Intelligent filling and reporting system for distribution network maintenance application

Abstract

The invention discloses a distribution network maintenance application intelligent filling system, which comprises an automatic identification maintenance application form filling requirement algorithm model, a receipt classification automatic matching template and a template automatic matching filling module, and is characterized in that the automatic identification maintenance application form filling requirement algorithm model comprises data collection and arrangement and image identification processing; the document classification automatic matching template comprises a template design and matching process, and the template automatic matching and filling module comprises a template automatic matching and filling method and an auditing and correcting method. According to the intelligent filling system for the distribution network maintenance application, the maintenance application filling quality is improved, the maintenance application circulation approval efficiency is improved, the power supply service quality is improved, and the power grid operation risk is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network maintenance, and in particular relates to an intelligent reporting system for distribution network maintenance applications. Background Art

[0002] In the maintenance of power distribution network, filling in maintenance application is a critical and tedious task. The traditional way of filling in maintenance application often relies on manual judgment and experience accumulation, which is not only inefficient, but also prone to errors and omissions due to human factors. With the increasing scale and complexity of power network, the traditional way of filling in has been difficult to meet the needs of modern power maintenance management; firstly, different types of maintenance work have different requirements for filling in the content, and the lack of standardized filling in templates has caused the filling in personnel to spend a lot of time to understand and organize relevant information, reducing work efficiency; at the same time, due to the differences in the professional skills of the filling in personnel, the accuracy and standardization of the filling in the content are also difficult to guarantee; secondly, when generating the standard description of the safety measures of power outage maintenance information, the traditional method often relies on manual writing, which not only increases the workload, but also easily leads to incomplete description of safety measures or loopholes due to human negligence, which brings safety hazards to the maintenance work; in order to solve the above problems and improve the efficiency and accuracy of the distribution network maintenance application filling in, it is particularly important to develop a distribution network maintenance application intelligent filling in auxiliary tool. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a distribution network maintenance application intelligent filling system that can improve the filling quality of maintenance applications, speed up the efficiency of maintenance application circulation and approval, improve the quality of power supply services and reduce the risk of power grid operation.

[0004] The technical solution of the present invention is: a distribution network maintenance application intelligent filling system, including: an automatic recognition maintenance application form filling requirement algorithm model, a document classification automatic matching template and a template automatic matching filling module, the automatic recognition maintenance application form filling requirement algorithm model includes: data collection and collation and image recognition processing, the data collection is to collect a large amount of maintenance form filling format data, real-time monitoring data and historical maintenance application form data, and classify and collect them, the image recognition processing includes: document digitization, image preprocessing, feature extraction, pattern matching, data preprocessing and expansion of training data set, the data preprocessing includes: text preprocessing and word segmentation processing, the word segmentation processing includes: calculating word frequency, calculating inverse document frequency and extracting keywords and classifying them;

[0005] The document classification automatic matching template includes: template design and matching process, the template design includes: template design and classification, template matching process, template optimization and template content, the template design and classification includes: analyzing requirements, designing templates and classified storage, the template matching process includes: text preprocessing, template feature extraction, template selection, the template selection is realized by keyword automatic matching template or classifier, the keyword automatic matching template includes: AC automaton, Trie tree and KMP algorithm, the matching process includes: text preprocessing, key feature extraction and template matching, the template matching specifically compares the extracted features with the templates in the template library to find the best matching template, and the template matching matching algorithm is similarity calculation or regular expression matching;

[0006] The template automatic matching and filling module includes: a template automatic matching and filling method and a review and correction method. The template automatic matching and filling method includes: template definition and standardization, automatic matching algorithm, intelligent filling and logic verification. The automatic matching algorithm is a logistic regression model. The expression of the logistic regression model is:

[0007] Y=σ(WX+b),

[0008]

[0009] z=WX+b,

[0010]

[0011] Among them, Y represents the output probability of the model, σ represents the Sigmoid function, W represents the weight vector, X represents the input feature vector, and b represents the bias term. Represents the classification label. If Y>0.5, the model predicts that the sample belongs to category 1. If Y≤0.5, the model predicts that the sample belongs to category 0.

[0012] The intelligent filling is to automatically extract relevant information from the data source using a recurrent neural network model and fill it into the corresponding fields of the maintenance order. The expression of the recurrent neural network model is:

[0013] h t =f(W xh ,X t +W hh ·h t-1 +b),

[0014] Among them, h t represents the hidden state at the current time t, f represents the nonlinear activation function, W xh represents the weight of the input to the hidden state, X t represents the input features at the current time t, Whh represents the weight from hidden state to hidden state, h t-1 represents the hidden state at the previous moment t-1, and b represents the bias term;

[0015] The template content is specifically as follows: all designed templates are stored in a template library, which is a database, file system or cloud storage service, and each template contains common fields and formats in the application form of this category, including: equipment name, fault description, maintenance time, application unit, and approval process;

[0016] The review and correction method includes: manual review, intelligent correction suggestions, feedback and iteration, and process optimization. The intelligent correction suggestions specifically provide correction suggestions for errors or omissions in the automatic filling process based on historical data, user feedback or machine learning models. The process optimization introduces automated testing or continuous integration to regularly evaluate and optimize the automated processing process.

[0017] Furthermore, the document digitization is to quickly convert the image into an electronic document through optical character recognition, the image preprocessing is specifically to perform denoising, binarization and tilt correction operations on the input image, the feature extraction is to extract the text features in the image through edge detection and morphological processing, the text features include stroke thickness and curvature, the pattern matching is to compare the extracted text features with the characters in the database through the Tesseract-OCR model, the data preprocessing is to further optimize the image quality through sharpening and contrast stretching, the expanded training data set is: to expand the training data set for the field of distribution network maintenance technology and Chinese and English, and the method of expanding the training data set includes: data collection, data enhancement, transfer learning and pre-training model.

[0018] Furthermore, the text preprocessing is specifically: preprocessing the text description in the distribution network maintenance application form, including removing punctuation marks, spaces, line breaks, and performing unified processing of text encoding, the word segmentation processing is to use the Jieba word segmentation algorithm and select the precise mode to segment the preprocessed text, and the expression for calculating word frequency is:

[0019]

[0020] Among them, TF(t, d) represents the frequency of word t in document d, and f t，d represents the number of times word t appears in document d, ∑ k f k，d Represents the total number of times all words appear in document d;

[0021] The expression for calculating the inverse document frequency is:

[0022]

[0023] Where IDF(t) represents the inverse document frequency of word t, N represents the total number of documents, and |{d∈D:t∈d}| represents the number of applications for word t.

[0024] The extraction and classification of keywords specifically includes: sorting the words according to the TF-IDF value of each word, selecting the words with the highest values ​​as the keywords of the application form, and outputting the extracted key information in a structured form for subsequent processing.

[0025] Furthermore, the analysis requirements are: analyzing various types, common fields, and format specifications of distribution network maintenance application forms to determine specific requirements for template design; the design template is: designing multiple templates based on the requirements analysis results, each template corresponding to a specific type of application form, the template contains fields that must be filled in the application form of this category, and sets the arrangement order and format of the fields; the classification storage is: classifying the designed templates according to the equipment type, maintenance nature, and application unit dimensions, and storing them in a template library, which is a database, file system, or cloud storage service; the template feature extraction is: extracting key features from the preprocessed text, including: keywords, key phrases, dates and times in a specific format; the AC automaton includes the following steps: constructing a prefix tree, adding mismatch pointers, and pattern matching; the Trie tree includes the following operation steps: inserting, deleting, and querying; the time complexity of the query operation is proportional to the length of the target string; the KMP algorithm includes The following steps include: constructing a prefix table, initialization, matching process and determining the sliding position. Specifically, the prefix table is constructed as follows: the i-th element of the prefix table represents the longest common prefix and suffix length of the substring consisting of the first i+1 characters of the pattern string. Specifically, the initialization is as follows: before starting the match, initialize two pointers i and j to point to the starting positions of the text string and the pattern string respectively. At the same time, create a prefix table to store the longest common prefix and suffix information of the pattern string. Specifically, the matching process is as follows: starting from the beginning of the text string, compare the text string and the pattern string character by character. When the characters do not match, determine the sliding position of the pattern string according to the prefix table. Specifically, the sliding position is determined as follows: assuming that the pattern string is "ABCABD" and the text string is "AAACAABDA", during the matching process, when the fifth character 'C' of the text string does not match the third character 'C' of the pattern string, according to the prefix table, the pattern string will slide forward 2 places so that the first character of the pattern string is aligned with the fifth character of the text string.

[0026] Furthermore, the template matching is to automatically classify the templates by a support vector machine, and the specific steps of the template matching are:

[0027] (1) Data preparation: First, prepare a dataset for training and testing. The dataset contains labeled samples, each of which has a set of features and a corresponding category label. Then, use the train_test_split function to divide the dataset into a training set and a test set, with a ratio of 70% training set and 30% test set.

[0028] (2) Feature extraction and selection: Extract appropriate features from the original data as input, perform feature scaling, standardization or normalization operations on the data, and use StandardScaler for feature scaling;

[0029] (3) Data preprocessing: Use filtering or smoothing to denoise the data and fill or delete missing values;

[0030] (4) Model building and training: According to the problem requirements, select a linear kernel function, a polynomial kernel function, or a Gaussian radial basis function kernel and parameter configuration corresponding to the problem requirements to build a support vector machine classification model, input the prepared training data into the libsvm library, and perform model training based on the selected kernel function and parameters;

[0031] (5) Model evaluation: Use the test data set to evaluate the trained model, calculate the classification accuracy, precision, recall rate, F1 value, draw the confusion matrix, and analyze the performance of the model in different categories;

[0032] (6) Model optimization: According to the evaluation results, adjust the parameter configuration of the support vector machine model, including the parameters of the kernel function and the regularization parameter, and use k-fold cross validation to evaluate the stability and generalization ability of the model;

[0033] (7) Prediction and classification: Input the features of new unknown samples into the model and classify them into different categories according to the decision boundary of the model.

[0034] Furthermore, the steps of using StandardScaler to perform feature scaling are:

[0035] (1) Import library: Import StandardScaler from Scikit-learn;

[0036] (2) Prepare feature datasets;

[0037] (3) Instantiate StandardScaler: Create a StandardScaler object;

[0038] (4) Fitting and transforming data: Use the fit_transform method to fit, transform, and standardize features of the training data, and use the transform method to transform the test set;

[0039] (5) View scaled data: Output scaled data to verify whether the features have been standardized;

[0040] The expression of the standardized feature is:

[0041]

[0042] Among them, μ represents the mean, σ represents the standard deviation, and x′ i represents the eigenvalue of the i-th sample after standardization, x i represents the characteristic value of the i-th sample, and N represents the total number of samples in the data set.

[0043] Furthermore, the steps of k-fold cross validation to evaluate the stability and generalization ability of the model are:

[0044] (1) Split the data set:

[0045]

[0046] Among them, D represents the entire data set, S i represents the i-th subset or fold, represents the union of all subsets, S j represents the jth subset or fold, and k represents the number of subsets or folds;

[0047] (2) The i-th training and testing: Training set: T i =D\S i , test set: S i ;

[0048] (3) Model evaluation: In the i-th time, train the model and calculate the performance index E i , the performance index E i The expression is:

[0049] E i =Performance(M,S i ),

[0050] Among them, M represents the training model, S i represents the i-th subset or fold;

[0051] (4) Calculate the average performance:

[0052]

[0053] Where k represents the number of subsets or folds, E i Indicates performance index, E mean Indicates the stability and generalization ability of the model on the dataset.

[0054] Furthermore, the template definition and standardization are as follows: defining the structure and necessary fields of the maintenance order template, including: equipment name, fault description, maintenance steps, collecting and organizing the data sources used to fill out the maintenance order, including: equipment information library, historical fault records, spare parts inventory, and cleaning and formatting the data sources; the logical verification includes setting automated verification rules and using machine learning models to identify potential errors or anomalies.

[0055] Furthermore, the automated verification rules include: required field check, data type and format check, logical relationship check, range check, and uniqueness check; the machine learning model is a text analysis model: an NLP model.

[0056] Beneficial effects of the present invention:

[0057] 1. Maintenance application template formulation: According to the different types and scenarios of distribution network maintenance, standardized maintenance application filling templates are designed and formulated. Each template contains necessary fields, format requirements, preset options and default values ​​to meet the needs of different maintenance types. The template supports dynamic adjustment, allowing system administrators to add, delete, modify and check operations according to actual needs to ensure the timeliness and accuracy of the template;

[0058] 2. Intelligent identification and reporting: Users select the corresponding maintenance type through the user interface, and the system automatically loads the corresponding reporting template. Users enter the basic information of the power outage equipment, such as equipment name, number, location, etc. The system uses intelligent identification technology to automatically retrieve and match the detailed information of the equipment from the database. The system intelligently identifies and associates all related primary and secondary equipment based on the primary equipment type entered by the user, providing basic data for the subsequent generation of power outage maintenance information security measures;

[0059] 3. Intelligent generation of safety measures: Establish a safety measures template library, covering the description of safety measures under various equipment types and maintenance scenarios. The templates in the template library can be updated and maintained as needed. When the user completes the input of the power outage equipment information, the system intelligently matches the corresponding template from the safety measures template library according to the associated primary and secondary equipment information. Using intelligent filling technology, key information such as power outage scope and equipment number is automatically filled into the safety measures template to generate a complete safety measures description;

[0060] 4. Data verification and validation: Before users submit maintenance applications, the system will conduct strict verification and validation of the submitted data, including data integrity, consistency, and logical rationality. For errors or abnormalities found, the system provides intelligent error prompts and suggestions to help users quickly correct them. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention.

[0062] Figure 1 It is a schematic diagram of a user operation interface of a maintenance report form of an intelligent reporting system for distribution network maintenance application of the present invention.

[0063] Figure 2 It is a schematic diagram of a user operation interface for selecting power outage equipment in an intelligent reporting system for distribution network maintenance application of the present invention.

[0064] Figure 3 It is a schematic diagram of a user operation interface generated by a specification description of an intelligent reporting system for distribution network maintenance application of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0066] like Figure 1-3 As shown, a distribution network maintenance application intelligent filling system includes: an algorithm model for automatically identifying maintenance application form filling requirements, a document classification automatic matching template and a template automatic matching filling module, the algorithm model for automatically identifying maintenance application form filling requirements includes: data collection and sorting and image recognition processing, the data collection is to collect a large amount of maintenance form filling format data, real-time monitoring data and historical maintenance application form data, and classify and collect them, the image recognition processing includes: document digitization, image preprocessing, feature extraction, pattern matching, data preprocessing and expansion of training data set, the data preprocessing includes: text preprocessing and word segmentation processing, the word segmentation processing includes: calculating word frequency, calculating inverse document frequency and extracting keywords and classifying them;

[0067] The document classification automatic matching template includes: template design and matching process, the template design includes: template design and classification, template matching process, template optimization and template content, the template design and classification includes: analyzing requirements, designing templates and classified storage, the template matching process includes: text preprocessing, template feature extraction, template selection, the template selection is realized by keyword automatic matching template or classifier, the keyword automatic matching template includes: AC automaton, Trie tree and KMP algorithm, the matching process includes: text preprocessing, key feature extraction and template matching, the template matching specifically compares the extracted features with the templates in the template library to find the best matching template, and the template matching matching algorithm is similarity calculation or regular expression matching;

[0068] The template automatic matching and filling module includes: a template automatic matching and filling method and a review and correction method. The template automatic matching and filling method includes: template definition and standardization, automatic matching algorithm, intelligent filling and logic verification. The automatic matching algorithm is a logistic regression model. The expression of the logistic regression model is:

[0069] Y=σ(Wx+b),

[0070]

[0071] z=WX+b,

[0072]

[0073] Among them, Y represents the output probability of the model, σ represents the Sigmoid function, W represents the weight vector, X represents the input feature vector, and b represents the bias term. Represents the classification label. If Y>0.5, the model predicts that the sample belongs to category 1. If Y≤0.5, the model predicts that the sample belongs to category 0.

[0074] The intelligent filling is to automatically extract relevant information from the data source using a recurrent neural network model and fill it into the corresponding fields of the maintenance order. The expression of the recurrent neural network model is:

[0075] h t =f(W xh ·X t +W hh ·h t-1 +b),

[0076] Among them, h t represents the hidden state at the current time t, f represents the nonlinear activation function, W xh represents the weight of the input to the hidden state, X t represents the input features at the current time t, Whh represents the weight from hidden state to hidden state, h t-1 represents the hidden state at the previous moment t-1, and b represents the bias term;

[0077] The template content is specifically as follows: all designed templates are stored in a template library, which is a database, file system or cloud storage service, and each template contains common fields and formats in the application form of this category, including: equipment name, fault description, maintenance time, application unit, and approval process;

[0078] The review and correction method includes: manual review, intelligent correction suggestions, feedback and iteration, and process optimization. The intelligent correction suggestions specifically provide correction suggestions for errors or omissions in the automatic filling process based on historical data, user feedback or machine learning models. The process optimization introduces automated testing or continuous integration to regularly evaluate and optimize the automated processing process.

[0079] Preferably, the document digitization is to quickly convert the image into an electronic document through optical character recognition, the image preprocessing is specifically to perform denoising, binarization and tilt correction operations on the input image, the feature extraction is to extract text features in the image through edge detection and morphological processing, the text features include stroke thickness and curvature, the pattern matching is to compare the extracted text features with the characters in the database through the Tesseract-OCR model, the data preprocessing is to further optimize the image quality through sharpening and contrast stretching, the expanded training data set is: to expand the training data set for the field of distribution network maintenance technology and Chinese and English, and the method of expanding the training data set includes: data collection, data enhancement, transfer learning and pre-training model.

[0080] Preferably, the text preprocessing is specifically: preprocessing the text description in the distribution network maintenance application form, including removing punctuation marks, spaces, line breaks, and performing unified processing of text encoding, the word segmentation processing is to use the Jieba word segmentation algorithm and select the precise mode to segment the preprocessed text, and the expression for calculating word frequency is:

[0081]

[0082] Among them, TF(t, d) represents the frequency of word t in document d, and f t，d represents the number of times word t appears in document d, ∑ k f k，d Represents the total number of times all words appear in document d;

[0083] The expression for calculating the inverse document frequency is:

[0084]

[0085] Where IDF(t) represents the inverse document frequency of word t, N represents the total number of documents, and |{d∈D:t∈d}| represents the number of applications for word t.

[0086] The extraction and classification of keywords specifically includes: sorting the words according to the TF-IDF value of each word, selecting the words with the highest values ​​as the keywords of the application form, and outputting the extracted key information in a structured form for subsequent processing.

[0087] Preferably, the analysis requirements are: analyzing various types, common fields, and format specifications of distribution network maintenance application forms to determine specific requirements for template design; the design template is: designing multiple templates based on the requirements analysis results, each template corresponding to a specific type of application form, the template contains fields that must be filled in the application form of this category, and sets the arrangement order and format of the fields; the classification storage is: classifying the designed templates according to the equipment type, maintenance nature, and application unit dimensions, and storing them in a template library, which is a database, file system, or cloud storage service; the template feature extraction is: extracting key features from the preprocessed text, including: keywords, key phrases, dates and times in a specific format; the AC automaton includes the following steps: constructing a prefix tree, adding mismatch pointers, and pattern matching; the Trie tree includes the following operation steps: inserting, deleting, and querying; the time complexity of the query operation is proportional to the length of the target string; the KMP algorithm includes The following steps include: constructing a prefix table, initialization, matching process and determining the sliding position. Specifically, the prefix table is constructed as follows: the i-th element of the prefix table represents the longest common prefix and suffix length of the substring consisting of the first i+1 characters of the pattern string. Specifically, the initialization is as follows: before starting the match, initialize two pointers i and j to point to the starting positions of the text string and the pattern string respectively. At the same time, create a prefix table to store the longest common prefix and suffix information of the pattern string. Specifically, the matching process is as follows: starting from the beginning of the text string, compare the text string and the pattern string character by character. When the characters do not match, determine the sliding position of the pattern string according to the prefix table. Specifically, the sliding position is determined as follows: assuming that the pattern string is "ABCABD" and the text string is "AAACAABDA", during the matching process, when the fifth character 'C' of the text string does not match the third character 'C' of the pattern string, according to the prefix table, the pattern string will slide forward 2 places so that the first character of the pattern string is aligned with the fifth character of the text string.

[0088] Preferably, the template matching is to automatically classify the templates using a support vector machine, and the specific steps of the template matching are:

[0089] (1) Data preparation: First, prepare a dataset for training and testing. The dataset contains labeled samples, each of which has a set of features and a corresponding category label. Then, use the train_test_split function to divide the dataset into a training set and a test set, with a ratio of 70% training set and 30% test set.

[0090] (2) Feature extraction and selection: Extract appropriate features from the original data as input, perform feature scaling, standardization or normalization operations on the data, and use StandardScaler for feature scaling;

[0091] (3) Data preprocessing: Use filtering or smoothing to denoise the data and fill or delete missing values;

[0092] (4) Model building and training: According to the problem requirements, select a linear kernel function, a polynomial kernel function, or a Gaussian radial basis function kernel and parameter configuration corresponding to the problem requirements to build a support vector machine classification model, input the prepared training data into the libsvm library, and perform model training based on the selected kernel function and parameters;

[0093] (5) Model evaluation: Use the test data set to evaluate the trained model, calculate the classification accuracy, precision, recall rate, F1 value, draw the confusion matrix, and analyze the performance of the model in different categories;

[0094] (6) Model optimization: According to the evaluation results, adjust the parameter configuration of the support vector machine model, including the parameters of the kernel function and the regularization parameter, and use k-fold cross validation to evaluate the stability and generalization ability of the model;

[0095] (7) Prediction and classification: Input the features of new unknown samples into the model and classify them into different categories according to the decision boundary of the model.

[0096] More preferably, the step of using StandardScaler to perform feature scaling is:

[0097] (1) Import library: Import StandardScaler from Scikit-learn;

[0098] (2) Prepare feature datasets;

[0099] (3) Instantiate StandardScaler: Create a StandardScaler object;

[0100] (4) Fitting and transforming data: Use the fit_transform method to fit, transform, and standardize features of the training data, and use the transform method to transform the test set;

[0101] (5) View scaled data: Output scaled data to verify whether the features have been standardized;

[0102] The expression of the standardized feature is:

[0103]

[0104] Among them, μ represents the mean, σ represents the standard deviation, and x′ i represents the eigenvalue of the i-th sample after standardization, x i represents the characteristic value of the i-th sample, and N represents the total number of samples in the data set.

[0105] More preferably, the steps of evaluating the stability and generalization ability of the model by k-fold cross validation are:

[0106] (1) Split the data set:

[0107]

[0108] Among them, D represents the entire data set, S i represents the i-th subset or fold, represents the union of all subsets, S j represents the jth subset or fold, and k represents the number of subsets or folds;

[0109] (2) The i-th training and testing: Training set: T i =D\S i , test set: S i ;

[0110] (3) Model evaluation: In the i-th time, train the model and calculate the performance index E i , the performance index E i The expression is:

[0111] E i =Performance(M,S i ),

[0112] Among them, M represents the training model, S i represents the i-th subset or fold;

[0113] (4) Calculate the average performance:

[0114]

[0115] Where k represents the number of subsets or folds, E i Indicates performance index, E mean Indicates the stability and generalization ability of the model on the dataset.

[0116] Preferably, the template definition and standardization are as follows: defining the structure and necessary fields of the maintenance order template, including: equipment name, fault description, maintenance steps, collecting and organizing the data sources used to fill out the maintenance order, including: equipment information library, historical fault records, spare parts inventory, and cleaning and formatting the data sources; the logical verification includes setting automated verification rules and using machine learning models to identify potential errors or anomalies.

[0117] Preferably, the automated verification rules include: required field check, data type and format check, logical relationship check, range check, and uniqueness check; the machine learning model is a text analysis model: an NLP model.

[0118] Actual deployment and use:

[0119] 1. System architecture and deployment:

[0120] System architecture:

[0121] The client-server (C / S) or browser-server (B / S) architecture is adopted, and it is selected according to actual needs. The system includes user interface layer, business logic layer, data processing layer and data storage layer.

[0122] The user interface layer provides a graphical interface for personnel to input maintenance application information; the business logic layer contains the core intelligent algorithms and logic processing modules; the data processing layer performs specific data processing tasks; and the data storage layer stores the data required by the system.

[0123] System deployment:

[0124] Deployed in the internal network of power companies to ensure data security and convenient access;

[0125] The system server is configured with high-performance hardware and a stable operating system to support the processing and concurrent access of large amounts of data.

[0126] 2. Template development and management:

[0127] Template development:

[0128] According to the different types and scenarios of distribution network maintenance, the system administrator or professional technicians shall develop standardized maintenance application filling templates.

[0129] The template contains necessary fields, format requirements, preset options and default values ​​to ensure the standardization and consistency of the reported content.

[0130] Template Management:

[0131] It provides functions of adding, deleting, modifying and checking templates, allowing system administrators to adjust template contents according to actual needs; the template library supports dynamic updates to ensure the timeliness and accuracy of templates.

[0132] 3. Intelligent filling process:

[0133] User login and template selection:

[0134] The reporting personnel log in to the system using their username and password;

[0135] Select the corresponding maintenance type on the system main interface, and the system will automatically load the corresponding reporting template.

[0136] Enter the power outage equipment information:

[0137] The reporting personnel fill in the basic information of the power outage equipment in the template, such as equipment name, number, location, etc.;

[0138] The system uses intelligent recognition technology to automatically retrieve and match the detailed information of the device from the database.

[0139] Intelligent generation of safety measures:

[0140] Based on the primary device type input by the user, the system intelligently identifies and associates all relevant primary and secondary devices; intelligently matches the corresponding template from the safety measure template library, and uses intelligent filling technology to automatically fill key information into the template to generate a complete safety measure description.

[0141] Data verification and submission:

[0142] The system conducts strict verification and validation on the reported data to ensure the integrity and consistency of the data;

[0143] For errors or abnormal situations found, the system provides intelligent error prompts and suggestions to help users correct them quickly;

[0144] The person filling out the form shall confirm that it is correct and submit the maintenance application.

[0145] 4. Subsequent processing and feedback:

[0146] Approval Process:

[0147] After the maintenance application is submitted, it enters the approval process. The approval personnel will review the application through the system and give approval opinions;

[0148] The approval process supports multi-level approval and process customization to ensure the rigor and flexibility of approval.

[0149] Feedback and Optimization:

[0150] The system collects user feedback and usage data to continuously optimize and improve reporting templates and intelligent algorithms;

[0151] Maintain and upgrade the system regularly to ensure system stability and security.

[0152] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent reporting system for distribution network maintenance application, characterized in that: include: Automatic identification of maintenance application form filling requirements algorithm model, document classification automatic matching template and template automatic matching filling module, the automatic identification of maintenance application form filling requirements algorithm model includes: data collection and collation and image recognition processing, the data collection is to collect a large amount of maintenance form filling format data, real-time monitoring data and historical maintenance application form data, and classify and collect them, the image recognition processing includes: document digitization, image preprocessing, feature extraction, pattern matching, data preprocessing and expansion of training data set, the data preprocessing includes: text preprocessing and word segmentation processing, the word segmentation processing includes: calculating word frequency, calculating inverse document frequency and extracting keywords and classifying them; The document classification automatic matching template includes: template design and matching process, the template design includes: template design and classification, template matching process, template optimization and template content, the template design and classification includes: analyzing requirements, designing templates and classified storage, the template matching process includes: text preprocessing, template feature extraction, template selection, the template selection is realized by keyword automatic matching template or classifier, the keyword automatic matching template includes: AC automaton, Trie tree and KMP algorithm, the matching process includes: text preprocessing, key feature extraction and template matching, the template matching specifically compares the extracted features with the templates in the template library to find the best matching template, and the template matching matching algorithm is similarity calculation or regular expression matching; The template automatic matching and filling module includes: a template automatic matching and filling method and a review and correction method. The template automatic matching and filling method includes: template definition and standardization, automatic matching algorithm, intelligent filling and logic verification. The automatic matching algorithm is a logistic regression model. The expression of the logistic regression model is: Y=σ(WX+b), z=WX+b, Among them, Y represents the output probability of the model, σ represents the Sigmoid function, W represents the weight vector, X represents the input feature vector, and b represents the bias term. Represents the classification label. If Y>0.5, the model predicts that the sample belongs to category 1. If Y≤0.5, the model predicts that the sample belongs to category 0. The intelligent filling is to automatically extract relevant information from the data source using a recurrent neural network model and fill it into the corresponding fields of the maintenance order. The expression of the recurrent neural network model is: h t =f(W xh ·X t +W hh ·h t-1 +b), Among them, h t represents the hidden state at the current time t, f represents the nonlinear activation function, W xh represents the weight of the input to the hidden state, X t represents the input features at the current time t, W hh represents the weight from hidden state to hidden state, h t-1 represents the hidden state at the previous moment t-1, and b represents the bias term; The template content is specifically as follows: all designed templates are stored in a template library, which can be a database, file system or cloud storage service, and each template contains common fields and formats in the application form of this category, including: equipment name, fault description, maintenance time, application unit, and approval process; The review and correction method includes: manual review, intelligent correction suggestions, feedback and iteration, and process optimization. The intelligent correction suggestions specifically provide correction suggestions for errors or omissions in the automatic filling process based on historical data, user feedback or machine learning models. The process optimization introduces automated testing or continuous integration to regularly evaluate and optimize the automated processing process.

2. According to claim 1, a distribution network maintenance application intelligent reporting system is characterized in that: The document digitization is to quickly convert an image into an electronic document through optical character recognition, the image preprocessing is specifically to perform denoising, binarization and tilt correction operations on the input image, the feature extraction is to extract text features in the image through edge detection and morphological processing, and the text features include stroke thickness and curvature, the pattern matching is to compare the extracted text features with the characters in the database through the Tesseract-OCR model, the data preprocessing is to further optimize the image quality through sharpening and contrast stretching, the expanded training data set is: to expand the training data set for the field of distribution network maintenance technology and Chinese and English, and the method for expanding the training data set includes: data collection, data enhancement, transfer learning and pre-training model.

3. According to claim 1, a distribution network maintenance application intelligent reporting system is characterized in that: The text preprocessing is specifically: preprocessing the text description in the distribution network maintenance application form, including removing punctuation marks, spaces, line breaks, and performing unified text encoding processing. The word segmentation processing is to use the Jieba word segmentation algorithm and select the precise mode to segment the preprocessed text. The expression for calculating word frequency is: Among them, TF(t, d) represents the frequency of word t in document d, and f t，d represents the number of times word t appears in document d, ∑ k f k，d Represents the total number of times all words appear in document d; The expression for calculating the inverse document frequency is: Where IDF(t) represents the inverse document frequency of word t, N represents the total number of documents, and |{d∈D:t∈d}| represents the number of applications for word t. The extraction and classification of keywords specifically includes: sorting the words according to the TF-IDF value of each word, selecting the words with the highest values ​​as the keywords of the application form, and outputting the extracted key information in a structured form for subsequent processing.

4. According to claim 1, a distribution network maintenance application intelligent reporting system is characterized in that: The analysis requirements are: analyzing various types, common fields, and format specifications of distribution network maintenance application forms to determine specific requirements for template design. The design template is: designing multiple templates based on the requirements analysis results, each template corresponds to a specific type of application form, and the template contains fields that must be filled in the application form of this category, and sets the arrangement order and format of the fields. The classification storage is: classifying the designed templates according to the equipment type, maintenance nature, and application unit dimensions, and storing them in a template library. The template library is a database, file system, or cloud storage service. The template feature extraction is: extracting key features from the preprocessed text, including: keywords, key phrases, dates and times in a specific format. The AC automaton includes the following steps: constructing a prefix tree, adding mismatch pointers, and pattern matching. The Trie tree includes the following operation steps: inserting, deleting, and querying. The time complexity of the query operation is proportional to the length of the target string. The KMP algorithm includes the following Steps: constructing a prefix table, initialization, matching process and determining the sliding position. The construction of the prefix table is specifically as follows: the i-th element of the prefix table represents the longest common prefix and suffix length of the substring composed of the first i+1 characters of the pattern string. The initialization is specifically as follows: before starting the match, initialize two pointers i and j to point to the starting positions of the text string and the pattern string respectively. At the same time, create a prefix table to store the longest common prefix and suffix information of the pattern string. The matching process is specifically as follows: starting from the beginning of the text string, compare the text string and the pattern string character by character. When the characters do not match, determine the sliding position of the pattern string according to the prefix table. The determination of the sliding position is specifically as follows: assuming that the pattern string is "ABCABD" and the text string is "AAACAABDA", during the matching process, when the fifth character 'C' of the text string does not match the third character 'C' of the pattern string, according to the prefix table, the pattern string will slide forward 2 places so that the first character of the pattern string is aligned with the fifth character of the text string.

5. According to claim 1, a distribution network maintenance application intelligent reporting system is characterized in that: The template matching is to automatically classify the templates through a support vector machine. The specific steps of the template matching are: (1) Data preparation: First, prepare a dataset for training and testing. The dataset contains labeled samples, each of which has a set of features and a corresponding category label. Then, use the train_test_split function to divide the dataset into a training set and a test set, with a ratio of 70% training set and 30% test set. (2) Feature extraction and selection: Extract appropriate features from the original data as input, perform feature scaling, standardization or normalization operations on the data, and use StandardScaler for feature scaling; (3) Data preprocessing: Use filtering or smoothing to denoise the data and fill or delete missing values; (4) Model building and training: According to the problem requirements, select a linear kernel function, a polynomial kernel function, or a Gaussian radial basis function kernel and parameter configuration corresponding to the problem requirements to build a support vector machine classification model, input the prepared training data into the libsvm library, and perform model training based on the selected kernel function and parameters; (5) Model evaluation: Use the test data set to evaluate the trained model, calculate the classification accuracy, precision, recall rate, F1 value, draw the confusion matrix, and analyze the performance of the model in different categories; (6) Model optimization: According to the evaluation results, adjust the parameter configuration of the support vector machine model, including the parameters of the kernel function and the regularization parameter, and use k-fold cross validation to evaluate the stability and generalization ability of the model; (7) Prediction and classification: Input the features of new unknown samples into the model and classify them into different categories according to the decision boundary of the model.

6. According to claim 5, a distribution network maintenance application intelligent reporting system is characterized in that: The steps of using StandardScaler for feature scaling are: (1) Import library: Import StandardScaler from Scikit-learn; (2) Prepare feature datasets; (3) Instantiate StandardScaler: Create a StandardScaler object; (4) Fitting and transforming data: Use the fit_transform method to fit, transform, and standardize features of the training data, and use the transform method to transform the test set; (5) View scaled data: Output scaled data to verify whether the features have been standardized; The expression of the standardized feature is: Among them, μ represents the mean, σ represents the standard deviation, and x′ i represents the eigenvalue of the i-th sample after standardization, x i represents the characteristic value of the i-th sample, and N represents the total number of samples in the data set.

7. The intelligent reporting system for distribution network maintenance application according to claim 5 is characterized in that: The steps of the k-fold cross validation to evaluate the stability and generalization ability of the model are: (1) Split the data set: Among them, D represents the entire data set, S i represents the i-th subset or fold, represents the union of all subsets, S j represents the jth subset or fold, and k represents the number of subsets or folds; (2) The i-th training and testing: Training set: T i =D\S i , test set: S i ; (3) Model evaluation: In the i-th time, train the model and calculate the performance index E i , the performance index E i The expression is: E i =Performance(M,S i ), Among them, M represents the training model, S i represents the i-th subset or fold; (4) Calculate the average performance: Where k represents the number of subsets or folds, E i Indicates performance index, E mean Indicates the stability and generalization ability of the model on the dataset.

8. The intelligent reporting system for distribution network maintenance application according to claim 1 is characterized in that: The template definition and standardization are as follows: defining the structure and necessary fields of the maintenance order template, including: equipment name, fault description, maintenance steps, collecting and organizing the data sources used to fill out the maintenance order, including: equipment information library, historical fault records, spare parts inventory, and cleaning and formatting the data sources; the logical verification includes setting automated verification rules and using machine learning models to identify potential errors or anomalies.

9. The intelligent reporting system for distribution network maintenance application according to claim 8 is characterized in that: The automated verification rules include: required field check, data type and format check, logical relationship check, range check, and uniqueness check; the machine learning model is a text analysis model: an NLP model.

Images