A fault intelligent detection system for power distribution cabinet

Abstract

The present invention discloses a fault intelligent detection system for a power distribution cabinet, including a data collection module, an enhanced processing module, a fault diagnosis module, a repair generation module and an intelligent detection module. The present invention belongs to the field of fault detection and prediction technology for power distribution cabinets, and specifically is a fault intelligent detection system for power distribution cabinets. The present invention adopts a fault diagnosis and repair solution generation method combining frequency difference grouping and frequency acquisition data and multi-index data analysis, and performs fault diagnosis by combining multi-dimensional data of three different acquisition frequencies, and generates repair results of repair solutions, which provides powerful data reference and support for fault detection and maintenance of power distribution cabinets; adopts a bidirectional long short-term memory model combined with frequency difference grouping noise optimization for fault diagnosis, which effectively combines the integrated analysis of high-frequency and low-frequency acquisition data; adopts a decision tree regression model method based on data-driven to perform repair generation, thereby improving the availability and reliability of the system.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault detection and prediction of power distribution cabinets, and specifically relates to an intelligent fault detection system for power distribution cabinets. Background Art

[0002] The intelligent fault detection system for distribution cabinets uses machine learning algorithms and integrates sensors, data acquisition and analysis algorithms to monitor the working status of distribution cabinets. The system collects key parameters such as current, voltage, temperature, etc., combined with fault pattern recognition, and can quickly detect potential faults to prevent equipment damage and power outages, thereby improving the safety and reliability of the power system.

[0003] However, in the existing intelligent detection system for distribution cabinet faults, there are traditional methods that mainly focus on simple physical faults of distribution cabinets, while in the actual needs for intelligent detection of distribution cabinet faults, there are still complex and diverse safety, environmental adaptability, maintenance convenience and real-time requirements. The simple physical fault analysis cannot meet the intelligent needs of distribution cabinet fault detection. The technical problem; in the existing intelligent diagnosis method for distribution cabinet faults, there is a technical problem that fault detection for distribution cabinet sensor data needs to ensure the accuracy of detection to avoid unnecessary power outages and maintenance, and also needs to perform multi-feature analysis from the perspective of environmental conditions, and the existing method cannot take into account the two aspects for integrated detection. The technical problem; in the existing intelligent repair method for distribution cabinet faults, there is a technical problem that the existing method is difficult to combine the diagnosed faults to further provide intelligent support for the maintenance of the distribution cabinet. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a fault intelligent detection system for distribution cabinets. In the existing distribution cabinet fault intelligent detection system, there is a traditional method that mainly focuses on the simple physical faults of the distribution cabinet. In the actual distribution cabinet fault intelligent detection needs, there are still complex and diverse safety, environmental adaptability, maintenance convenience and real-time requirements. The simple physical fault analysis cannot meet the technical problem of the intelligent needs of distribution cabinet fault detection. This solution creatively adopts a fault diagnosis and repair solution generation method that combines frequency difference grouping and frequency acquisition data and multi-index data analysis. By combining multi-dimensional data with three different acquisition frequencies, fault diagnosis is performed, and by generating the repair results of the repair solution, a powerful data reference and support is provided for the distribution cabinet fault detection and maintenance. In the existing distribution cabinet fault intelligent diagnosis method, there is a fault detection for the distribution cabinet sensor data that needs to ensure the accuracy of the detection to avoid unnecessary power outages and maintenance, and also needs to perform multi-feature analysis from the perspective of environmental conditions, and the existing method cannot take both into account. In order to solve the technical problem of integrated detection of the contents of the project, this solution creatively adopts a bidirectional long short-term memory model combined with frequency difference grouping noise optimization for fault diagnosis. By combining the multi-frequency acquisition data of the frequency difference grouping data, richer deep fault features are extracted, and then the misjudgment rate of the algorithm is reduced by designing a noise optimization operator. In combination with the bidirectional time series network integrated random forest classifier, the integrated analysis of high-frequency and low-frequency acquisition data is effectively combined, thereby improving the accuracy of detection, while taking into account the in-depth analysis of various environmental characteristics, and improving the interpretability and availability of the detection results. In view of the technical problem that the existing methods are difficult to combine the diagnosed faults to further provide intelligent support for the maintenance of the distribution cabinet in the existing intelligent repair methods for distribution cabinet faults, this solution creatively adopts a data-driven decision tree regression model method for repair generation. By training the decision tree and regression model, they are responsible for identifying the cause of the fault, matching the fault repair plan and predicting the fault repair effect respectively, providing full intelligent support for the maintenance process of the fault repair plan, and improving the availability and reliability of the distribution cabinet fault intelligent detection system.

[0005] The technical solution adopted by the present invention is as follows: A fault intelligent detection system for a power distribution cabinet provided by the present invention comprises a data collection module, an enhanced processing module, a fault diagnosis module, a repair generation module and an intelligent detection module;

[0006] Furthermore, the data collection module is used to collect the original data set required for intelligent fault detection of the power distribution cabinet, specifically by deploying a sensor cluster and performing data collection to obtain the original data of comprehensive detection of the power distribution cabinet;

[0007] The power distribution cabinet comprehensively detects the original data, specifically including current data, temperature data, humidity data, vibration frequency data and historical maintenance and repair data;

[0008] The historical maintenance and repair data specifically includes historical maintenance record data, historical fault repair solution record data, and historical fault repair and maintenance result data;

[0009] The data collection specifically refers to group frequency difference data collection, including collecting data once every 1 second, collecting data once every 5 seconds and collecting data once every 10 seconds, and obtaining the original data of the comprehensive detection of the distribution cabinet through the group frequency difference data collection, and storing the original data of the comprehensive detection of the distribution cabinet in the local database.

[0010] Furthermore, the enhanced processing module is used to optimize and enhance the original data, specifically by sequentially performing data cleaning and feature processing on the original data of the comprehensive detection of the distribution cabinet to obtain enhanced distribution cabinet detection data, and the specific steps include: data cleaning, feature processing, data standardization and enhanced comprehensive processing;

[0011] The enhanced distribution cabinet detection data specifically includes a cleaning optimization data set, a feature data set and a standardized data set.

[0012] Furthermore, the fault diagnosis module is used to analyze and identify potential faults during the operation of the power distribution cabinet, specifically, based on the enhanced power distribution cabinet detection data, a bidirectional long short-term memory model combined with frequency difference grouping noise optimization is used to perform fault diagnosis to obtain power distribution cabinet fault diagnosis data;

[0013] The bidirectional long short-term memory model combined with frequency difference grouping noise optimization specifically includes a frequency difference set empirical mode decomposition operator, a frequency difference feature optimization operator, a bidirectional long short-term memory sub-model, a random forest classification sub-model and a parameter optimization operator;

[0014] The step of using a bidirectional long short-term memory model combined with frequency difference grouping noise optimization to perform fault diagnosis and obtain distribution cabinet fault diagnosis data includes: constructing a frequency difference set empirical mode decomposition operator, constructing a frequency difference feature optimization operator, constructing a bidirectional long short-term memory sub-model, constructing a random forest classification sub-model, constructing a parameter optimization operator, fault diagnosis model training and fault diagnosis;

[0015] The frequency difference set empirical mode decomposition operator is constructed to optimize the characteristic data in the enhanced distribution cabinet detection data and reduce the interference of noise on data quality. Specifically, the set empirical mode decomposition adaptive noise algorithm is used to perform adaptive mode decomposition on the enhanced distribution cabinet detection data to obtain frequency difference intrinsic characteristic data;

[0016] The constructing of the frequency difference feature optimization operator specifically includes performing frequency difference feature optimization extraction on the frequency difference intrinsic feature data to obtain frequency difference feature data, and performing standard deviation and mean calculation on the frequency difference feature data to obtain an optimized frequency difference feature data set;

[0017] The bidirectional long short-term memory sub-model is specifically constructed by constructing a standard bidirectional long short-term memory neural network, and performing deep extraction of time series features according to the acquisition frequency to obtain frequency difference hidden state output data. The calculation formula is:

[0018] ;

[0019] In the formula, h t is the frequency difference hidden state output data, which is used to represent the feature output data after weighted integration according to different sampling frequency weights. a1 is the 1 second sampling frequency weight, a5 is the 5 second sampling frequency weight, and a 10 is the 10-second sampling frequency weight, h t,1 is the frequency difference hidden state output corresponding to the 1 second sampling frequency, h t,5 is the frequency difference hidden state output corresponding to the 5-second sampling frequency, h t,10 It is the frequency difference hidden state output corresponding to the 10-second sampling frequency;

[0020] The construction of the random forest classification sub-model specifically includes respectively training the classifier of the standard random forest model according to the optimized frequency difference feature data sets corresponding to different sampling frequencies, and classifying the fault type by taking the frequency difference hidden state output data as the total input of the trained random forest classification sub-model to obtain the model classification output data;

[0021] The constructing parameter optimization operator specifically adopts a tree-structured Bayesian optimization algorithm to optimize model parameters, obtains a model hyperparameter optimization combination, and applies the model hyperparameter optimization combination to fault diagnosis model training;

[0022] The fault diagnosis model training is specifically to perform fault diagnosis model training through the frequency difference set empirical mode decomposition operator, the frequency difference feature optimization operator, the bidirectional long short-term memory sub-model, the random forest classification sub-model and the parameter optimization operator to obtain the fault diagnosis model Model IF ;

[0023] The fault diagnosis is specifically performed using the fault diagnosis model Model IF , performing fault diagnosis based on the enhanced power distribution cabinet detection data to obtain power distribution cabinet fault diagnosis data;

[0024] The distribution cabinet fault diagnosis data specifically includes fault probability reference data, fault type reference data and fault diagnosis index reference data;

[0025] The fault probability reference data is used to represent the quantitative confidence value of the fault, and the value range is [0,100%];

[0026] The fault type reference data is used to indicate identifiable fault categories, including overload fault, short circuit fault, poor contact fault, aging fault, overheating fault, abnormal moisture fault and insulation fault;

[0027] The fault diagnosis indicator reference data is used to provide an identification basis to optimize the interpretability of fault diagnosis, and specifically includes fault occurrence time record prediction data, fault duration prediction data and fault impact location reference data.

[0028] Furthermore, the repair generation module is used to identify the cause of the fault, generate a repair plan and predict the repair effect. Specifically, based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, a data-driven decision tree regression model method is used to perform repair generation to obtain distribution cabinet fault repair reference data. The specific steps include: data integration, feature coding optimization, building a decision tree model, building a regression model, repair generation model training and repair plan generation;

[0029] The data integration specifically includes integrating the enhanced power distribution cabinet detection data and the power distribution cabinet fault diagnosis data to obtain an integrated repair generation data set, and performing a data cleaning operation on the integrated repair generation data set to obtain an integrated repair generation optimization data set;

[0030] The feature encoding optimization specifically comprises performing a feature extraction operation on the integrated repair generation optimization data set to obtain a repair generation feature data set, and obtaining a coding feature matrix set by performing one-hot encoding on the data features in the repair generation feature data set;

[0031] The repair generates a feature data set, specifically including fault type features, fault frequency features, historical repair solution features, repair method features and environmental condition auxiliary features;

[0032] The construction of the decision tree model specifically includes using a decision tree algorithm to perform fault repair pre-classification training to obtain a fault cause pre-classification identification model;

[0033] The constructing of the regression model specifically comprises using the fault cause pre-classification and identification model to perform fault cause pre-classification and identification according to the coding feature matrix set to obtain fault cause pre-identification data, and adopting a rule-based method to extract a repair solution according to the fault cause pre-identification data from the integrated repair generation optimization data set to obtain a fault repair candidate list, and performing regression model training according to the fault repair candidate list to obtain a repair effect prediction regression model;

[0034] The repair effect prediction regression model is used to predict the effect of the fault repair solutions in the fault repair candidate list to obtain a repair effect prediction output;

[0035] The repair generation model training is specifically to perform repair generation model training through the data integration, the feature coding optimization, the construction of the decision tree model and the construction of the regression model to obtain the repair generation model Model RP ;

[0036] The repair solution generation is specifically to use the repair generation model Model RP , generating a repair plan based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, and obtaining distribution cabinet fault repair reference data through prediction output integration;

[0037] The distribution cabinet fault repair reference data specifically includes fault cause pre-identification data, fault repair solution prediction data and repair effect prediction data.

[0038] Furthermore, the intelligent detection module is used to perform comprehensive intelligent detection in combination with the distribution cabinet fault diagnosis data and the distribution cabinet fault repair reference data. Specifically, the distribution cabinet fault diagnosis data is obtained, the distribution cabinet fault information is recorded, and the distribution cabinet fault repair reference data is obtained by applying the repair generation module. A repair plan is selected from the distribution cabinet fault repair reference data to perform intelligent detection and intelligent repair of the distribution cabinet fault, and a distribution cabinet fault intelligent detection reference data set is obtained. The fault data, repair plan data and the results of the repair plan in the distribution cabinet fault intelligent detection reference data set are manually recorded for system optimization and maintenance.

[0039] The data collection module is used for data collection, and obtains the original data of the comprehensive detection of the power distribution cabinet through data collection, and sends the original data of the comprehensive detection of the power distribution cabinet to the enhanced processing module;

[0040] The enhanced processing module is used for enhanced processing, and obtains enhanced distribution cabinet detection data through enhanced processing, and sends the enhanced distribution cabinet detection data to the fault diagnosis module and the repair generation module;

[0041] The fault diagnosis module is used for fault diagnosis, obtains the distribution cabinet fault diagnosis data through fault diagnosis, and sends the distribution cabinet fault diagnosis data to the repair generation module and the intelligent detection module;

[0042] The repair generation module is used for repair generation, obtains distribution cabinet fault repair reference data through repair generation, and sends the distribution cabinet fault repair reference data to the intelligent detection module;

[0043] The intelligent detection module is used for intelligent detection, and obtains a reference data set for intelligent detection of distribution cabinet faults through intelligent detection.

[0044] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0045] (1) In the existing intelligent detection system for distribution cabinet faults, the traditional methods mainly focus on simple physical faults of distribution cabinets. However, in the actual intelligent detection needs for distribution cabinet faults, there are still complex and diverse safety, environmental adaptability, maintenance convenience and real-time requirements. Simple physical fault analysis cannot meet the technical problem of intelligent needs for distribution cabinet fault detection. This solution creatively adopts a fault diagnosis and repair solution generation method that combines frequency difference grouping and frequency collection data with multi-index data analysis. By combining multi-dimensional data of three different collection frequencies, fault diagnosis is performed, and by generating the repair results of the repair solution, it provides a strong data reference and support for distribution cabinet fault detection and maintenance;

[0046] (2) In the existing intelligent diagnosis methods for distribution cabinet faults, there is a technical problem that the fault detection for distribution cabinet sensor data needs to ensure the accuracy of the detection to avoid unnecessary power outages and maintenance, and also needs to perform multi-feature analysis from the perspective of environmental conditions. The existing methods cannot take both into account for integrated detection. This solution creatively adopts a bidirectional long short-term memory model combined with frequency difference grouping noise optimization for fault diagnosis. By combining the multi-frequency acquisition data of the frequency difference grouping data, richer deep fault features are extracted, and then the misjudgment rate of the algorithm is reduced by designing a noise optimization operator. In addition, the bidirectional time series network is combined with a random forest classifier to effectively combine the integrated analysis of high-frequency and low-frequency acquisition data, thereby improving the accuracy of detection. It also takes into account the in-depth analysis of multiple environmental features, and improves the interpretability and usability of the detection results.

[0047] (3) In view of the technical problem that the existing intelligent repair methods for distribution cabinet faults are difficult to combine with the diagnosed faults to further provide intelligent support for the maintenance of the distribution cabinet, this solution creatively adopts a data-driven decision tree regression model method to generate repairs. By training the decision tree and regression model to identify the cause of the fault, match the fault repair plan, and predict the fault repair effect, it provides full intelligent support for the maintenance process of the fault repair plan, thereby improving the availability and reliability of the distribution cabinet fault intelligent detection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A structural schematic diagram of a fault intelligent detection system for a power distribution cabinet provided by the present invention;

[0049] Figure 2 A schematic flow chart of steps executed by a fault intelligent detection system for a power distribution cabinet provided by the present invention;

[0050] Figure 3 A flowchart of the steps performed by the enhancement processing module;

[0051] Figure 4 A flowchart of the steps performed by the fault diagnosis module;

[0052] Figure 5 A flowchart of the steps performed to repair the generated module;

[0053] Figure 6 A visualization diagram of the fault probability reference data corresponding to the 1s acquisition frequency drawn for the fault diagnosis module;

[0054] Figure 7 A visualization diagram of the fault probability reference data corresponding to the 5s acquisition frequency drawn for the fault diagnosis module;

[0055] Figure 8 A visualization diagram of the fault probability reference data corresponding to the 10s acquisition frequency drawn for the fault diagnosis module;

[0056] Fig. 9 Visualization diagram of the weighted integrated fault probability reference data drawn for the fault diagnosis module.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] Example 1, see Figure 1 and Figure 2 , the present invention provides a fault intelligent detection system for a power distribution cabinet, comprising a data collection module, an enhanced processing module, a fault diagnosis module, a repair generation module and an intelligent detection module;

[0060] The data collection module is used for data collection, and obtains the original data of the comprehensive detection of the power distribution cabinet through data collection, and sends the original data of the comprehensive detection of the power distribution cabinet to the enhanced processing module;

[0061] The enhanced processing module is used for enhanced processing, and obtains enhanced distribution cabinet detection data through enhanced processing, and sends the enhanced distribution cabinet detection data to the fault diagnosis module and the repair generation module;

[0062] The fault diagnosis module is used for fault diagnosis, obtains the distribution cabinet fault diagnosis data through fault diagnosis, and sends the distribution cabinet fault diagnosis data to the repair generation module and the intelligent detection module;

[0063] The repair generation module is used for repair generation, obtains distribution cabinet fault repair reference data through repair generation, and sends the distribution cabinet fault repair reference data to the intelligent detection module;

[0064] The intelligent detection module is used for intelligent detection, and obtains a reference data set for intelligent detection of distribution cabinet faults through intelligent detection.

[0065] By performing the above operations, in the existing intelligent fault detection system for distribution cabinets, there is a traditional method that mainly focuses on simple physical faults of distribution cabinets, while in the actual intelligent fault detection needs for distribution cabinets, there are still complex and diverse safety, environmental adaptability, maintenance convenience and real-time requirements. The simple physical fault analysis cannot meet the technical problem of intelligent needs of distribution cabinet fault detection. This solution creatively adopts a fault diagnosis and repair solution generation method that combines frequency difference grouping and frequency collection data and multi-indicator data analysis. By combining multi-dimensional data of three different collection frequencies, fault diagnosis is performed, and by generating the repair results of the repair solution, it provides powerful data reference and support for distribution cabinet fault detection and maintenance.

[0066] Embodiment 2, this embodiment is based on the above embodiment, see Figure 1 and Figure 2 , the data collection module is used to collect the original data set required for intelligent fault detection of the distribution cabinet, specifically by deploying a sensor cluster and performing data collection to obtain the original data of comprehensive detection of the distribution cabinet;

[0067] The power distribution cabinet comprehensively detects the original data, specifically including current data, temperature data, humidity data, vibration frequency data and historical maintenance and repair data;

[0068] Table 1 is a reference comparison table of the relationship between the sensors and data. As shown in the table, the sensor cluster specifically includes a current sensor, a temperature sensor, a humidity sensor and a vibration sensor;

[0069] The current sensor is used to collect real-time current value data and monitor the current intensity of the power distribution cabinet;

[0070] The temperature sensor is used to collect real-time temperature data and monitor the internal and external temperatures of the power distribution cabinet;

[0071] The humidity sensor is used to collect real-time humidity value data and monitor the internal humidity of the power distribution cabinet;

[0072] The vibration sensor is used to collect real-time distribution cabinet vibration frequency data and monitor the vibration conditions of the distribution cabinet and its internal accessories;

[0073] Table 1 Sensor and data relationship reference comparison table

[0074]

[0075] The historical maintenance and repair data specifically includes historical maintenance record data, historical fault repair solution record data, and historical fault repair and maintenance result data;

[0076] The data collection specifically refers to group frequency difference data collection, including collecting data once every 1 second, collecting data once every 5 seconds and collecting data once every 10 seconds, and obtaining the original data of the comprehensive detection of the distribution cabinet through the group frequency difference data collection, and storing the original data of the comprehensive detection of the distribution cabinet in the local database.

[0077] Embodiment 3, this embodiment is based on the above embodiment, see Figure 2 and Figure 3 The enhanced processing module is used to optimize and enhance the original data, specifically by sequentially performing data cleaning and feature processing on the original data of the comprehensive detection of the distribution cabinet to obtain enhanced distribution cabinet detection data, and the specific steps include: data cleaning, feature processing, data standardization and enhanced comprehensive processing;

[0078] The data cleaning specifically refers to performing a data integrity check on the raw data of the comprehensive detection of the power distribution cabinet, filling the missing values ​​with the mean filling method, and removing the outliers through outlier detection to obtain a cleaned and optimized data set;

[0079] The feature processing specifically refers to extracting features from the data given in the cleaning optimization data set using an artificial feature extraction method to obtain a feature data set;

[0080] The characteristic data set specifically includes current characteristics, temperature characteristics, humidity characteristics and vibration characteristics;

[0081] The current characteristics specifically include current mean characteristics, maximum current characteristics, minimum current characteristics and current standard deviation characteristics;

[0082] The temperature characteristics specifically include temperature mean characteristics and temperature change rate characteristics;

[0083] The humidity characteristics specifically include humidity mean characteristics and humidity change rate characteristics;

[0084] The vibration characteristics specifically include vibration frequency mean characteristics and vibration frequency peak characteristics;

[0085] The data standardization specifically adopts a Z-score standardization method to standardize the feature data set to obtain a standardized data set;

[0086] The enhanced comprehensive processing is specifically to obtain enhanced distribution cabinet detection data through the data cleaning, feature processing and data standardization;

[0087] The enhanced distribution cabinet detection data specifically includes a cleaning optimization data set, a feature data set and a standardized data set.

[0088] Embodiment 4, this embodiment is based on the above embodiment, see Figure 2 and Figure 4 The fault diagnosis module is used to analyze and identify potential faults during the operation of the distribution cabinet, specifically, based on the enhanced distribution cabinet detection data, a bidirectional long short-term memory model combined with frequency difference grouping noise optimization is used to perform fault diagnosis to obtain distribution cabinet fault diagnosis data;

[0089] The bidirectional long short-term memory model combined with frequency difference grouping noise optimization specifically includes a frequency difference set empirical mode decomposition operator, a frequency difference feature optimization operator, a bidirectional long short-term memory sub-model, a random forest classification sub-model and a parameter optimization operator;

[0090] The step of using a bidirectional long short-term memory model combined with frequency difference grouping noise optimization to perform fault diagnosis and obtain distribution cabinet fault diagnosis data includes: constructing a frequency difference set empirical mode decomposition operator, constructing a frequency difference feature optimization operator, constructing a bidirectional long short-term memory sub-model, constructing a random forest classification sub-model, constructing a parameter optimization operator, fault diagnosis model training and fault diagnosis;

[0091] The frequency difference set empirical mode decomposition operator is constructed to optimize the characteristic data in the enhanced distribution cabinet detection data and reduce the interference of noise on data quality. Specifically, the set empirical mode decomposition adaptive noise algorithm is used to perform adaptive mode decomposition on the enhanced distribution cabinet detection data to obtain the frequency difference intrinsic characteristic data. The calculation formula is:

[0092] ;

[0093] Where, X freq (·) is the frequency difference intrinsic feature data, where freq is a frequency difference index variable used to represent different acquisition frequencies, and the specific value range of the frequency difference index variable freq is {1,5,10}, n is the total number of intrinsic mode functions, j is the intrinsic mode function index, IMF(·) is the intrinsic mode function, r(·) is the noise residual signal representation function, and t is the time index;

[0094] The constructing of the frequency difference feature optimization operator specifically includes performing frequency difference feature optimization extraction on the frequency difference intrinsic feature data to obtain frequency difference feature data, and performing standard deviation and mean calculation on the frequency difference feature data to obtain an optimized frequency difference feature data set;

[0095] The bidirectional long short-term memory sub-model is specifically constructed by constructing a standard bidirectional long short-term memory neural network, and performing deep extraction of time series features according to the acquisition frequency to obtain frequency difference hidden state output data. The calculation formula is:

[0096] ;

[0097] In the formula, is the hidden state output of the forward LSTM neural network, is the hidden state output of the backward LSTM neural network, LSTM(·) is the standard LSTM neural network model representation function, x t,freq is the original input data, used to represent the optimized frequency difference feature data set, is the hidden state output of the forward long short-term memory neural network at the previous moment, is the hidden state output of the backward long short-term memory neural network at the previous moment, is the cell state parameter of the forward long short-term memory neural network at the previous moment, is the cell state parameter of the backward long short-term memory neural network at the previous moment, h t,freq is the hidden state output of the bidirectional network, which is used to represent the output connection vector of the forward and backward neural networks, where t is the time index, freq is the frequency difference index variable, [·;·] is the vector connection operator, and h t It is the frequency difference hidden state output data, which is used to represent the feature output data after weighted integration according to different sampling frequency weights. a1 is the 1 second sampling frequency weight, and the specific value is 0.2. a5 is the 5 second sampling frequency weight, and the specific value is 0.5. 10 is the 10-second sampling frequency weight, the specific value is 0.3, h t,1 is the frequency difference hidden state output corresponding to the 1 second sampling frequency, h t,5 is the frequency difference hidden state output corresponding to the 5-second sampling frequency, h t,10 It is the frequency difference hidden state output corresponding to the 10-second sampling frequency;

[0098] The random forest classification sub-model is constructed by respectively training the classifier of the standard random forest model according to the optimized frequency difference feature data sets corresponding to different sampling frequencies, and classifying the fault type by taking the frequency difference hidden state output data as the total input of the trained random forest classification sub-model to obtain the model classification output data. The calculation formula is:

[0099] ;

[0100] In the formula, F freq (x) is the prediction result of the random forest model, which is used to represent the classifier prediction result of the random forest model corresponding to different sampling frequencies. freq is the total number of random forest submodel trees corresponding to the frequency difference index variable freq, m is the tree index, h m,freq (x) is the predicted output of the mth random forest model tree corresponding to the frequency difference index variable freq, and x is the input feature of the random forest model;

[0101] F out (x) is the model classification output data, which is used to represent the final output of the random forest classification sub-model. a1 is the 1-second sampling frequency weight, with a specific value of 0.2. a5 is the 5-second sampling frequency weight, with a specific value of 0.5. 10 is the weight of the 10-second sampling frequency, with a specific value of 0.3. F1(·) is the prediction result of the random forest sub-model corresponding to the 1-second sampling frequency. F5(·) is the prediction result of the random forest sub-model corresponding to the 5-second sampling frequency. 10 (·) is the prediction result of the random forest model corresponding to a sampling frequency of 10 seconds;

[0102] The constructing parameter optimization operator specifically adopts a tree-structured Bayesian optimization algorithm to optimize model parameters, obtains a model hyperparameter optimization combination, and applies the model hyperparameter optimization combination to fault diagnosis model training;

[0103] Table 2 is a schematic table of the optimization parameter range of the model hyperparameter optimization combination, as shown in the table, the model hyperparameter optimization combination specifically includes the number of units, learning rate, training cycle, batch size and inactivation rate;

[0104] Table 2 Schematic diagram of optimization parameter ranges for model hyperparameter optimization combinations

[0105]

[0106] The fault diagnosis model training is specifically to perform fault diagnosis model training through the frequency difference set empirical mode decomposition operator, the frequency difference feature optimization operator, the bidirectional long short-term memory sub-model, the random forest classification sub-model and the parameter optimization operator to obtain the fault diagnosis model Model IF ;

[0107] The fault diagnosis is specifically performed using the fault diagnosis model Model IF , performing fault diagnosis based on the enhanced power distribution cabinet detection data to obtain power distribution cabinet fault diagnosis data;

[0108] The distribution cabinet fault diagnosis data specifically includes fault probability reference data, fault type reference data and fault diagnosis index reference data;

[0109] The fault probability reference data is used to represent the quantitative confidence value of the fault, and the value range is [0,100%];

[0110] The fault type reference data is used to indicate identifiable fault categories, including overload fault, short circuit fault, poor contact fault, aging fault, overheating fault, abnormal moisture fault and insulation fault;

[0111] The fault diagnosis indicator reference data is used to provide an identification basis to optimize the interpretability of fault diagnosis, and specifically includes fault occurrence time record prediction data, fault duration prediction data and fault impact location reference data.

[0112] By performing the above operations, in the existing intelligent diagnosis methods for distribution cabinet faults, there is a technical problem that fault detection for distribution cabinet sensor data needs to ensure the accuracy of detection to avoid unnecessary power outages and maintenance, and also needs to perform multi-feature analysis from the perspective of environmental conditions, and the existing methods cannot take into account both aspects for integrated detection. This solution creatively adopts a bidirectional long short-term memory model combined with frequency difference grouping noise optimization for fault diagnosis. By combining the multi-frequency acquisition data of frequency difference grouping data, richer deep fault features are extracted, and then the error rate of the algorithm is reduced by designing a noise optimization operator. In addition, the bidirectional time series network is combined with an integrated random forest classifier to effectively combine the integrated analysis of high-frequency and low-frequency acquisition data, thereby improving the accuracy of detection, while also taking into account the in-depth analysis of multiple environmental characteristics, and improving the interpretability and availability of the detection results.

[0113] Embodiment 5, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 5 The repair generation module is used to identify the cause of the fault, generate a repair plan and predict the repair effect. Specifically, based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, a data-driven decision tree regression model method is used to perform repair generation to obtain distribution cabinet fault repair reference data. The specific steps include: data integration, feature coding optimization, building a decision tree model, building a regression model, repair generation model training and repair plan generation;

[0114] The data integration specifically includes integrating the enhanced power distribution cabinet detection data and the power distribution cabinet fault diagnosis data to obtain an integrated repair generation data set, and performing a data cleaning operation on the integrated repair generation data set to obtain an integrated repair generation optimization data set;

[0115] The feature encoding optimization specifically comprises performing a feature extraction operation on the integrated repair generation optimization data set to obtain a repair generation feature data set, and obtaining a coding feature matrix set by performing one-hot encoding on the data features in the repair generation feature data set;

[0116] The repair generates a feature data set, specifically including fault type features, fault frequency features, historical repair solution features, repair method features and environmental condition auxiliary features;

[0117] The construction of the decision tree model specifically includes using a decision tree algorithm to perform fault repair pre-classification training to obtain a fault cause pre-classification identification model;

[0118] The decision tree algorithm specifically uses the CART algorithm to perform decision tree training;

[0119] The constructing of the regression model specifically comprises using the fault cause pre-classification and identification model to perform fault cause pre-classification and identification according to the coding feature matrix set to obtain fault cause pre-identification data, and adopting a rule-based method to extract a repair solution according to the fault cause pre-identification data from the integrated repair generation optimization data set to obtain a fault repair candidate list, and performing regression model training according to the fault repair candidate list to obtain a repair effect prediction regression model;

[0120] The regression model training is specifically carried out using a linear regression algorithm;

[0121] The repair effect prediction regression model is used to predict the effect of the fault repair solutions in the fault repair candidate list to obtain a repair effect prediction output;

[0122] The repair generation model training is specifically to perform repair generation model training through the data integration, the feature coding optimization, the construction of the decision tree model and the construction of the regression model to obtain the repair generation model Model RP ;

[0123] The repair solution generation is specifically to use the repair generation model Model RP , generating a repair plan based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, and obtaining distribution cabinet fault repair reference data through prediction output integration;

[0124] The distribution cabinet fault repair reference data specifically includes fault cause pre-identification data, fault repair solution prediction data and repair effect prediction data.

[0125] By performing the above operations, in order to address the technical problem that the existing intelligent repair methods for distribution cabinet faults are difficult to combine with the diagnosed faults to further provide intelligent support for the maintenance of the distribution cabinet, this solution creatively adopts a data-driven decision tree regression model method to generate repairs. By training the decision tree and regression model, which are responsible for identifying the cause of the fault, matching the fault repair plan, and predicting the fault repair effect, it provides full intelligent support for the maintenance process of the fault repair plan, and improves the availability and reliability of the intelligent fault detection system for the distribution cabinet.

[0126] Embodiment 6, this embodiment is based on the above embodiment, see Figure 1 and Figure 2The intelligent detection module is used to perform comprehensive intelligent detection in combination with the distribution cabinet fault diagnosis data and the distribution cabinet fault repair reference data. Specifically, the distribution cabinet fault diagnosis data is obtained, the distribution cabinet fault information is recorded, and the distribution cabinet fault repair reference data is obtained by applying the repair generation module. A repair plan is selected from the distribution cabinet fault repair reference data to perform intelligent detection and intelligent repair of the distribution cabinet fault, and a distribution cabinet fault intelligent detection reference data set is obtained. The fault data, repair plan data and the results of the repair plan in the distribution cabinet fault intelligent detection reference data set are manually recorded for system optimization and maintenance.

[0127] Embodiment 7, this embodiment is based on the above embodiment, see Figure 6 , Figure 6 A visualization diagram of the fault probability reference data corresponding to a 1s acquisition frequency drawn for the fault diagnosis module, where the horizontal coordinate is the time index in seconds, and the vertical coordinate is the change trend of the fault probability prediction value corresponding to the specific 1s acquisition frequency.

[0128] Embodiment 8, this embodiment is based on the above embodiment, see Figure 7 , Figure 7 A visualization diagram of the fault probability reference data corresponding to a 5s acquisition frequency drawn for the fault diagnosis module, where the horizontal coordinate is the time index in seconds, and the vertical coordinate is the change trend of the fault probability prediction value corresponding to the specific 5s acquisition frequency.

[0129] Embodiment 9, this embodiment is based on the above embodiment, see Figure 8 , Figure 8 A visualization diagram of the fault probability reference data corresponding to a 10s acquisition frequency drawn for the fault diagnosis module, where the horizontal coordinate is the time index in seconds, and the vertical coordinate is the change trend of the fault probability prediction value corresponding to the specific 10s acquisition frequency.

[0130] Embodiment 10, this embodiment is based on the above embodiment, see Fig. 9 , Fig. 9 A visualization diagram of the weighted integrated fault probability reference data drawn for the fault diagnosis module, where the horizontal coordinate is the time index in seconds and the vertical coordinate is the specific trend of the fault probability prediction value after weighted integration.

[0131] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process and method including a series of elements include not only those elements, but also other elements not explicitly listed, or also include elements inherent to such process and method.

[0132] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0133] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent fault detection system for a power distribution cabinet, characterized in that: It includes data collection module, enhanced processing module, fault diagnosis module, repair generation module and intelligent detection module; The data collection module is used to collect the original data set required for intelligent fault detection of the distribution cabinet, specifically by deploying a sensor cluster and performing data collection to obtain the original data of comprehensive detection of the distribution cabinet, and send the original data of comprehensive detection of the distribution cabinet to the enhanced processing module; The data collection specifically refers to group frequency difference data collection, including collecting data once every 1 second, collecting data once every 5 seconds and collecting data once every 10 seconds, and obtaining the original data of the comprehensive detection of the distribution cabinet through the group frequency difference data collection, and storing the original data of the comprehensive detection of the distribution cabinet in the local database; The enhanced processing module is used to optimize and enhance the original data, specifically by sequentially performing data cleaning and feature processing on the original data of the comprehensive detection of the distribution cabinet to obtain enhanced distribution cabinet detection data, and sending the enhanced distribution cabinet detection data to the fault diagnosis module and the repair generation module. The specific steps include: data cleaning, feature processing, data standardization and enhanced comprehensive processing; The fault diagnosis module is used to analyze and identify potential faults during the operation of the power distribution cabinet. Specifically, based on the enhanced power distribution cabinet detection data, a bidirectional long short-term memory model combined with frequency difference grouping noise optimization is used to perform fault diagnosis, obtain power distribution cabinet fault diagnosis data, and send the power distribution cabinet fault diagnosis data to the repair generation module and the intelligent detection module; The bidirectional long short-term memory model combined with frequency difference grouping noise optimization specifically includes a frequency difference set empirical mode decomposition operator, a frequency difference feature optimization operator, a bidirectional long short-term memory sub-model, a random forest classification sub-model and a parameter optimization operator; The repair generation module is used to identify the cause of the fault, generate a repair plan and predict the repair effect. Specifically, based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, a data-driven decision tree regression model method is used to perform repair generation, obtain distribution cabinet fault repair reference data, and send the distribution cabinet fault repair reference data to the intelligent detection module; The intelligent detection module is used to perform comprehensive intelligent detection in combination with the distribution cabinet fault diagnosis data and the distribution cabinet fault repair reference data. Specifically, through intelligent detection and intelligent repair of distribution cabinet faults, a distribution cabinet fault intelligent detection reference data set is obtained, and the fault data, repair plan data and repair plan results in the distribution cabinet fault intelligent detection reference data set are manually recorded for system optimization and maintenance.

2. The intelligent fault detection system for a power distribution cabinet according to claim 1, characterized in that: In the data collection module, the distribution cabinet comprehensively detects the original data, specifically including current data, temperature data, humidity data, vibration frequency data and historical maintenance and repair data; the historical maintenance and repair data specifically includes historical maintenance record data, historical fault repair plan record data and historical fault repair maintenance result data.

3. The intelligent fault detection system for a power distribution cabinet according to claim 2, characterized in that: In the fault diagnosis module, the step of using a bidirectional long short-term memory model combined with frequency difference grouping noise optimization to perform fault diagnosis and obtain distribution cabinet fault diagnosis data includes: constructing a frequency difference set empirical mode decomposition operator, constructing a frequency difference feature optimization operator, constructing a bidirectional long short-term memory sub-model, constructing a random forest classification sub-model, constructing a parameter optimization operator, fault diagnosis model training and fault diagnosis; The frequency difference set empirical mode decomposition operator is constructed to optimize the feature data in the enhanced distribution cabinet detection data and reduce the interference of noise on data quality, specifically using a set empirical mode decomposition adaptive noise algorithm to perform adaptive mode decomposition on the enhanced distribution cabinet detection data; The constructing of the frequency difference feature optimization operator specifically includes performing frequency difference feature optimization extraction on the frequency difference intrinsic feature data to obtain frequency difference feature data, and performing standard deviation and mean calculation on the frequency difference feature data to obtain an optimized frequency difference feature data set; The bidirectional long short-term memory sub-model is specifically constructed by constructing a standard bidirectional long short-term memory neural network, and performing deep extraction of time series features according to the acquisition frequency to obtain frequency difference hidden state output data. The calculation formula is: ； h t is the frequency difference hidden state output data, which is used to represent the feature output data after weighted integration according to different sampling frequency weights. a1 is the 1 second sampling frequency weight, a5 is the 5 second sampling frequency weight, and a 10 is the 10-second sampling frequency weight, h t,1 is the frequency difference hidden state output corresponding to the 1 second sampling frequency, h t,5 is the frequency difference hidden state output corresponding to the 5-second sampling frequency, h t,10 It is the frequency difference hidden state output corresponding to the 10-second sampling frequency; The construction of the random forest classification sub-model specifically includes respectively training the classifier of the standard random forest model according to the optimized frequency difference feature data sets corresponding to different sampling frequencies, and classifying the fault type by taking the frequency difference hidden state output data as the total input of the trained random forest classification sub-model to obtain the model classification output data; The constructing parameter optimization operator specifically adopts a tree-structured Bayesian optimization algorithm to optimize model parameters, obtains a model hyperparameter optimization combination, and applies the model hyperparameter optimization combination to fault diagnosis model training; The fault diagnosis model training is specifically to perform fault diagnosis model training through the frequency difference set empirical mode decomposition operator, the frequency difference feature optimization operator, the bidirectional long short-term memory sub-model, the random forest classification sub-model and the parameter optimization operator to obtain the fault diagnosis model Model IF ; The fault diagnosis is specifically performed using the fault diagnosis model Model IF , perform fault diagnosis based on the enhanced distribution cabinet detection data to obtain distribution cabinet fault diagnosis data.

4. The intelligent fault detection system for a power distribution cabinet according to claim 3 is characterized in that: The distribution cabinet fault diagnosis data specifically includes fault probability reference data, fault type reference data and fault diagnosis index reference data; The fault probability reference data is used to represent the quantitative confidence value of the fault, and the value range is [0,100%]; The fault type reference data is used to indicate identifiable fault categories, including overload fault, short circuit fault, poor contact fault, aging fault, overheating fault, abnormal moisture fault and insulation fault; The fault diagnosis index reference data is used to provide an identification basis to optimize the interpretability of fault diagnosis, and specifically includes fault occurrence time record prediction data, fault duration prediction data and fault impact location reference data.

5. The intelligent fault detection system for a power distribution cabinet according to claim 4, characterized in that: In the repair generation module, a data-driven decision tree regression model method is used to perform repair generation and obtain reference data for distribution cabinet fault repair. The specific steps include: data integration, feature coding optimization, building a decision tree model, building a regression model, repair generation model training and repair plan generation.

6. The intelligent fault detection system for a power distribution cabinet according to claim 5, characterized in that: The data integration specifically includes integrating the enhanced power distribution cabinet detection data and the power distribution cabinet fault diagnosis data to obtain an integrated repair generation data set, and performing a data cleaning operation on the integrated repair generation data set to obtain an integrated repair generation optimization data set; The feature encoding optimization specifically comprises performing a feature extraction operation on the integrated repair generation optimization data set to obtain a repair generation feature data set, and obtaining a coding feature matrix set by performing one-hot encoding on the data features in the repair generation feature data set; The repair generates a feature data set, specifically including fault type features, fault frequency features, historical repair solution features, repair method features and environmental condition auxiliary features; The construction of the decision tree model specifically includes using a decision tree algorithm to perform fault repair pre-classification training to obtain a fault cause pre-classification identification model; The constructing of the regression model specifically comprises using the fault cause pre-classification and identification model to perform fault cause pre-classification and identification according to the coding feature matrix set to obtain fault cause pre-identification data, and adopting a rule-based method to extract a repair solution according to the fault cause pre-identification data from the integrated repair generation optimization data set to obtain a fault repair candidate list, and performing regression model training according to the fault repair candidate list to obtain a repair effect prediction regression model; The repair effect prediction regression model is used to predict the effect of the fault repair solutions in the fault repair candidate list to obtain a repair effect prediction output; The repair generation model training is specifically to perform repair generation model training through the data integration, the feature coding optimization, the construction of the decision tree model and the construction of the regression model to obtain the repair generation model Model RP ; The repair solution generation is specifically to use the repair generation model Model RP , generating a repair plan based on the enhanced distribution cabinet detection data and the distribution cabinet fault diagnosis data, and obtaining distribution cabinet fault repair reference data through prediction output integration; The distribution cabinet fault repair reference data specifically includes fault cause pre-identification data, fault repair solution prediction data and repair effect prediction data.

7. The intelligent fault detection system for a power distribution cabinet according to claim 6, characterized in that: The intelligent detection module is used to perform comprehensive intelligent detection in combination with fault diagnosis results and fault repair references. Specifically, the fault diagnosis data of the distribution cabinet is obtained, the fault information of the distribution cabinet is recorded, and the repair generation module is applied to obtain the fault repair reference data of the distribution cabinet. A repair plan is selected from the fault repair reference data of the distribution cabinet to perform intelligent detection and intelligent repair of the distribution cabinet fault, and a reference data set for intelligent detection of distribution cabinet faults is obtained. The fault data, repair plan data and results of the repair plan in the reference data set for intelligent detection of distribution cabinet faults are manually recorded for system optimization and maintenance.

Images