Intelligent analysis method and device for power failure

Abstract

The invention discloses an intelligent power failure analysis method and device, and the method comprises the steps: carrying out the deduplication of failure text data, and obtaining a failure text data list which comprises a failure cause and a processing mode; the method comprises the following steps: segmenting fault text data according to text lengths of fault causes, generating a classification framework by combining a first segment of data with a large model, classifying new fault causes into an existing fault category or generating a new fault category according to text similarity between each new fault cause in subsequent segments of data and the existing fault causes, and classifying the new fault causes into the existing fault category. Obtaining a classification result data set; according to the fault occurrence frequency of each fault category in the classification result data set, performing high-frequency screening on fault causes and processing modes thereof; and performing fault category name merging on each fault category in the screened classification result data set to generate a fault maintenance report. The problems that a classification framework cannot realize dynamic adjustment, the classification precision is low and the classification efficiency is low are effectively solved, and the accuracy and efficiency of fault analysis are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power failure analysis, in particular to a power failure intelligent analysis method and device. BACKGROUND

[0002] In the field of power equipment failure analysis and maintenance management, the existing technology mainly focuses on the classification, statistics and report generation of failure records, and has experienced a development process from manual dominance to semi-automation and automation. Specifically, the development of technology can include the following stages: The traditional manual classification and statistics method relies on the experience and judgment of professionals, and classifies and counts failure records manually, which is low in efficiency, strong in subjectivity, and easy to lead to non-uniform classification standards and statistical errors.

[0003] The automatic classification method based on pre-defined rules classifies failure records automatically through rules pre-set by humans, such as the correspondence between fault codes and fault types. Although it improves the classification efficiency, the classification framework is rigid and lacks adaptability.

[0004] The classification method based on traditional machine learning uses machine learning algorithms such as decision trees and support vector machines to extract features and train models for failure records to achieve automatic fault classification. To some extent, it reduces the dependence on manual rules, but the data labeling cost of machine learning model training is high, the generalization ability is limited, which leads to a decrease in classification accuracy, and the inability to handle multi-value fields leads to information loss.

[0005] The static classification method based on early large models uses the text understanding ability of large models to classify all failure records at once without the need for manual feature extraction, and can understand complex semantics. However, when the data volume is too large, the attention allocation for long text is imbalanced, leading to classification bias, and the classification framework cannot be dynamically adjusted. Secondly, the current maintenance report generation technology still relies on manual analysis of core content, which is low in report generation efficiency and prone to content bias. SUMMARY

[0006] The present application provides a power failure intelligent analysis method and device, which can solve the technical problems of existing large model failure analysis technology, such as the inability of the classification framework to dynamically adjust, low classification accuracy, and low classification efficiency.

[0007] To achieve the above-mentioned purpose, in a first aspect, the present application provides a power failure intelligent analysis method, which comprises: De-duplicating the failure text data to obtain a failure text data list, the failure text data list including failure causes and treatment methods.

[0008] The fault text data is segmented according to the text length of the fault cause. The first segment of data is combined with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments, the new fault causes are either assigned to existing fault categories or new fault categories are generated, resulting in a classification result dataset.

[0009] Based on the frequency of failures in each failure category in the classification results dataset, high-frequency screening is performed to identify the causes of failures and their handling methods.

[0010] The fault category names in the filtered classification result dataset are merged to generate a fault repair report.

[0011] Furthermore, in one embodiment, the cause of the fault and the handling method are obtained by semantic extraction of the fault text data that has not been deduplicated using a large model; If the number of fault causes and handling methods in the fault text data list is inconsistent, the number will be aligned through the list completion mechanism.

[0012] Furthermore, in one embodiment, the deduplication of the faulty text data includes: The character similarity of the fault cause and the character similarity of the processing method are calculated using the string edit distance similarity formula, and the semantic similarity of the fault cause and the semantic similarity of the processing method are calculated using the cosine similarity formula.

[0013] The average of the character similarity and semantic similarity of the fault cause is used to obtain the text similarity of the fault cause, and the average of the character similarity and semantic similarity of the processing method is used to obtain the text similarity of the processing method.

[0014] If the text similarity of the fault cause of two faults is greater than or equal to the preset fault cause text similarity threshold, the text similarity of the processing method is greater than or equal to the preset processing method text similarity threshold, and the fault keywords belong to the same physical system, retain the fault cause name and processing method name with the shortest text length, and merge the two faults and their processing methods.

[0015] Furthermore, in one embodiment, segmenting the fault text data according to the text length of the fault cause includes: Calculate the average text length of all fault causes in the fault text data list.

[0016] If the average text length is less than or equal to the first length threshold, the segmentation threshold is the base threshold; if the average text length is greater than the first length threshold and less than or equal to the second length threshold, the segmentation threshold is 2 / 3 of the base threshold; if the average text length is greater than the second length threshold, the segmentation threshold is 1 / 2 of the base threshold.

[0017] The faulty text data is segmented based on the segmentation threshold.

[0018] Furthermore, in one embodiment, based on the text similarity between each new fault cause and an existing fault cause in subsequent data segments, the new fault cause is either categorized into an existing fault category or a new fault category is generated, including: If the text similarity between a new fault cause and at least one existing fault cause in an existing fault category is greater than or equal to a preset text similarity threshold, the new fault cause is assigned to the fault category containing the fault cause with the highest similarity.

[0019] If the text similarity between the new fault cause and all existing fault causes is lower than the preset text similarity threshold, and none of them belong to the preset fuzzy similarity matching range, the new fault cause will be used to generate a new fault category based on the professional terminology database.

[0020] Furthermore, in one embodiment, after obtaining the classification result dataset, the method further includes: Calculate the semantic similarity of the names of each newly added fault category. If the semantic similarity of the names of two newly added fault categories is greater than or equal to the preset semantic similarity threshold, retain the name of the earliest generated newly added fault category and merge the fault text data in the two newly added fault categories.

[0021] Furthermore, in one embodiment, the step of performing high-frequency screening of fault causes and their handling methods based on the frequency of fault occurrences for each fault category in the classification result dataset includes: The system counts the number of times each fault occurs for each fault cause in each single fault category, sorts the fault causes according to the number of occurrences, and retains a preset number of high-frequency fault causes and their handling methods in each single fault category.

[0022] Furthermore, in one embodiment, the step of merging the fault category names in the filtered classification result dataset includes: Based on the similarity of the names of each fault category in the filtered classification result dataset, fault keyword constraints, name text length, and fault type coverage, fault category names are merged, and the merged fault text data is arranged in order of the number of faults occurring in each individual fault category.

[0023] Furthermore, in one embodiment, the step of merging fault category names based on the name text similarity, fault keyword constraints, name text length, and fault type coverage of each fault category in the filtered classification result dataset includes: If the name text similarity of two fault categories is greater than or equal to the preset name text similarity threshold and the fault keywords belong to the same physical system, the fault text data in the two fault categories will be merged, and the fault category name with the shortest name text length and the highest fault type coverage will be retained.

[0024] The fault coverage is the ratio of the number of fault cause types in a fault category to the total number of fault cause types in the merged fault categories.

[0025] Secondly, this application provides an intelligent power fault analysis device, the device comprising: The deduplication module is used to deduplicatize faulty text data to obtain a list of faulty text data, which includes the cause of the fault and the handling method.

[0026] The classification module is used to segment the fault text data according to the text length of the fault cause. It combines the first segment of data with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments of data, the new fault cause is classified into the existing fault category or a new fault category is generated, resulting in a classification result dataset.

[0027] The filtering module is used to perform high-frequency filtering of fault causes and their handling methods based on the number of fault occurrences for each fault category in the classification result dataset.

[0028] The report generation module is used to merge the fault category names in the filtered classification result dataset and generate a fault repair report.

[0029] The beneficial effects of the technical solutions provided in this application include: By deduplicating fault text data, data redundancy is effectively reduced, improving data processing efficiency. Furthermore, the fault text data is segmented based on the text length of the fault cause, and a dynamic classification framework is generated using a large model. New fault causes are assigned to existing fault categories or new fault categories are generated through text similarity calculation. This allows for dynamic adjustment of the classification structure based on new data, improving classification accuracy. Simultaneously, high-frequency filtering of fault causes and their handling methods is performed based on the frequency of fault occurrence, ensuring the practicality and relevance of fault repair reports. Finally, the fault category names in the filtered classification result dataset are merged, optimizing the fault category structure and further improving classification efficiency. Therefore, the technical solution of this application effectively solves the technical problems of existing large-model fault analysis techniques, such as the inability to dynamically adjust the classification framework, low classification accuracy, and low classification efficiency, significantly improving the accuracy and efficiency of fault analysis. Attached Figure Description

[0030] Figure 1This is a flowchart of the intelligent power fault analysis method according to an embodiment of this application.

[0031] Figure 2 This is a block diagram of the intelligent power fault analysis device according to an embodiment of this application.

[0032] Figure 3 A complete flowchart for generating the fault repair report for this application is provided. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0034] First, some of the technical terms used in this application will be explained to help those skilled in the art understand this application.

[0035] TF-IDF stands for Term Frequency-Inverse Document Frequency. Term frequency represents the frequency of a word in a text, while inverse document frequency measures the importance of a word. The process of generating a TF-IDF vector is as follows: Calculate word frequencies using formulas (1) and (2) respectively. and inverse document frequency : (1), (2).

[0036] Calculate the TF-IDF weights using formula (3): (3).

[0037] For text T Generate vectors V Each element （T） TF-IDF Representing text T The TF-IDF weight of each word in the text.

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0039] In one aspect, embodiments of this application provide a method for intelligent analysis of power faults.

[0040] In one embodiment, see Figure 1 As shown, the above-mentioned intelligent power fault analysis method includes: S1. Remove duplicates from the fault text data to obtain a fault text data list, which includes the fault cause and the handling method.

[0041] S2. Segment the fault text data according to the text length of the fault cause. Combine the first segment of data with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments of data, classify the new fault causes into the existing fault categories or generate new fault categories to obtain the classification result dataset.

[0042] S3. Based on the number of faults occurring in each fault category in the classification result dataset, perform high-frequency screening of fault causes and their handling methods.

[0043] S4. Merge the fault category names in the filtered classification result dataset to generate a fault repair report.

[0044] In this embodiment, by deduplicating the fault text data, a list of fault text data containing fault causes and handling methods is generated, reducing the interference of duplicate data on the analysis and improving the efficiency of subsequent processing. The fault text data is segmented according to the text length of the fault causes, and a classification framework is generated using the first segment data combined with a large model. This segmentation method effectively addresses the problem of uneven attention distribution that may occur when processing long text data, ensuring that each segment receives sufficient and accurate analysis. The dynamically generated classification framework can be continuously adjusted and optimized based on new data, improving the adaptability and flexibility of classification. By calculating the text similarity between new fault causes and existing fault causes, the new fault causes are either assigned to existing fault categories or new fault categories are generated. This process considers not only the surface character similarity of the text but also semantic similarity for comprehensive judgment, thereby enabling more accurate identification and classification of fault causes. Based on the frequency of fault occurrence for each fault category in the classification result dataset, the causes of faults and their handling methods are filtered for high frequency. Then, the fault category names in the filtered classification result dataset are merged to finally generate a fault repair report. High frequency filtering can highlight the frequently occurring fault causes and their handling methods, helping maintenance personnel to quickly locate and solve major problems. Merging fault category names further optimizes the structure and content of the report, making it clearer, easier to understand and use, thereby improving the quality and practicality of the fault repair report.

[0045] Furthermore, in one embodiment, the fault cause and handling method in step S1 above are obtained by semantic extraction of the undeduplicated fault text data using a large model. Specifically: The database connection module establishes interaction with the fault text database, and performs structured queries based on the target fault code, time range, or fault name to extract fault text data from the fault text database, including fields such as loss record ID, downtime start time, manufacturer, model series, standard library fault name, fault code, loss description (including fault cause), and loss handling status (including handling method).

[0046] The loss description and loss handling information in the fault text data are combined into a text that serves as the semantic task input for the large model, with additional explanations of technical terms and output format constraints. The large model extracts the fault causes and handling methods from the combined text and forms a semantically structured output in the format of "fault cause + handling method". The output results are checked for field quantity matching, null values ​​and illegal characters. If the number of fields for fault cause and handling method in the output is inconsistent, or if there are null values ​​or illegal characters in the output results, a retry or manual review is triggered.

[0047] Since the fault causes or handling methods output by the large model may contain multiple values, the output results are further split into multi-value fields. Strings are separated by commas, and leading and trailing spaces are removed from each string element to form independent lists of fault causes and handling methods. If the number of fault causes and handling methods after splitting is inconsistent, a list completion mechanism is used to align the numbers. Specifically: If the difference between the number of fault causes and the number of handling methods is 1, it will be automatically corrected through the short list completion mechanism. If the fault cause list is longer, the last handling method will be copied to the same length. If the difference between the number of fault causes and the number of handling methods is greater than 1, the manual alignment workflow will be triggered, and the operation and maintenance personnel will be prompted to supplement the missing data through a pop-up window. The completion mechanism can ensure a 1:1 causal relationship between the "fault cause" list and the "handling method" list, and avoid analysis bias caused by misalignment of multi-value fields.

[0048] After alignment, TF-IDF vectors are used to encode the "fault cause-handling method" pairs. The cosine similarity algorithm is used to verify the semantic correlation of each "cause-handling" pair. When the similarity is less than the preset similarity threshold (i.e., the semantic correlation between the two is extremely weak, such as sensor fault corresponding to gearbox replacement), it is marked as semantic mismatch and pushed to manual confirmation to ensure the rationality of the causal relationship.

[0049] Terminology standardization is performed on the fault text data after splitting multi-valued fields, unifying the fault causes and handling methods into standardized terms to eliminate semantic noise caused by synonyms, heterogeneity, or human input differences. Terminology standardization is achieved through a mapping table and similarity matching rules, ensuring consistency of expression across devices and personnel.

[0050] In the standardized fault text data, fields with no valid content are labeled "No Record," and abnormally long fields are identified using the Z-score algorithm, as shown in the following formula: (4), in, Z This represents outliers, field length, mean field length of the same batch of data, and standard deviation. In this embodiment, if... If it is found to be abnormal, a manual verification process will be triggered.

[0051] The information density of terminology-standardized fault text data is verified using text entropy values, as shown in the following formula: (5), in, Represents text entropy. Indicates the first i The probability of each character appearing. m This represents the total number of distinct characters in the text. In this embodiment, if... If the percentage of repeated characters exceeds 80%, the data is considered to have low information density and will be subject to manual review.

[0052] Furthermore, in one embodiment, the deduplication of faulty text data in step S1 above specifically involves the following steps: For fault causes with completely duplicate text, perform initial deduplication, accumulate the number of fault occurrences, retain the processing method with the shortest text length, and generate a fault text data list containing the fault cause, processing method, and number of occurrences, as shown in the example below:

[0053] The character similarity of the fault cause and the character similarity of the processing method are calculated using the string edit distance similarity formula, and the semantic similarity of the fault cause and the semantic similarity of the processing method are calculated using the cosine similarity formula.

[0054] String edit distance similarity measures the degree of direct matching at the character level, and the calculation formula is as follows: (6), in, This represents the similarity of the edit distance between strings. and Represents text, indicating that Convert to The minimum number of insert, delete, and replace operations required.

[0055] Cosine similarity measures the degree of semantic matching, and is calculated as follows: Use jieba (a Python library for Chinese text segmentation) to segment Chinese text, such as... T 1 = “IGBT overheating” is segmented into [IGBT, overheating], T 2 = “IGBT module overheating” is segmented into [IGBT, module, overheating].

[0056] The formula for cosine similarity is as follows: (7), in, Represents cosine similarity. Representing text TF-IDF vector, Representing text TF-IDF vector, This indicates the total dimension of the vocabulary. express The i One portion, express The i Each component.

[0057] The average of the character similarity and semantic similarity of the fault cause is used to obtain the text similarity of the fault cause, and the average of the character similarity and semantic similarity of the processing method is used to obtain the text similarity of the processing method.

[0058] Text similarity The formula is as follows: (8).

[0059] If the text similarity of the fault cause of two faults is greater than or equal to the preset fault cause text similarity threshold, the text similarity of the processing method is greater than or equal to the preset processing method text similarity threshold, and the fault keywords belong to the same physical system, retain the fault cause name and processing method name with the shortest text length, and merge the two faults and their processing methods.

[0060] The constraint that fault keywords belong to the same physical system is to force faults of the same cause to belong to the same physical system. The fault keywords in the fault cause are matched with a pre-built keyword library to obtain a unique physical system. The physical system to which two identical fault keywords belong must be consistent.

[0061] Furthermore, in one embodiment, in step S2 above, the fault text data is segmented according to the text length of the fault cause. The specific steps are as follows: The formula for calculating the average text length of all fault causes in the fault text data list is as follows: (9), in, This represents the average text length. This indicates the total number of causes of the failure.

[0062] If the average text length is less than or equal to the first length threshold, the segmentation threshold is the base threshold; if the average text length is greater than the first length threshold and less than or equal to the second length threshold, the segmentation threshold is 2 / 3 of the base threshold; if the average text length is greater than the second length threshold, the segmentation threshold is 1 / 2 of the base threshold.

[0063] In this embodiment, the base threshold is 30 texts per segment. The segment count is calculated based on the base threshold and adjusted according to the average text length of the cause of the fault. The specific formula is as follows: (10) And satisfy This is to ensure that the amount of data in each segment is appropriate.

[0064] The fault text data is segmented according to the segmentation threshold, and then segmented according to the segmentation threshold. K Split the full fault data into M To avoid overload from a single input, each segment of data consists of continuous fault data, and the starting index of each segment is recorded. start i With Section Chief M i ， The index of the first data segment is i=0 The index of all segment data is 0~K-1 .

[0065] (11), (12).

[0066] Furthermore, in one embodiment, in step S2 above, the first data segment is combined with the large model to generate a classification framework.

[0067] The classification criterion is "core component + fault manifestation" (e.g., "pitch control unit + electrical component fault"). Each category contains multiple "fault causes + handling methods". Each "fault cause + handling method" can only be classified into one category and cannot belong to multiple categories simultaneously. The category output format is {"fault category":["fault 1","fault 2",...]}, resulting in the initial category set G0={g1, g2, ..., g p (In this embodiment, p≤15), and perform category elimination, retaining categories with a category coverage of ≥8% (category coverage = total number of failures in a single category / total number of failures in all categories), and determine the classification framework G. int=G0' (G0' is the filtered valid category).

[0068] Furthermore, in the embodiment, in step S2 above, based on the text similarity between each new fault cause and an existing fault cause in subsequent data segments, the new fault cause is either categorized into an existing fault category or a new fault category is generated. The specific steps are as follows: For the i ( i≥1 For the data segment from the second to the last paragraph, perform the following operations based on the above classification framework: Index range by segment ( start i end i (end) i =min((i+1)×K,N) extracts data and organizes it into a structured text format associated with the local index, i.e., k. Cause of failure: T local1global[k] Processing method: M local1global[k] .

[0069] If the text similarity between a new fault cause and at least one existing fault cause in an existing fault category is greater than or equal to a preset text similarity threshold, the new fault cause is assigned to the fault category containing the fault cause with the highest similarity. In this embodiment, the preset text similarity threshold is 0.7.

[0070] If the text similarity between the new fault cause and all existing fault causes is lower than a preset text similarity threshold, and none of them belong to a preset fuzzy similarity matching interval, the new fault cause will be used to generate a new fault category based on a terminology database. In this embodiment, the preset fuzzy similarity matching interval is [0.5, 0.7]. If the text similarity between the new fault cause and all existing fault causes is lower than the preset text similarity threshold, and they belong to a preset fuzzy similarity matching interval, the new fault cause will be automatically marked as "requires review" and pushed to professionals for confirmation of category attribution.

[0071] When a new fault category needs to be added, it is first semantically matched with the power industry terminology database (including professional terms such as equipment components, fault types, and handling measures). The similarity between the new fault category name and the term in the database must be greater than or equal to a preset similarity. The word vector similarity is calculated using the TF-IDF vectorization method. In this embodiment, the preset similarity value is 0.6. Otherwise, it is judged as a non-professional term, requiring manual adjustment of the category description. Then, the Longest Common Subsequence (LCS) algorithm is used to calculate the similarity between the new fault category and all existing fault categories. If the similarity with any existing category is greater than or equal to a preset category similarity value (in this embodiment, the preset category similarity value is 0.7), indicating high semantic overlap (e.g., "slip ring contamination" and "slip ring oil accumulation"), a category merging suggestion or manual decision is triggered to ensure the mutual exclusivity of professional terms in the classification system.

[0072] The data in subsequent segments are classified according to the above classification method to generate the first... i Segment mapping relationship map i ={g:[k 1 , k 2 ,...]} ,Will map i The local index list corresponding to each category is calculated using the formula. global_idx=start i +k Convert to a global index, and for each category mapped in the current segment, calculate the number of failures for newly added categories. The calculation formula is as follows: (13) in, This indicates the number of failures that occurred in the newly added category. Indicates the current paragraph order The number of times each fault occurred.

[0073] After all segmented processing is completed, hash verification and distribution statistics are used to check whether the classification results of the same fault cause are consistent in different segments. If there is an inconsistency (such as "sensor communication abnormality" being classified as "communication system failure" in the first segment and "sensor hardware failure" in the second segment), the segment matching log is traced back, and the optimal classification is recommended by the large model or corrected manually to ensure the consistency of the global classification logic.

[0074] Furthermore, in one embodiment, after obtaining the classification result dataset, step S2 above further includes the following step: Calculate the semantic similarity of the names of each newly added fault category. If the semantic similarity of the names of two newly added fault categories is greater than or equal to the preset semantic similarity threshold, retain the name of the earliest generated newly added fault category and merge the fault text data in the two newly added fault categories.

[0075] Furthermore, in one embodiment, in step S3 above, based on the frequency of fault occurrence for each fault category in the classification result dataset, high-frequency filtering is performed on the causes of faults and their handling methods. The specific steps are as follows: The system counts the number of times each fault occurs for each fault cause in each single fault category, sorts the fault causes according to the number of occurrences, and retains a preset number of high-frequency fault causes and their handling methods in each single fault category.

[0076] Furthermore, in one embodiment, in step S4 above, the fault category names in the filtered classification result dataset are merged. The specific steps are as follows: Based on the similarity of the names of each fault category in the filtered classification result dataset, fault keyword constraints, name text length, and fault type coverage, fault category names are merged, and the merged fault text data is arranged in order of the number of faults occurring in each individual fault category.

[0077] If the name text similarity of two fault categories is greater than or equal to a preset name text similarity threshold and the fault keywords belong to the same physical system, the fault text data in the two fault categories are merged, retaining the fault category name with the shortest name text length and the highest fault type coverage, and the index lists are merged. G is the set of category names to be merged.

[0078] The fault coverage is the ratio of the number of fault cause types in a fault category to the total number of fault cause types in the merged fault categories.

[0079] If the number of fault causes under a fault category is greater than the preset number of faults, in this embodiment, the preset number of faults is set to 20, and the fault causes are further broken down according to core components or fault types (such as "sensor abnormality" being broken down into "temperature sensor abnormality" or "position sensor abnormality"), to ensure that each fault category focuses on a single core fault feature.

[0080] Verify that all errors in the full index have been categorized into valid fault categories to ensure a low omission rate. O The omission rate is less than or equal to a preset omission rate threshold. In this embodiment, the preset omission rate threshold is 3%. If it is not met, manual review and supplementary classification are performed. O The calculation formula is: (14) (15) in, The total number of failures indicating the cause of the failure. This represents the set of indexes for unclassified fault causes.

[0081] The faults within each individual fault category are sorted in descending order of frequency, generating a categorized report that includes "Category Name, Core Sub-Fault, Handling Method, Total Frequency, and Fault Percentage." The fault percentage is the ratio of the number of faults occurring in a single fault category to the total number of faults occurring across all fault categories, calculated using the following formula: (16) (17) in, This represents the total number of failures for a single category. This indicates the percentage of failures, and the calculation result is rounded to two decimal places.

[0082] The calculation logic of the statistical results in the automatic verification report uses a reverse deduction method to deduce the merging process of faulty text data from the statistical results. If there are problems such as duplicate calculation, numerical overflow, or the percentage is not 100%, the statistical process will be automatically re-executed.

[0083] The system calls upon the historical maintenance results database to match the historical validity of the processing methods in the reports. If the repair success rate is less than the preset success rate threshold, or the average working hours exceed the industry benchmark preset multiple, it automatically recommends better historical solutions to ensure the feasibility and economy of maintenance suggestions. In this embodiment, the preset success rate threshold is set to 60%, and the preset multiple is set to 2.

[0084] Secondly, this application provides an embodiment of an intelligent power fault analysis device. See [link to related document]. Figure 2 As shown, the above-mentioned device includes a deduplication module, a classification module, a filtering module, and a report generation module, specifically: The deduplication module is used to deduplicatize faulty text data to obtain a list of faulty text data, which includes the cause of the fault and the handling method.

[0085] The classification module is used to segment the fault text data according to the text length of the fault cause. It combines the first segment of data with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments of data, the new fault cause is classified into the existing fault category or a new fault category is generated, resulting in a classification result dataset.

[0086] The filtering module is used to perform high-frequency filtering of fault causes and their handling methods based on the number of fault occurrences for each fault category in the classification result dataset.

[0087] The report generation module is used to merge the fault category names in the filtered classification result dataset and generate a fault repair report.

[0088] See Figure 3 As shown, a flowchart illustrating the entire process of fault repair report generation, from data input to output, is presented. It consists of four layers, with each layer's modules working together to achieve a closed-loop process of "data preprocessing - intelligent classification - report generation," as detailed below: Data Source Layer: Serving as the "data foundation," this layer includes a fault record database and a power industry terminology library. The fault record database stores raw fault data throughout the entire lifecycle of power equipment, covering fields such as fault time, fault description, handling measures, and downtime, providing "raw material" for analysis. The power industry terminology library collects professional terms and standardized definitions in the power field (e.g., pitch safety chain: emergency shutdown protection device), ensuring consistency in professional terminology throughout the process and avoiding ambiguity that could interfere with analysis results.

[0089] The data processing layer includes a data cleaning module and a format conversion module. These modules clean and convert the data to provide high-quality, standardized, and segmentable data for intelligent classification. The data cleaning module addresses the issues of "dirty, messy, and incomplete" data by performing operations such as null value marking, outlier detection (e.g., Z-score algorithm, triggering manual verification when |Z|>3), and deduplication, thereby improving data quality. The format conversion module unifies multi-source heterogeneous data into a standard format that can be parsed by large models, breaking down data format barriers.

[0090] Intelligent classification layer: Based on the "large model + self-growing classification" technology, it realizes the "dynamic construction and optimization" of the fault classification system, including an adaptive segmentation module, a classification framework generation module, a dynamic matching growth module, and a category merging optimization module.

[0091] Adaptive segmentation module: Dynamically adjusts the segmentation strategy based on real-time data volume and large model token efficiency (e.g., when the data volume surges, automatically optimizes the number of records in each segment to ensure token utilization ≥ 80%), improves processing efficiency, and avoids information loss or model performance degradation when processing long texts.

[0092] Classification framework generation module: Uses a large model to perform semantic analysis on the first segment of data to generate a fault classification framework (such as a fault category structure of "core components + fault manifestations" to ensure appropriate classification granularity).

[0093] Dynamic matching and growth module: For data segments 2 to N, the cosine similarity algorithm is used to match existing fault categories; if the matching degree is <0.7, a new category is added, and the professionalism of the fault category and its similarity with existing fault categories are verified by combining the "power industry terminology library", so as to realize the "self-growth" of the classification system.

[0094] Category merging optimization module: Calculates the similarity of faulty categories using the longest common subsequence algorithm; if the similarity is ≥0.7, merges semantically similar categories, retains high-frequency category names, and avoids category redundancy.

[0095] The report summary layer transforms the intelligent classification results into reports that directly guide maintenance. This includes a statistical indicator calculation module, a batch report generation module, and a full report integration module. The statistical indicator calculation module automatically calculates the frequency and percentage of each fault category, quantifying fault distribution patterns. The batch report generation module generates Markdown-formatted phased reports according to classification segments. Each segment covers a preset number of fault categories, including fault causes and handling methods. The full report integration module merges all segmented reports, supplementing them with classification logic explanations, high-frequency faults, and maintenance priority suggestions, ultimately forming a complete maintenance handover report that directly supports power equipment maintenance decisions.

[0096] The flowchart for generating fault repair reports clearly presents the flow logic of data from the bottom layer to the top layer: the data source layer provides "raw materials" → the data processing layer performs "preprocessing" → the intelligent classification layer realizes "core analysis and dynamic classification" → the report summary layer outputs "decision results". The entire process highlights the innovation of "large model empowerment + self-growing classification", solving the pain points of traditional fault classification such as "fixed and difficult to expand" and "limited analysis of long texts", and providing systematic support for intelligent analysis and maintenance optimization of power equipment faults.

[0097] This application addresses the issues of high noise levels and low reliability in direct analysis of raw data through a comprehensive data cleaning mechanism. The Z-score algorithm identifies abnormally long fields and filters out erroneous data. Multi-valued fields are split and aligned, and "short list completion" ensures a one-to-one correspondence between cause and treatment method, avoiding causal misalignment. Duplicates are removed and frequencies are merged based on "standard fault names" to ensure accurate statistical data and provide a reliable data foundation for subsequent classification.

[0098] The dynamic growth classification mechanism solves the problem of rigidity in traditional classification systems. By segmenting the entire dataset into small batches, it avoids the rigidity of a framework caused by one-time classification. New data is matched with existing categories using cosine similarity; when the matching degree falls below 50%, a new category is automatically added, ensuring the classification system can expand as fault types evolve. Simultaneously, redundant categories are quantified and merged to resolve the issue of "different names for the same cause."

[0099] By employing a segmentation and iterative processing strategy, the bottleneck of traditional large-scale model data processing capabilities is overcome. The segmentation threshold is dynamically set according to the model's token limit (default 50 records / segment, with the number of tokens in a single segment less than 80% of the model's limit), ensuring that each batch of data is fully understood by the model. A "first segment framework building + subsequent segment matching and growth" model is adopted: the first segment of data generates an initial classification, and subsequent segments are matched with the existing framework based on cosine similarity, ensuring classification consistency while avoiding excessively large data volumes processed at once. Global validation (comparing with all historical categories) is performed on each newly added category in each segment, ensuring that segmentation processing does not affect the coherence of the overall classification logic, thereby improving the model's ability to process long texts and batches of data.

[0100] The automated processing workflow solves the problems of excessive manual intervention, low efficiency, and strong subjectivity. The entire classification process, from data segmentation and similarity calculation to category merging, is algorithm-driven, eliminating the need for manual definition of classification rules. The statistical analysis automatically calculates the fault percentage, avoiding errors from manual calculations. The reporting process automatically generates structured tables in batches, reducing the report generation time for 1000 records from one week to 10 minutes, while ensuring consistent formatting and reproducible results, significantly improving the efficiency and objectivity of fault analysis.

[0101] By using quantitative indicators and process traceability, the problems of poor interpretability of analysis results and difficulty in tracing the classification basis were solved. Each classification step outputs a quantifiable matching degree, clearly defining the classification basis. During segmented processing, classification logs for each batch are retained, including the verification results of newly added categories. When merging similar categories, the similarity calculation process is output synchronously, ensuring that the entire process is traceable and interpretable, improving maintenance personnel's understanding and trust in the classification logic.

[0102] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0103] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0104] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0105] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0107] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for intelligent analysis of power faults, characterized in that, The method includes: The fault text data is deduplicated to obtain a fault text data list, which includes the fault cause and the handling method. The fault text data is segmented according to the text length of the fault cause. The first segment of data is combined with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments, the new fault causes are classified into existing fault categories or new fault categories are generated to obtain the classification result dataset. Based on the frequency of failures in each failure category in the classification result dataset, high-frequency screening is performed to identify the causes of failures and their handling methods. The fault category names in the filtered classification result dataset are merged to generate a fault repair report.

2. The intelligent power fault analysis method as described in claim 1, characterized in that, The causes and handling methods of the faults are obtained by semantic extraction of the fault text data that has not been deduplicated using a large model; If the number of fault causes and handling methods in the fault text data list is inconsistent, the number will be aligned through the list completion mechanism.

3. The intelligent power fault analysis method as described in claim 1, characterized in that, The process of deduplicating faulty text data includes: The character similarity of the cause of the fault and the character similarity of the processing method are calculated according to the string edit distance similarity formula, and the semantic similarity of the cause of the fault and the semantic similarity of the processing method are calculated according to the cosine similarity formula. The average of the character similarity and semantic similarity of the fault cause is used to obtain the text similarity of the fault cause; the average of the character similarity and semantic similarity of the processing method is used to obtain the text similarity of the processing method. If the text similarity of the fault cause of two faults is greater than or equal to the preset fault cause text similarity threshold, the text similarity of the processing method is greater than or equal to the preset processing method text similarity threshold, and the fault keywords belong to the same physical system, retain the fault cause name and processing method name with the shortest text length, and merge the two faults and their processing methods.

4. The intelligent power fault analysis method as described in claim 1, characterized in that, The segmentation of fault text data based on the text length of the fault cause includes: Calculate the average text length of all fault causes in the fault text data list; If the average text length is less than or equal to the first length threshold, the segmentation threshold is the base threshold; if the average text length is greater than the first length threshold and less than or equal to the second length threshold, the segmentation threshold is 2 / 3 of the base threshold; if the average text length is greater than the second length threshold, the segmentation threshold is 1 / 2 of the base threshold. The faulty text data is segmented based on the segmentation threshold.

5. The intelligent power fault analysis method as described in claim 1, characterized in that, Based on the text similarity between each new fault cause and existing fault causes in subsequent data segments, the new fault causes are either categorized into existing fault categories or new fault categories are generated, including: If the text similarity between a new fault cause and at least one existing fault cause in an existing fault category is greater than or equal to a preset text similarity threshold, the new fault cause is assigned to the fault category containing the fault cause with the highest similarity. If the text similarity between the new fault cause and all existing fault causes is lower than the preset text similarity threshold, and none of them belong to the preset fuzzy similarity matching range, the new fault cause will be used to generate a new fault category based on the professional terminology database.

6. The intelligent power fault analysis method as described in claim 5, characterized in that, After obtaining the classification result dataset, it also includes: Calculate the semantic similarity of the names of each newly added fault category. If the semantic similarity of the names of two newly added fault categories is greater than or equal to the preset semantic similarity threshold, retain the name of the earliest generated newly added fault category and merge the fault text data in the two newly added fault categories.

7. The intelligent power fault analysis method as described in claim 1, characterized in that, The process of high-frequency filtering of fault causes and their handling methods based on the frequency of fault occurrences for each fault category in the classification result dataset includes: The system counts the number of times each fault occurs for each fault cause in each single fault category, sorts the fault causes according to the number of occurrences, and retains a preset number of high-frequency fault causes and their handling methods in each single fault category.

8. The intelligent power fault analysis method as described in claim 1, characterized in that, The step of merging the fault category names in the filtered classification result dataset includes: Based on the similarity of the names of each fault category in the filtered classification result dataset, fault keyword constraints, name text length, and fault type coverage, fault category names are merged, and the merged fault text data is arranged in order of the number of faults occurring in each individual fault category.

9. The intelligent power fault analysis method as described in claim 8, characterized in that, The process of merging fault category names based on the similarity of name text, fault keyword constraints, name text length, and fault type coverage of each fault category in the filtered classification result dataset includes: If the name text similarity of two fault categories is greater than or equal to the preset name text similarity threshold and the fault keywords belong to the same physical system, the fault text data in the two fault categories will be merged, and the fault category name with the shortest name text length and the highest fault type coverage will be retained. The fault coverage is the ratio of the number of fault cause types in a fault category to the total number of fault cause types in the merged fault categories.

10. A power fault intelligent analysis device, characterized in that, The device includes: The deduplication module is used to deduplicat the fault text data to obtain a fault text data list, which includes the fault cause and the handling method; The classification module is used to segment the fault text data according to the text length of the fault cause. It combines the first segment of data with the large model to generate a classification framework. Based on the text similarity between each new fault cause and the existing fault causes in the subsequent segments of data, the new fault cause is classified into the existing fault category or a new fault category is generated to obtain the classification result dataset. The filtering module is used to perform high-frequency filtering of fault causes and their handling methods based on the number of fault occurrences in each fault category in the classification result dataset. The report generation module is used to merge the fault category names in the filtered classification result dataset and generate a fault repair report.

Images