon, cause, confirmation method, treatment measures, and material spare parts; use a trained language model to convert the fault field into a vector; store the fault field and the vector in association to generate a preset fault knowledge base.

[0106] First, equipment fault manuals, maintenance records and other documents are used as data sources. Then, valuable data is extracted from redundant documents through knowledge extraction. Then, knowledge template construction, knowledge processing and generation, and knowledge review are used to standardize knowledge and finally complete the classification and management of knowledge. The specific processing process is as follows: Figure 3As shown. Through the analysis and investigation of lithium battery winding process and equipment, the fault knowledge system contains 8 key fields: fault code, equipment, components, fault phenomenon, cause, confirmation method, treatment measures, and material spare parts. In addition, the Sentence-Transformer pre-trained language model is used to convert the fault phenomenon field into a vector for subsequent semantic retrieval. Finally, ElasticSearch (ES for short) is used to store fault knowledge, including 8 text type fields and 1 vector type field.

[0107] In one embodiment, the method further comprises:

[0108] Receive the alarm information sent by the lithium battery device; parse the alarm information and extract the alarm content carried in the alarm information; match the fault handling measures in the preset fault knowledge base based on the alarm content to obtain the fault handling measures data.

[0109] When an abnormal situation occurs in a lithium battery device (such as a winding device) and meets the alarm conditions, the system will immediately trigger the alarm mechanism and send an alarm message. After receiving the alarm message, the server extracts the alarm content from the alarm message, matches the alarm content with the existing alarms and solutions in the knowledge base, and finally recommends one or more solutions.

[0110] Specifically, if Figure 4 As shown. The focus here is on matching prescriptions (treatment measures) for fault alarms. The main idea is to complete prescription recommendations based on the text similarity between the alarm information and the text recorded in the knowledge base. First, the alarm information sent by the device is preprocessed. The alarm information includes two parts: the alarm code (for example, "E00014") and the alarm content (for example, "Finishing glue YV904 rubber roller vacuum stretch alarm [X901.22]"). The alarm code is meaningless for this processing, so the alarm content is preprocessed. By deleting meaningless strings, stop words, etc., the key alarm content is extracted to improve the accuracy of subsequent similarity calculations. Secondly, BM25 similarity, edit distance similarity, and Jaccard similarity are selected as the main algorithms for text similarity calculation.

[0111] In one embodiment, fault handling measures are matched in a preset fault knowledge base based on the alarm content, and the obtained fault handling measures data include:

[0112] According to the alarm content, a rough sorting based on BM25 similarity is performed in the preset fault knowledge base to obtain the initial similar fault description results; the initial similar fault description results are secondarily sorted using edit distance similarity and Jaccard similarity, and the fault description results that meet the preset edit distance similarity threshold and the preset Jaccard similarity threshold are searched in the initial similar fault description results to obtain the target fault description results; the fault handling measures data corresponding to the target fault description results are searched in the preset fault knowledge base.

[0113] In this embodiment, the records in the knowledge base are sorted by combining the three similarity characteristics. A rough sort is performed according to the BM25 similarity, and a preset (for example, 10) results are screened out, and then a fine sort is performed using the edit distance similarity and Jaccard similarity. Finally, the thresholds of the edit distance similarity and the Jaccard similarity are set respectively, and a match is considered successful when the threshold conditions are met.

[0114] like Figure 5 As shown, in one embodiment, S300 includes:

[0115] S320: Calculate the BM25 similarity and cosine similarity of each candidate question in the candidate question set and the user question data.

[0116] In the candidate question set, BM25 similarity and cosine similarity are calculated for each candidate question and the user question data.

[0117] S340: Arrange the candidate questions in descending order of similarity based on the BM25 similarity calculation result, and adjust the descending order queue based on the cosine similarity calculation to obtain an initial candidate question sorting queue.

[0118] Here, candidate questions are sorted mainly using BM25 similarity and cosine similarity. First, they are sorted in descending order according to BM2 similarity, and then the order is fine-tuned according to cosine similarity.

[0119] In one embodiment, each candidate question is sorted in descending order of similarity based on the BM25 similarity calculation result, and the queue in descending order is adjusted based on the cosine similarity calculation, so that the initial candidate question sorting queue includes:

[0120] Step 1: Get the minimum threshold and maximum recall number K of BM25 similarity, and the minimum threshold and maximum threshold of cosine similarity.

[0121] Get the minimum threshold (e.g., 0.3) and maximum recall K (e.g., top 10) for BM25 similarity. Get the minimum threshold (e.g., 0.5) and maximum threshold (e.g., 0.8, for quick adjustments in special cases) for cosine similarity.

[0122] Step 2: Based on the BM25 similarity calculation results, recall K candidate questions whose BM25 similarity is greater than the minimum threshold of BM25 similarity from each candidate question.

[0123] Recall K candidate questions from the candidate question set whose BM25 similarity is greater than the minimum threshold. This step is to quickly narrow the candidate range and retain the most likely relevant questions.

[0124] Step 3: Based on the cosine similarity calculation result, recall all questions from each candidate question whose cosine similarity is greater than the minimum threshold of cosine similarity and are not repeated with the K candidate questions.

[0125] Recall all questions from the candidate question set whose cosine similarity is greater than the minimum threshold, while ensuring that these questions are not repeated with the K candidate questions initially screened. These additional recalled questions are sorted by cosine similarity and are ready to be inserted into the initial list of K candidate questions.

[0126] Step 4: Start sorting from position K+1. If the cosine similarity corresponding to the current candidate question is not less than the maximum threshold of the cosine similarity, the current candidate question is ranked first to obtain the initial candidate question sorting queue.

[0127] Starting from position K+1, traverse the additional recalled questions. If the cosine similarity of a question is not less than the maximum threshold, it means that it is highly similar to the user's question in semantics, and it can be sorted to the first place (or adjusted according to specific needs, but usually refers to a very prominent match). If it does not meet the maximum threshold condition, but still has a high cosine similarity, it can be inserted in order after the initial K candidate questions and sorted normally.

[0128] Specifically, in the above embodiment, first, four thresholds are set: the minimum threshold Min of BM25 similarity bm25 And the maximum recall number K, the minimum threshold of cosine similarity Min cos and the maximum threshold Max cos Then, taking BM25 similarity as the benchmark, recall is greater than Min bm25 At the same time, the recall cosine similarity is greater than Min cos , and all questions that are not repeated with the previous K questions are sorted starting from position K+1. cos , indicating that the semantics of the user's question and the candidate question are almost identical, so the candidate question is ranked first. Finally, the candidate question ranking based on similarity is implemented.

[0129] In one embodiment, based on the multiple similarity calculation results, screening the target candidate questions includes:

[0130] Step 1: Based on the multiple similarity calculation results, obtain the first candidate question.

[0131] Based on the sorting of multiple similarity calculation results (such as BM25, cosine similarity, etc.), the first candidate question is obtained.

[0132] Step 2: Calculate the cosine similarity between the first candidate question and the user's question data.

[0133] Calculate the cosine similarity between the first candidate question and the user's question data.

[0134] Step 3: If the cosine similarity is not greater than the preset cosine similarity threshold, the Jaccard similarity between the first candidate question and the user question data is calculated.

[0135] If the cosine similarity verification fails, the Jaccard similarity between the first candidate question and the user's question data is calculated. Jaccard similarity is usually used to measure the similarity between two sets, but in text processing, the text can be regarded as a word set or a bag of words model for calculation. If the cosine similarity is greater than the preset cosine similarity threshold (cosine similarity verification passed), the first candidate question is the target candidate question.

[0136] Step 4: If the Jaccard similarity is not greater than the preset Jaccard similarity threshold, the edit distance similarity between the first candidate question and the user question data is calculated.

[0137] If the Jaccard similarity is not greater than the preset Jaccard similarity threshold (indicating that the text content overlap is not high enough), the next step of verification is performed. If the Jaccard similarity is greater than the preset Jaccard similarity threshold, the first candidate question is the target candidate question. If the Jaccard similarity verification also fails, the edit distance similarity between the first candidate question and the user's question data is calculated. The edit distance (also known as the Levenshtein distance) measures the minimum number of single-character edits (insertion, deletion, or substitution) required to convert one string into another. Here, the edit distance similarity can be converted in some way (such as taking the inverse or normalization) to represent the similarity.

[0138] Step 5: If the edit distance similarity is not greater than the preset edit distance similarity threshold, M candidate questions with higher cosine similarity are selected from the multiple similarity calculation results to obtain the target candidate question.

[0139] If the edit distance similarity is not greater than the preset edit distance similarity threshold (indicating that the direct modification distance between the texts is far), then an alternative solution is selected. If all the above similarity verifications fail, then M candidate questions with higher cosine similarity are selected as alternatives from the remaining candidate questions of the multiple similarity calculation results. Here M is a preset number that can be adjusted according to actual conditions. These M candidate questions will be used as target candidate questions for subsequent answer queries and feedback.

[0140] Specifically, if Figure 6 As shown, in this embodiment, when outputting the answer, it is necessary to check the similarity between the first candidate question and the user's question again to avoid wrong or irrelevant answers.

[0141] First, the cosine similarity between the first candidate question and the user's question is calculated. If it is greater than the threshold, the answer corresponding to the candidate question is returned. Otherwise, the Jaccard similarity is calculated. If it is greater than the threshold, the corresponding answer is returned. Otherwise, the edit distance similarity is calculated. If it is greater than the threshold, the answer is also returned. Finally, if none of the above conditions are met, the correlation between the user's question and the candidate question is weak, and the M candidate questions with the highest similarity are returned for the user to choose. Since the importance of each similarity is different and the similarity calculation requires a certain amount of time, the four similarities are calculated in a serial order instead of calculating the four similarities for all questions at once. Cosine similarity is the most important because the sentence-transformer pre-trained language model can accurately describe the text similarity from a semantic level. Using cosine similarity, the candidate question with the highest semantic similarity can be quickly selected. The core of BM25 similarity, Jaccard similarity, and edit distance similarity is the degree of character overlap, which is complementary to cosine similarity. When the maximum value of cosine similarity is less than the threshold, the other three similarities are used for testing. If the conditions are met, the candidate question with the highest similarity can also be selected. If the conditions are not met, it may be that the user's question is vague or there is no highly matching question in the knowledge base. In order to avoid directly outputting wrong or irrelevant answers, a list of recommended questions with the highest similarity will be returned for the user's reference.

[0142] In a specific application example, the overall processing flow of the intelligent question-answering method for lithium battery equipment fault diagnosis in this application is as follows: Figure 7 shown.

[0143] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0144] Based on the same inventive concept, the embodiment of the present application also provides an intelligent question-and-answer device for lithium battery fault diagnosis for realizing the intelligent question-and-answer method for lithium battery fault diagnosis involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the intelligent question-and-answer device for lithium battery fault diagnosis provided below can refer to the limitations of the intelligent question-and-answer method for lithium battery fault diagnosis above, and will not be repeated here.

[0145] In one embodiment, Figure 8 As shown, an intelligent question-answering device for lithium battery equipment fault diagnosis is provided, comprising:

[0146] The entity recognition module 100 is used to obtain user question data, perform entity recognition on the user question data, and obtain entity query information;

[0147] The problem initial screening module 200 is used to screen candidate problems in a preset fault knowledge base based on entity query information to obtain a candidate problem set. The preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data;

[0148] A similarity calculation module 300 is used to perform multiple similarity calculations on each candidate question in the candidate question set and the user question data;

[0149] A question secondary screening module 400 is used to screen target candidate questions based on multiple similarity calculation results;

[0150] The query feedback module 500 is used to query the answers to the target candidate questions and provide feedback.

[0151] In one embodiment, the problem initial screening module 200 is used to obtain lithium battery equipment fault operation and maintenance data, which includes lithium battery equipment manuals, lithium battery equipment instructions and operation and maintenance record data; perform knowledge extraction on the lithium battery equipment fault operation and maintenance data to obtain knowledge extraction data; extract fault fields from the knowledge extraction data based on preset key fields, and the preset key fields include fault codes, equipment, components, fault phenomena, causes, confirmation methods, processing measures, and material spare parts; use a pre-trained language model to convert the fault field into a vector; store the fault field and the vector in association to generate a preset fault knowledge base.

[0152] In one of the embodiments, the intelligent question-and-answer device for lithium battery equipment fault diagnosis also includes an alarm module for receiving alarm information sent by the lithium battery equipment; parsing the alarm information and extracting the alarm content carried in the alarm information; matching fault handling measures in a preset fault knowledge base based on the alarm content to obtain fault handling measures data.

[0153] In one of the embodiments, the alarm module is also used to perform a coarse sorting based on BM25 similarity in a preset fault knowledge base according to the alarm content to obtain an initial similar fault description result; use the edit distance similarity and Jaccard similarity to perform a secondary fine sorting on the initial similar fault description results, and search for fault description results that meet the preset edit distance similarity threshold and the preset Jaccard similarity threshold in the initial similar fault description results to obtain a target fault description result; search for fault handling measures data corresponding to the target fault description result in the preset fault knowledge base.

[0154] In one of the embodiments, the similarity calculation module 300 is also used to perform BM25 similarity and cosine similarity calculations on each candidate question in the candidate question set and the user question data; based on the BM25 similarity calculation results, each candidate question is arranged in descending order of similarity, and based on the cosine similarity calculation, the descending queue is adjusted to obtain an initial candidate question sorting queue.

[0155] In one embodiment, the similarity calculation module 300 is also used to obtain the minimum threshold and maximum recall number K of BM25 similarity, as well as the minimum threshold and maximum threshold of cosine similarity; based on the BM25 similarity calculation result, recall K candidate questions whose BM25 similarity is greater than the minimum threshold of BM25 similarity from each candidate question; based on the cosine similarity calculation result, recall all questions whose cosine similarity is greater than the minimum threshold of cosine similarity and are not repeated with the K candidate questions from each candidate question; start sorting from position K+1, and if the cosine similarity corresponding to the current candidate question is not less than the maximum threshold of cosine similarity, the current candidate question is ranked first to obtain an initial candidate question sorting queue.

[0156] In one of the embodiments, the question secondary screening module 400 is also used to obtain a first candidate question based on multiple similarity calculation results; calculate the cosine similarity between the first candidate question and the user question data; if the cosine similarity is not greater than a preset cosine similarity threshold, calculate the Jaccard similarity between the first candidate question and the user question data; if the Jaccard similarity is not greater than the preset Jaccard similarity threshold, calculate the edit distance similarity between the first candidate question and the user question data; if the edit distance similarity is not greater than the preset edit distance similarity threshold, screen M candidate questions with higher cosine similarities from the multiple similarity calculation results to obtain the target candidate question.

[0157] Each module in the intelligent question-and-answer device for lithium battery equipment fault diagnosis can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0158] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store preset data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an intelligent question-and-answer method for fault diagnosis of a lithium battery device is implemented.

[0159] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0160] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the intelligent question-answering method for fault diagnosis of a lithium battery device is implemented.

[0161] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the intelligent question-and-answer method for fault diagnosis of a lithium battery device is implemented.

[0162] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the intelligent question-answering method for lithium battery device fault diagnosis.

[0163] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0164] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.

[0165] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0166] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An intelligent question-answering method for lithium battery equipment fault diagnosis, characterized in that: The method comprises: Acquire user question data, perform entity recognition on the user question data, and obtain entity query information; Based on the entity query information, candidate questions are screened in a preset fault knowledge base to obtain a candidate question set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data; Performing multiple similarity calculations on each candidate question in the candidate question set and the user question data; Based on the results of multiple similarity calculations, target candidate questions are screened; Query the answer to the target candidate question and provide feedback.

2. The method according to claim 1, characterized in that Before the step of screening candidate questions in a preset fault knowledge base based on the entity query information to obtain a candidate question set, the method further includes: Obtaining lithium battery equipment failure operation and maintenance data, wherein the lithium battery equipment failure operation and maintenance data includes a lithium battery equipment manual, a lithium battery equipment instruction manual, and operation and maintenance record data; Performing knowledge extraction on the lithium battery equipment fault operation and maintenance data to obtain knowledge extraction data; Extracting fault fields from the knowledge extraction data based on preset key fields, wherein the preset key fields include fault code, equipment, component, fault phenomenon, cause, confirmation method, treatment measures, and material spare parts; Using a pre-trained language model to convert the fault field into a vector; The fault field and the vector are associated and stored to generate a preset fault knowledge base.

3. The method according to claim 1, characterized in that The method further comprises: Receive alarm information sent by lithium battery equipment; Parsing the alarm information and extracting the alarm content carried in the alarm information; Based on the alarm content, fault handling measures are matched in the preset fault knowledge base to obtain fault handling measure data.

4. The method according to claim 3, characterized in that The matching of fault handling measures in the preset fault knowledge base based on the alarm content to obtain fault handling measures data includes: According to the alarm content, a rough sorting based on BM25 similarity is performed in the preset fault knowledge base to obtain an initial similar fault description result; The initial similar fault description results are sorted for a second time using edit distance similarity and Jaccard similarity, and a fault description result that satisfies a preset edit distance similarity threshold and a preset Jaccard similarity threshold is searched in the initial similar fault description results to obtain a target fault description result; The preset fault knowledge base is searched for fault handling measure data corresponding to the target fault description result.

5. The method according to claim 1, characterized in that The performing multiple similarity calculations on each candidate question in the candidate question set and the user question data comprises: Calculate the BM25 similarity and cosine similarity of each candidate question in the candidate question set and the user question data; Based on the BM25 similarity calculation result, each candidate question is arranged in descending order of similarity, and based on the cosine similarity calculation, the descending queue is adjusted to obtain the initial candidate question sorting queue.

6. The method according to claim 5, characterized in that The candidate questions are arranged in descending order of similarity based on the BM25 similarity calculation result, and the queues arranged in descending order are adjusted based on the cosine similarity calculation to obtain the initial candidate question sorting queue, which includes: Get the minimum threshold and maximum recall number K of BM25 similarity, as well as the minimum threshold and maximum threshold of cosine similarity; Based on the BM25 similarity calculation result, recall K candidate questions whose BM25 similarity is greater than a minimum threshold of BM25 similarity from the candidate questions; Based on the cosine similarity calculation result, recall from the candidate questions all questions whose cosine similarity is greater than a minimum threshold of the cosine similarity and are not repeated with the K candidate questions; Starting from position K+1, if the cosine similarity corresponding to the current candidate question is not less than the maximum threshold of the cosine similarity, the current candidate question is ranked first to obtain the initial candidate question ranking queue.

7. The method according to claim 1, characterized in that The screening of target candidate questions based on the multiple similarity calculation results includes: Based on the multiple similarity calculation results, the first candidate question is obtained; Calculate the cosine similarity between the first candidate question and the user question data; If the cosine similarity is not greater than a preset cosine similarity threshold, calculating the Jaccard similarity between the first candidate question and the user question data; If the Jaccard similarity is not greater than a preset Jaccard similarity threshold, calculating the edit distance similarity between the first candidate question and the user question data; If the edit distance similarity is not greater than a preset edit distance similarity threshold, M candidate questions with relatively high cosine similarities are screened from the multiple similarity calculation results to obtain a target candidate question.

8. An intelligent question-answering device for lithium battery equipment fault diagnosis, characterized in that: The device comprises: An entity recognition module is used to obtain user question data, perform entity recognition on the user question data, and obtain entity query information; A problem initial screening module, used to screen candidate problems in a preset fault knowledge base based on the entity query information to obtain a candidate problem set, wherein the preset fault knowledge base is constructed based on lithium battery equipment fault operation and maintenance data; A similarity calculation module, used to perform multiple similarity calculations on each candidate question in the candidate question set and the user question data; The question secondary screening module is used to screen target candidate questions based on multiple similarity calculation results; The query feedback module is used to query the answers to the target candidate questions and provide feedback.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images