A method for detecting faults in a distribution box
Through sensor data acquisition and high-dimensional feature extraction, combined with historical fault cases and loop topology data, the rapid and accurate detection and processing of distribution box faults is achieved, and the problem of insufficient fault detection and analysis capabilities in traditional methods is solved, and fault identification is improved accuracy and efficiency.
Patent Information
- Application Number
- CN202411312989.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-20
AI Technical Summary
The traditional distribution box fault detection methods have problems such as insufficient fault detection and analysis capabilities and limited fault feature extraction and identification capabilities, which makes it difficult to ensure the timeliness and accuracy of fault detection and positioning.
Data is collected through temperature and current sensors, multi-dimensional data preprocessing and high-dimensional feature extraction are carried out, fault feature matrix is constructed, differential analysis and pattern recognition is performed based on historical fault case data, fault correlation factor data is collected for probability evaluation, and loop topology data is used for Fault area location and cause analysis.
It realizes rapid and accurate detection and processing of distribution box faults, improves the accuracy and efficiency of fault identification, solves the shortcomings of traditional methods in fault analysis capabilities and feature extraction, and realizes the intelligent detection of distribution box faults. , precision and efficiency.
Smart Images

Figure CN118818202B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of distribution box fault detection, and in particular to a distribution box fault detection method. Background Art
[0002] With the continuous development of power systems and the continuous growth of electricity demand, the scale and complexity of distribution networks are also increasing. As a key device in the distribution network, the fault detection and detection of distribution boxes are crucial to ensure the safe and stable operation of the power system. However, traditional distribution box fault detection methods have the problems of insufficient fault detection and analysis capabilities and limited fault feature extraction and identification capabilities.
[0003] Insufficient fault detection and analysis capabilities: Traditional distribution box fault detection mainly relies on manual inspection and fault analysis, which has problems such as slow response speed and low analysis efficiency. When a fault occurs, a lot of time and manpower are required for on-site investigation and data collection, and the timeliness and accuracy of fault detection and location are difficult to guarantee.
[0004] Limited ability to extract and identify fault features: Distribution box faults are diverse and complex, and faults of different types and causes have different features. Traditional methods have limited ability to extract and identify fault features, making it difficult to accurately distinguish and identify different fault modes. Summary of the invention
[0005] Based on this, it is necessary to provide a distribution box fault detection method to solve at least one of the above technical problems.
[0006] To achieve the above object, a distribution box fault detection method comprises the following steps:
[0007] Step S1: collecting sensor data from the distribution box through temperature and current sensors and performing multi-dimensional data preprocessing to obtain a multi-dimensional time series data set;
[0008] Step S2: constructing a high-dimensional feature vector for the multi-dimensional time series data set to obtain a high-dimensional feature vector set; performing dynamic feature extraction on the high-dimensional feature vector set to obtain a multi-dimensional feature matrix;
[0009] Step S3: Obtain historical distribution box failure case data; perform difference analysis of different feature subsets on the multidimensional feature matrix to obtain a feature difference statistic matrix; extract fault feature fragments and reconstruct fault feature vectors based on the feature difference statistic matrix and historical distribution box failure case data to obtain a fault feature vector;
[0010] Step S4: performing case failure mode feature description analysis based on historical distribution box failure case data to obtain a failure mode feature description library; performing failure mode identification based on the failure feature vector and the failure mode feature description library to obtain a candidate failure mode ranking list; generating a failure similarity vector based on the candidate failure mode ranking list to obtain a failure similarity vector;
[0011] Step S5: Collect the data of fault-related factors of the distribution box to obtain a data set of fault-related factors; perform a fault mode probability assessment based on the historical distribution box fault case data and the data set of fault-related factors to obtain a calibrated fault mode probability vector; perform a fault probability correction on the calibrated fault mode probability vector based on the data set of fault-related factors to obtain fault probability distribution data;
[0012] Step S6: Sort the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list; obtain the distribution box loop topology data; perform fault mode loop association analysis according to the fault mode sorting list and the distribution box loop topology data to obtain a fault mode loop association table; locate the fault area according to the fault mode loop association table and the fault mode sorting list to obtain fault area location result data; perform fault cause analysis according to the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
[0013] The present invention constructs a high-quality, structured multi-dimensional time series data set by real-time acquisition and preprocessing of sensor data, laying a solid data foundation for subsequent fault feature extraction and pattern recognition, and realizing the transformation of fault detection from passive response to active perception. Through multi-level feature extraction, enhancement, fusion, selection and dimensionality reduction, representative fault features are extracted from the original data, and a structured multi-dimensional feature matrix is constructed, which provides a data foundation with rich information, high discrimination and appropriate dimensions for subsequent fault pattern recognition and positioning. Through operations such as time series pattern clustering, feature association analysis, difference analysis, weighted fusion and fault time window positioning, the effective extraction and integration of key fault features are realized, and a more accurate and effective fault feature expression is constructed, which lays a solid foundation for subsequent fault pattern recognition and positioning. By constructing a fault case library, performing fault pattern clustering analysis and feature description, and performing fault pattern recognition and similarity calculation, the effective use of historical fault cases is realized, providing a reference basis for the detection and positioning of current faults, and improving the accuracy and efficiency of fault recognition. By collecting fault correlation factor data, evaluating and correcting fault mode probabilities, and generating fault probability distribution, quantitative analysis of fault mode probabilities is achieved, providing a more reliable basis for subsequent fault location and cause analysis. Through the steps of fault mode loop correlation analysis, candidate loop screening, loop fault probability update, fault area location, and fault cause analysis, the leap from fault mode identification to fault area location is achieved, providing an effective technical means for the rapid detection and processing of distribution box faults. Therefore, the present invention provides a distribution box fault detection method to solve the problems of insufficient fault detection and analysis capabilities and limited fault feature extraction and recognition capabilities in traditional distribution box fault detection methods. By collecting multi-dimensional time series data and introducing fault correlation factors, a more comprehensive fault information library is constructed to provide a richer data basis for fault analysis. The method adopts multi-level feature extraction and fusion technology, and combines the fault mode feature description library and loop topology data correlation analysis to achieve accurate identification of multiple fault modes and accurate positioning of fault areas, effectively solving the deficiencies of traditional methods in fault analysis capabilities, feature extraction and recognition capabilities, and realizing intelligent, precise and efficient distribution box fault detection.
[0014] Preferably, step S1 comprises the following steps:
[0015] Step S11: collecting sensor data from the distribution box through temperature and current sensors to obtain sensor data;
[0016] Step S12: Cleaning the sensor data to obtain sensor cleaning data;
[0017] Step S13: performing time-space alignment on the sensor cleaning data to obtain time-space alignment data;
[0018] Step S14: performing data integration on the spatiotemporal aligned data to obtain a multi-dimensional time series data set.
[0019] The present invention deploys sensors at key nodes of the distribution box to collect temperature and current data in real time, providing the most direct data basis for subsequent fault detection, and realizing the transformation of fault detection from passive response to active perception. Through data cleaning operations such as outlier detection and missing value interpolation, the noise and errors in the sensor data are effectively removed, the data quality is improved, and reliable data guarantee is provided for subsequent feature extraction and analysis. Through time synchronization and spatial alignment, the time deviation and position difference between different sensors are eliminated, the uniformity of data in time and space is ensured, and an accurate reference basis is provided for subsequent multidimensional feature extraction and fault pattern recognition. The data is divided and organized according to the preset time window to construct a structured multidimensional time series data set, which provides a more convenient and efficient data processing format for subsequent feature engineering and pattern recognition, and improves the efficiency of the algorithm.
[0020] Preferably, step S2 comprises the following steps:
[0021] Step S21: performing time series segmentation on the multidimensional time series data set to obtain a time series segment set; performing preliminary extraction of univariate features of temperature and current data on each time segment in the time series segment set to obtain a univariate feature set;
[0022] Step S22: performing time series feature enhancement on the time series segment set according to the univariate feature set to obtain an enhanced time domain feature set; performing fast Fourier transform on each time segment in the enhanced time domain feature set to obtain a frequency domain feature set;
[0023] Step S23: constructing a high-dimensional feature vector for the univariate feature set, the enhanced time domain feature set, and the frequency domain feature set to obtain a high-dimensional feature vector set;
[0024] Step S24: performing feature correlation analysis on the high-dimensional feature vector set to obtain a feature correlation matrix; sorting the feature importance of the high-dimensional feature vector set according to the feature correlation matrix to obtain a key feature sorting set;
[0025] Step S25: perform feature dimensionality reduction on the key feature sorting set to obtain a reduced-dimensional feature set; perform feature time series reconstruction on the reduced-dimensional feature set to obtain a fault feature time series; and perform multidimensional feature matrix generation on the fault feature time series to obtain a multidimensional feature matrix.
[0026] The present invention converts the original multi-dimensional time series data into a feature form that is easier to analyze through time series segmentation and preliminary extraction of univariate features, and preliminarily extracts the basic statistical features of temperature and current signals, laying the foundation for subsequent feature enhancement and pattern recognition. Through time series feature enhancement, the contextual information of the time series is integrated into the feature expression, which improves the sensitivity of the feature to the time trend. At the same time, the frequency domain features are extracted by fast Fourier transform, the signal features are revealed from the frequency perspective, the diversity of features is enriched, and more comprehensive information is provided for fault pattern recognition. The univariate features, enhanced time domain features and frequency domain features are fused into high-dimensional feature vectors, which realizes the effective integration of multi-faceted information, constructs a more comprehensive and expressive feature representation, and provides richer information for subsequent feature selection and fault recognition. Through feature correlation analysis and importance sorting, the key features related to the fault are identified, redundant information and noise interference are eliminated, the feature dimension is reduced, the calculation efficiency is improved, and more targeted feature input is provided for subsequent fault pattern recognition. Through feature dimensionality reduction, the feature dimension is further reduced, the calculation efficiency is improved, and key information is retained. The reduced-dimensional features are reconstructed into time series and a multi-dimensional feature matrix is generated, providing a more concise and efficient data structure for subsequent fault mode identification and location.
[0027] Preferably, step S3 comprises the following steps:
[0028] Step S31: performing time series pattern clustering analysis on the time series in the multidimensional feature matrix to obtain a time series pattern cluster set; performing intra-cluster feature association analysis on the time series pattern cluster set to obtain an intra-cluster feature association matrix set;
[0029] Step S32: performing key feature subset identification according to the intra-cluster feature association matrix set to obtain the key feature subset; performing difference analysis on different feature subsets in the key feature subset to obtain a feature difference statistics matrix;
[0030] Step S33: Calculate feature weights according to the feature difference statistics matrix to obtain a feature weight vector; perform multi-feature weighted fusion on the key feature subset according to the feature weight vector to obtain a fused feature time series;
[0031] Step S34: Acquire historical distribution box failure case data; locate the failure time window according to the distribution box historical failure data in the historical distribution box failure case data to obtain failure time window data;
[0032] Step S35: extracting fault feature segments from the fused feature time series according to the fault time window data to obtain fault feature segment data; reconstructing fault feature vectors from the fault feature segment data to obtain fault feature vectors.
[0033] The present invention groups data with similar time dynamic features through time series pattern cluster analysis, which is helpful to identify potential fault modes. Intra-cluster feature association analysis reveals the relationship between different features under the same fault mode, providing a basis for subsequent feature selection and fusion. By analyzing the feature association matrix, the key feature subsets that contribute the most to distinguishing different fault modes are identified, avoiding redundant information and computational burden caused by using all features. Feature subset difference analysis quantifies the ability of different feature subsets to distinguish fault modes, providing a basis for feature weight calculation. Different feature subsets are given different weights according to feature differences, highlighting the role of important features, reducing the influence of noise and irrelevant features, and improving the accuracy of fault identification. The weighted fused feature time series integrates the information of multiple key features and can more comprehensively characterize the fault features. Using historical fault case data, the time window of the fault occurrence is located, and the extraction range of the fault features is narrowed to a period of time before and after the fault occurs, avoiding the interference of irrelevant time period data, and improving the efficiency and accuracy of feature extraction. The fragment corresponding to the fault time window is extracted from the fused feature time series, retaining the key feature information when the fault occurs. The feature fragments are reconstructed into fault feature vectors, providing structured data input for subsequent fault mode recognition and location.
[0034] Preferably, step S4 comprises the following steps:
[0035] Step S41: extract features from historical distribution box fault case data and perform correlation analysis to obtain a case distribution box fault feature vector; construct a preliminary fault case library for the case distribution box fault feature vector to obtain a preliminary fault case library;
[0036] Step S42: performing a fault case cluster analysis on the fault feature vectors of the distribution box in the preliminary fault case library to obtain a fault mode cluster;
[0037] Step S43: performing a fault mode feature description on the fault mode cluster to obtain a fault mode feature description library;
[0038] Step S44: performing fault mode identification according to the fault feature vector and the fault mode feature description library to obtain a candidate fault mode ranking list;
[0039] Step S45: perform similarity confidence calculation based on the candidate fault mode ranking list to obtain a fault mode confidence vector; generate a fault similarity vector for the candidate fault mode ranking list and the fault mode confidence vector to obtain a fault similarity vector.
[0040] The present invention constructs a case distribution box fault feature vector by performing feature extraction and association analysis on historical fault case data, converts historical fault cases into structured data for machine learning, and lays a data foundation for subsequent fault mode analysis and identification. Through fault case cluster analysis, fault cases with similar features are classified together to form different fault mode clusters, which is helpful to identify and distinguish different fault types and provide a basis for fault mode feature description. Each fault mode cluster is characterized, its common features are extracted, and a fault mode feature description library is constructed to provide a reference standard for subsequent fault mode identification. The current fault feature vector is compared and analyzed with the fault mode feature description library to identify several fault modes that are most similar to the current fault, and they are sorted to provide candidate solutions for fault detection. By calculating the fault similarity vector, the similarity between the current fault and each candidate fault mode is quantified, providing more refined reference information for subsequent fault probability assessment and positioning.
[0041] Preferably, step S44 includes the following steps:
[0042] Step S441: performing fault mode similarity calculation on the fault feature vector and the fault mode feature description library to obtain a fault mode similarity matrix;
[0043] Step S442: performing fault mode recognition weight allocation on the fault mode similarity matrix to obtain fault mode recognition weight data;
[0044] Step S443: performing similarity weighted fusion on the fault mode similarity matrix according to the fault mode recognition weight data to obtain a weighted fault mode similarity matrix;
[0045] Step S444: screening candidate fault modes of the weighted fault mode similarity matrix using a preset fault mode similarity threshold to obtain a candidate fault mode set;
[0046] Step S445: sorting the candidate fault mode set to obtain a candidate fault mode sorting list.
[0047] The present invention quantifies the matching degree between the current fault and the existing fault mode by calculating the similarity between the fault feature vector and each mode in the fault mode feature description library, and provides basic data for subsequent fault mode identification. According to the prior knowledge or historical data of the fault mode, different fault modes are given different identification weights, which can highlight important fault modes, reduce the risk of misjudgment, and improve the accuracy and reliability of fault identification. The similarity calculation result is integrated with the fault mode identification weight, and the data drive and prior knowledge are comprehensively considered, so that the fault mode identification result is more accurate and reasonable. By setting the similarity threshold, the fault mode with high similarity to the current fault feature vector is screened out, and the mode irrelevant to the current fault is excluded, the scope of fault investigation is narrowed, and the efficiency of fault identification is improved. The candidate fault modes are sorted according to the similarity score, and the fault modes with higher probability are arranged in front, which provides a priority reference for subsequent fault probability evaluation and positioning, and is convenient for quickly locking the cause of the fault.
[0048] Preferably, step S5 comprises the following steps:
[0049] Step S51: collecting fault-related factor data of the distribution box to obtain a fault-related factor data set, wherein the fault-related factor data set includes load rate data and equipment status data;
[0050] Step S52: performing a failure mode probability assessment based on historical distribution box failure case data and a failure correlation factor data set to obtain a calibrated failure mode probability vector;
[0051] Step S53: performing load factor probability correction on the calibration fault mode probability vector according to the load rate data to obtain a load corrected fault mode probability vector;
[0052] Step S54: performing device state probability correction on the load correction fault mode probability vector according to the device state data to obtain a device state correction fault mode probability vector;
[0053] Step S55: performing probability normalization processing on the equipment state correction fault mode probability vector to obtain a normalized fault mode probability vector; generating a fault probability distribution according to the normalized fault mode probability vector to obtain fault probability distribution data.
[0054] The present invention can more comprehensively characterize the background environment of the fault occurrence by collecting the load rate and equipment status data of the distribution box in real time and incorporating this information into the scope of fault analysis, thereby providing a richer reference basis for the subsequent fault mode probability assessment. By using historical fault case data and fault-related factor data sets to build a probability assessment model and calibrating the model output results, the probability of occurrence of each candidate fault mode can be more accurately assessed, thereby improving the reliability of fault identification. Taking into account the impact of the load rate on certain fault modes, the probability of the fault mode is corrected according to the current load rate level, which can more accurately reflect the possibility of fault occurrence under different load conditions and improve the pertinence of fault identification. By further correcting the fault mode probability in combination with the equipment status data, the fault mode related to the equipment status can be more accurately identified, thereby improving the accuracy of fault identification. Through probability normalization processing, the probability values of all fault modes are mapped to the [0,1] interval, and their sum is guaranteed to be 1, making the probability distribution more intuitive and easy to understand.
[0055] Preferably, step S52 includes the following steps:
[0056] Step S521: performing Bayesian network structure learning based on historical distribution box failure case data to obtain a Bayesian network structure diagram;
[0057] Step S522: using the fault correlation factor data set to perform network parameter learning on the Bayesian network structure diagram to obtain a parameterized Bayesian network;
[0058] Step S523: inputting the fault similarity vector into the parameterized Bayesian network to update the node state probability, thereby obtaining an updated Bayesian network;
[0059] Step S524: performing network reasoning and probability propagation calculation on the updated Bayesian network to obtain a posterior probability vector of the fault mode;
[0060] Step S525: Perform a posteriori probability calibration on the fault mode posterior probability vector to obtain a calibrated fault mode probability vector.
[0061] The present invention analyzes historical fault case data, learns the association between variables such as fault mode, load rate, and equipment status, and constructs a Bayesian network structure diagram, which provides a clear network expression framework for fault mode probability reasoning and improves the interpretability of the model. The fault correlation factor data set is used to perform parameter learning on the Bayesian network structure diagram, and the probability relationship between variables is quantified, so that the Bayesian network can perform quantitative probability reasoning, providing a more accurate basis for fault mode probability evaluation. The fault similarity vector is input into the Bayesian network as evidence, and the node state probability related to the fault mode is updated, so as to achieve the purpose of integrating the current fault feature information into the probability reasoning process, so that the probability evaluation result is more in line with the current fault situation. Based on the updated Bayesian network, network reasoning and probability propagation calculation are performed to obtain the posterior probability of each fault mode, which makes full use of prior knowledge and current fault information, and improves the accuracy and reliability of fault mode probability evaluation. Calibration of the posterior probability vector can effectively correct the influence of model deviation and data noise, and further improve the accuracy and reliability of fault mode probability evaluation.
[0062] Preferably, step S6 comprises the following steps:
[0063] Step S61: sorting the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list;
[0064] Step S62: acquiring the distribution box circuit topology data; performing a fault mode circuit association analysis according to the fault mode sorting list and the distribution box circuit topology data to obtain a fault mode circuit association table;
[0065] Step S63: Screen candidate circuits according to the fault mode circuit association table and the fault mode sorting list to obtain a candidate fault circuit set; calculate circuit characteristic indexes for the candidate fault circuit set to obtain a circuit characteristic index variation matrix;
[0066] Step S64: updating the loop fault probability according to the loop characteristic index variation matrix to obtain an updated loop fault probability vector; performing loop topology analysis on the distribution box loop topology data to obtain a loop topology relationship diagram;
[0067] Step S65: locating the fault area according to the updated loop fault probability vector and the loop topology relationship diagram to obtain fault area locating result data;
[0068] Step S66: Perform fault cause analysis based on the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
[0069] The present invention reorders the candidate fault modes by combining the fault probability distribution data, and puts the fault modes with higher probability in front, providing more accurate guidance for subsequent fault location, and improving the efficiency of fault troubleshooting. The distribution box loop topology data is used to analyze the impact range of different fault modes on each loop, establish the association relationship between the fault mode and the loop, provide necessary topology information for fault area location, and narrow the scope of fault troubleshooting. According to the fault mode loop association table and the fault mode sorting list, the candidate fault loops related to the high probability fault mode are screened out, and the change amount of their characteristic indicators is calculated to provide data support for the subsequent fault probability update, further improving the accuracy of fault location. According to the change amount of the loop characteristic indicator, the fault probability of the candidate fault loop is updated, and the loop topology relationship is visualized to provide more refined probability information and more intuitive topology display for fault area location. Combined with the updated loop fault probability and loop topology relationship diagram, the fault propagation model or graph search algorithm is used to accurately locate the fault area, identify the faulty loop or equipment, and provide direct guidance for fault repair. According to the fault area location results and fault mode characteristic description, the cause of the fault is analyzed and a fault cause analysis report is generated to provide a reference basis for fault handling and achieve rapid detection and handling of distribution box faults.
[0070] Preferably, step S62 includes the following steps:
[0071] Step S621: define component failure modes for the failure mode sorting list to obtain a component failure mode library;
[0072] Step S622: acquiring the distribution box circuit topology data; performing a fault mode propagation path analysis according to the component fault mode library and the distribution box circuit topology data to obtain a fault mode propagation path diagram;
[0073] Step S623: identifying loop sensitive components according to the fault mode propagation path diagram to obtain a fault mode sensitive component table;
[0074] Step S624: constructing a fault mode component impact feature vector for the fault mode sorting list according to the fault mode sensitive component table to obtain a fault mode component impact feature vector;
[0075] Step S625: constructing loop element feature vectors according to the distribution box loop topology data to obtain a loop element feature vector table;
[0076] Step S626: performing fault mode-loop correlation calculation on the fault mode component influence feature vector and the loop component feature vector table to obtain a fault mode-loop correlation matrix;
[0077] Step S627: setting a correlation threshold according to the fault mode-loop correlation matrix to obtain a correlation threshold; screening the fault mode-loop correlation matrix for fault mode correlation loops according to the correlation threshold to obtain a fault mode loop correlation table.
[0078] The present invention establishes a mapping relationship between the fault mode and the component fault by associating the abstract fault mode with the specific component fault, and provides a basis for the subsequent fault propagation path analysis. Combined with the component fault mode library and the distribution box loop topology data, the propagation path of the fault mode in the distribution box loop is analyzed, the fault impact range is intuitively displayed, and a basis is provided for the identification of loop sensitive components. By analyzing the fault mode propagation path diagram, the key components affected by the fault are identified, and the corresponding relationship between the fault mode and the sensitive component is established, which provides a more refined target for fault location. The influence of the fault mode on different components is quantified, and the fault mode component influence feature vector is constructed, which provides a data basis for the subsequent calculation of the correlation between the fault mode and the loop. The loop topology information is converted into the loop component feature vector, which provides another necessary data basis for calculating the correlation between the fault mode and the loop. By calculating the correlation between the fault mode and the loop, the influence of different fault modes on each loop is quantified, which provides a basis for the screening of the fault mode associated loop. By setting the correlation threshold, the loop with a high correlation with the fault mode is screened out, and a fault mode loop association table is established, which narrows the target range for fault area positioning and improves the efficiency of fault location.
[0079] The present invention constructs a high-quality, structured multi-dimensional time series data set by real-time acquisition and preprocessing of sensor data, laying a solid data foundation for subsequent fault feature extraction and pattern recognition, and realizing the transformation of fault detection from passive response to active perception. Through multi-level feature extraction, enhancement, fusion, selection and dimensionality reduction, representative fault features are extracted from the original data, and a structured multi-dimensional feature matrix is constructed, which provides a data foundation with rich information, high discrimination and appropriate dimensions for subsequent fault pattern recognition and positioning. Through operations such as time series pattern clustering, feature association analysis, difference analysis, weighted fusion and fault time window positioning, the effective extraction and integration of key fault features are realized, and a more accurate and effective fault feature expression is constructed, which lays a solid foundation for subsequent fault pattern recognition and positioning. By constructing a fault case library, performing fault pattern clustering analysis and feature description, and performing fault pattern recognition and similarity calculation, the effective use of historical fault cases is realized, providing a reference basis for the detection and positioning of current faults, and improving the accuracy and efficiency of fault recognition. By collecting fault correlation factor data, evaluating and correcting fault mode probabilities, and generating fault probability distribution, quantitative analysis of fault mode probabilities is achieved, providing a more reliable basis for subsequent fault location and cause analysis. Through the steps of fault mode loop correlation analysis, candidate loop screening, loop fault probability update, fault area location, and fault cause analysis, the leap from fault mode identification to fault area location is achieved, providing an effective technical means for the rapid detection and processing of distribution box faults. Therefore, the present invention provides a distribution box fault detection method to solve the problems of insufficient fault detection and analysis capabilities and limited fault feature extraction and recognition capabilities in traditional distribution box fault detection methods. By collecting multi-dimensional time series data and introducing fault correlation factors, a more comprehensive fault information library is constructed to provide a richer data basis for fault analysis. The method adopts multi-level feature extraction and fusion technology, and combines the fault mode feature description library and loop topology data correlation analysis to achieve accurate identification of multiple fault modes and accurate positioning of fault areas, effectively solving the deficiencies of traditional methods in fault analysis capabilities, feature extraction and recognition capabilities, and realizing intelligent, precise and efficient distribution box fault detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0080] Figure 1 A schematic diagram of the steps of a method for detecting a fault in a distribution box;
[0081] Figure 2 for Figure 1 Detailed implementation steps of step S4 in FIG.
[0082] Figure 3 for Figure 1 Detailed implementation steps of step S6 in FIG.
[0083] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION
[0084] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.
[0085] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.
[0086] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.
[0087] To achieve this, please refer to Figures 1 to 3 , a distribution box fault detection method, comprising the following steps:
[0088] Step S1: collecting sensor data from the distribution box through temperature and current sensors and performing multi-dimensional data preprocessing to obtain a multi-dimensional time series data set;
[0089] Step S2: constructing a high-dimensional feature vector for the multi-dimensional time series data set to obtain a high-dimensional feature vector set; performing dynamic feature extraction on the high-dimensional feature vector set to obtain a multi-dimensional feature matrix;
[0090] Step S3: Obtain historical distribution box failure case data; perform difference analysis of different feature subsets on the multidimensional feature matrix to obtain a feature difference statistic matrix; extract fault feature fragments and reconstruct fault feature vectors based on the feature difference statistic matrix and historical distribution box failure case data to obtain a fault feature vector;
[0091] Step S4: performing case failure mode feature description analysis based on historical distribution box failure case data to obtain a failure mode feature description library; performing failure mode identification based on the failure feature vector and the failure mode feature description library to obtain a candidate failure mode ranking list; generating a failure similarity vector based on the candidate failure mode ranking list to obtain a failure similarity vector;
[0092] Step S5: Collect the data of fault-related factors of the distribution box to obtain a data set of fault-related factors; perform a fault mode probability assessment based on the historical distribution box fault case data and the data set of fault-related factors to obtain a calibrated fault mode probability vector; perform a fault probability correction on the calibrated fault mode probability vector based on the data set of fault-related factors to obtain fault probability distribution data;
[0093] Step S6: Sort the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list; obtain the distribution box loop topology data; perform fault mode loop association analysis according to the fault mode sorting list and the distribution box loop topology data to obtain a fault mode loop association table; locate the fault area according to the fault mode loop association table and the fault mode sorting list to obtain fault area location result data; perform fault cause analysis according to the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
[0094] In the embodiment of the present invention, reference Figure 1 The above is a schematic diagram of the steps of the distribution box fault detection method of the present invention. In this example, the distribution box fault detection method includes the following steps:
[0095] Step S1: collecting sensor data from the distribution box through temperature and current sensors and performing multi-dimensional data preprocessing to obtain a multi-dimensional time series data set;
[0096] Step S2: constructing a high-dimensional feature vector for the multi-dimensional time series data set to obtain a high-dimensional feature vector set; performing dynamic feature extraction on the high-dimensional feature vector set to obtain a multi-dimensional feature matrix;
[0097] Step S3: Obtain historical distribution box failure case data; perform difference analysis of different feature subsets on the multidimensional feature matrix to obtain a feature difference statistic matrix; extract fault feature fragments and reconstruct fault feature vectors based on the feature difference statistic matrix and historical distribution box failure case data to obtain a fault feature vector;
[0098] Step S4: performing case failure mode feature description analysis based on historical distribution box failure case data to obtain a failure mode feature description library; performing failure mode identification based on the failure feature vector and the failure mode feature description library to obtain a candidate failure mode ranking list; generating a failure similarity vector based on the candidate failure mode ranking list to obtain a failure similarity vector;
[0099] Step S5: Collect the data of fault-related factors of the distribution box to obtain a data set of fault-related factors; perform a fault mode probability assessment based on the historical distribution box fault case data and the data set of fault-related factors to obtain a calibrated fault mode probability vector; perform a fault probability correction on the calibrated fault mode probability vector based on the data set of fault-related factors to obtain fault probability distribution data;
[0100] Step S6: Sort the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list; obtain the distribution box loop topology data; perform fault mode loop association analysis according to the fault mode sorting list and the distribution box loop topology data to obtain a fault mode loop association table; locate the fault area according to the fault mode loop association table and the fault mode sorting list to obtain fault area location result data; perform fault cause analysis according to the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
[0101] In an embodiment of the present invention, temperature and current sensors are installed at key nodes of the distribution box to collect data in real time and transmit it to the data processing center. The raw data is cleaned, aligned in time and space, and organized to finally obtain a distribution box operation data set containing a multidimensional time series. The multidimensional time series data set is divided into multiple segments according to the time window. The statistical features of temperature and current are extracted for each segment, and the time series feature enhancement and fast Fourier transform are performed to obtain the frequency domain features. The three features are combined to construct a high-dimensional feature vector set, and then feature correlation analysis and dimensionality reduction processing are performed to finally obtain a multidimensional feature matrix. The multidimensional feature matrix is clustered for time series patterns, and feature association analysis is performed for each cluster to identify key feature subsets. The feature difference statistic matrix is obtained by feature subset difference analysis, and the feature weight vector is calculated, and finally weighted fusion is performed to obtain a fused feature time series. The fault time window is located in combination with historical fault case data, and the fault feature fragments are extracted and reconstructed to obtain the fault feature vector. Feature extraction and association analysis are performed on the historical fault case data to construct a preliminary fault case library, and cluster analysis is performed to obtain a fault mode cluster set. Each cluster is characterized to form a fault mode feature description library. The fault feature vector and the fault mode feature description library are used to perform similarity calculation, weight assignment, similarity weighted fusion and threshold screening, and finally a candidate fault mode ranking list is obtained, and a fault similarity vector is calculated. The load rate and equipment status data of the distribution box are collected in real time to form a fault correlation factor data set. The fault mode probability assessment model is constructed using the historical fault case data and the fault correlation factor data set, and the candidate fault mode is probability assessed to obtain a calibrated fault mode probability vector. The probability vector is then corrected according to the load rate and equipment status data to finally obtain the fault probability distribution data. The candidate fault mode ranking list is reordered according to the fault probability distribution data. The distribution box loop topology data is obtained, and the component fault mode library is defined. The fault mode loop association table is obtained through fault mode propagation path analysis, loop sensitive component identification, feature vector construction and correlation calculation. Combined with the updated loop fault probability and loop topology relationship diagram, the fault area is located, and finally the fault cause analysis is performed based on the positioning results and the fault mode feature description library to generate a distribution box fault cause analysis report.
[0102] Preferably, step S1 comprises the following steps:
[0103] Step S11: collecting sensor data from the distribution box through temperature and current sensors to obtain sensor data;
[0104] Step S12: Cleaning the sensor data to obtain sensor cleaning data;
[0105] Step S13: performing time-space alignment on the sensor cleaning data to obtain time-space alignment data;
[0106] Step S14: performing data integration on the spatiotemporal aligned data to obtain a multi-dimensional time series data set.
[0107] In an embodiment of the present invention, temperature and current sensors are installed at key nodes of the distribution box, such as at the incoming line loop, the outgoing line loop, the busbar, etc. The sensor collects the temperature and current data of the distribution box in real time according to a preset sampling frequency (for example, once a minute), and transmits the collected data to the data processing center to form the original sensor data. The original sensor data is detected and processed for abnormal values, for example, the temperature and current data that are obviously deviated from the normal range are eliminated by using the 3σ criterion, and the missing values are interpolated, for example, by using linear interpolation or spline interpolation. The purpose of data cleaning is to eliminate abnormal data caused by factors such as sensor failure and data transmission errors, and to ensure the accuracy and reliability of the data. Since there may be slight deviations in the data acquisition time of each sensor, it is necessary to perform time synchronization processing on the sensor cleaning data, for example, by using timestamp alignment or linear interpolation methods. At the same time, it is also necessary to spatially align the data according to the installation location information of the sensor, and associate the temperature and current data at different locations with the corresponding distribution box components. The spatiotemporal aligned data is divided according to the preset time window (for example, 1 hour), and the temperature and current data in each time window are organized into a data segment. Each data segment contains the temperature and current information of each component of the distribution box within the same time period, forming a multi-dimensional time series data set, which provides basic data for subsequent feature extraction and fault analysis.
[0108] Preferably, step S2 comprises the following steps:
[0109] Step S21: performing time series segmentation on the multidimensional time series data set to obtain a time series segment set; performing preliminary extraction of univariate features of temperature and current data on each time segment in the time series segment set to obtain a univariate feature set;
[0110] Step S22: performing time series feature enhancement on the time series segment set according to the univariate feature set to obtain an enhanced time domain feature set; performing fast Fourier transform on each time segment in the enhanced time domain feature set to obtain a frequency domain feature set;
[0111] Step S23: constructing a high-dimensional feature vector for the univariate feature set, the enhanced time domain feature set, and the frequency domain feature set to obtain a high-dimensional feature vector set;
[0112] Step S24: performing feature correlation analysis on the high-dimensional feature vector set to obtain a feature correlation matrix; sorting the feature importance of the high-dimensional feature vector set according to the feature correlation matrix to obtain a key feature sorting set;
[0113] Step S25: perform feature dimensionality reduction on the key feature sorting set to obtain a reduced-dimensional feature set; perform feature time series reconstruction on the reduced-dimensional feature set to obtain a fault feature time series; and perform multidimensional feature matrix generation on the fault feature time series to obtain a multidimensional feature matrix.
[0114] In an embodiment of the present invention, a multidimensional time series data set is segmented according to a preset time window length (e.g., 10 minutes) to obtain multiple time segments, each of which contains temperature and current data within the time window. Then, statistical features of temperature and current, such as maximum value, minimum value, average value, standard deviation, peak-to-peak value, root mean square value, etc., are extracted for each time segment to form a univariate feature set. Using a sliding window method, the univariate features of the current time segment are combined with the univariate features of the previous several time segments. For example, the features of the current segment and the previous two segments can be spliced together to form a new feature vector, thereby enhancing the information expression capability of the time series and obtaining an enhanced time domain feature set. Subsequently, a fast Fourier transform (FFT) is performed on each time segment in the enhanced time domain feature set to extract its frequency domain features, such as the amplitude and phase of each frequency component, to form a frequency domain feature set. The univariate feature set, enhanced time domain feature set, and frequency domain feature set of each time segment are merged into a high-dimensional feature vector to represent the comprehensive features of the time segment. The high-dimensional feature vectors of all time segments constitute a high-dimensional feature vector set. Calculate the correlation coefficients between each feature in the high-dimensional feature vector set and construct a feature correlation matrix. Then, according to the feature correlation matrix, for example, use principal component analysis (PCA) or feature importance ranking algorithm to sort the features, screen out key features with high contribution to fault identification, and form a key feature ranking set. Perform dimensionality reduction processing on the key feature ranking set, such as using PCA or linear discriminant analysis (LDA) to reduce the feature dimension, reduce the amount of calculation, and eliminate redundant information between features to obtain a reduced-dimensional feature set. Then, arrange the reduced-dimensional feature set corresponding to each time segment in chronological order to reconstruct the fault feature time series. Finally, arrange the fault feature time series of all time segments in order to generate a multidimensional feature matrix for subsequent fault mode recognition and location.
[0115] Preferably, step S3 comprises the following steps:
[0116] Step S31: performing time series pattern clustering analysis on the time series in the multidimensional feature matrix to obtain a time series pattern cluster set; performing intra-cluster feature association analysis on the time series pattern cluster set to obtain an intra-cluster feature association matrix set;
[0117] Step S32: performing key feature subset identification according to the intra-cluster feature association matrix set to obtain the key feature subset; performing difference analysis on different feature subsets in the key feature subset to obtain a feature difference statistics matrix;
[0118] Step S33: Calculate feature weights according to the feature difference statistics matrix to obtain a feature weight vector; perform multi-feature weighted fusion on the key feature subset according to the feature weight vector to obtain a fused feature time series;
[0119] Step S34: Acquire historical distribution box failure case data; locate the failure time window according to the distribution box historical failure data in the historical distribution box failure case data to obtain failure time window data;
[0120] Step S35: extracting fault feature segments from the fused feature time series according to the fault time window data to obtain fault feature segment data; reconstructing fault feature vectors from the fault feature segment data to obtain fault feature vectors.
[0121] In an embodiment of the present invention, a cluster analysis is performed on the time series in the multidimensional feature matrix, for example, a dynamic time warping (DTW) distance or a model-based clustering method is used to divide the time series with similar change trends into the same cluster to form a time series pattern cluster. Then, the features in each cluster are subjected to association analysis, for example, the Pearson correlation coefficient or the Spearman correlation coefficient between the features is calculated, and an intra-cluster feature association matrix is constructed to characterize the mutual relationship between the features in the cluster. According to the intra-cluster feature association matrix, a feature subset with strong association and representativeness in each cluster is identified, for example, a feature combination with an absolute value of the association coefficient exceeding a certain threshold can be selected to form a key feature subset. Then, a difference analysis is performed on different feature subsets, for example, the Euclidean distance or the Mahalanobis distance between the feature subset means is calculated, and a feature difference statistic matrix is constructed to characterize the degree of difference between different feature subsets. According to the feature difference statistic matrix, for example, an entropy weight method or a distance-based weight calculation method is used to calculate the weight of each feature subset to form a feature weight vector. Then, the key feature subsets are weighted fused according to the feature weight vector, for example, the feature values of each feature subset are multiplied by the corresponding weight and then summed to obtain a fused feature time series, which is used to comprehensively characterize the information of different feature subsets. The historical distribution box fault case data is obtained from the database or file, including the time of the fault, the fault type, the fault location and other information. According to the fault occurrence time recorded in the historical fault data, the corresponding fault time window is located in the fused feature time series, and the feature data in the time window is extracted to form the fault time window data. According to the fault time window data, the time series segment corresponding to the fault time window is extracted from the fused feature time series to form the fault feature segment data. Then, the fault feature segment data is feature reconstructed, for example, the time series segment is converted into a feature vector, and statistical features (mean, variance, etc.) or other feature extraction methods can be used to form a fault feature vector, which is used to characterize the feature information at the time of the fault.
[0122] Preferably, step S4 comprises the following steps:
[0123] Step S41: extract features from historical distribution box fault case data and perform correlation analysis to obtain a case distribution box fault feature vector; construct a preliminary fault case library for the case distribution box fault feature vector to obtain a preliminary fault case library;
[0124] Step S42: performing a fault case cluster analysis on the fault feature vectors of the distribution box in the preliminary fault case library to obtain a fault mode cluster;
[0125] Step S43: performing a fault mode feature description on the fault mode cluster to obtain a fault mode feature description library;
[0126] Step S44: performing fault mode identification according to the fault feature vector and the fault mode feature description library to obtain a candidate fault mode ranking list;
[0127] Step S45: perform similarity confidence calculation based on the candidate fault mode ranking list to obtain a fault mode confidence vector; generate a fault similarity vector for the candidate fault mode ranking list and the fault mode confidence vector to obtain a fault similarity vector.
[0128] As an example of the present invention, refer to Figure 2 As shown, in this example, step S4 includes:
[0129] Step S41: extract features from historical distribution box fault case data and perform correlation analysis to obtain a case distribution box fault feature vector; construct a preliminary fault case library for the case distribution box fault feature vector to obtain a preliminary fault case library;
[0130] In the embodiment of the present invention, the historical distribution box fault case data is analyzed to extract key information related to the fault, such as the time of the fault, the type of fault, the location of the fault, the operating environment parameters at that time, etc. This information is associated with the fault feature fragment data extracted in step S3 to construct a case distribution box fault feature vector, which contains the feature information and fault labels of the historical case. The fault feature vectors and corresponding fault labels of all historical cases are integrated together to construct a preliminary fault case library.
[0131] Step S42: performing a fault case cluster analysis on the fault feature vectors of the distribution box in the preliminary fault case library to obtain a fault mode cluster;
[0132] In an embodiment of the present invention, cluster analysis is performed on the distribution box fault feature vectors in the preliminary fault case library, for example, using a K-means clustering algorithm or a density clustering algorithm to divide fault cases with similar features into the same cluster to form a fault mode cluster set, each cluster representing a typical fault mode.
[0133] Step S43: performing a fault mode feature description on the fault mode cluster to obtain a fault mode feature description library;
[0134] In an embodiment of the present invention, each fault mode cluster is analyzed to extract its features, such as calculating the mean, variance, distribution and other statistical information of the features within the cluster, and combining expert experience and domain knowledge to describe the fault mode, such as the fault phenomenon, fault cause, fault location, fault degree, etc., to form a fault mode feature description library.
[0135] Step S44: performing fault mode identification according to the fault feature vector and the fault mode feature description library to obtain a candidate fault mode ranking list;
[0136] In the embodiment of the present invention, the fault feature vector extracted in step S3 is compared and analyzed with each fault mode in the fault mode feature description library, for example, the similarity between the feature vector and each fault mode is calculated, and indicators such as Euclidean distance and cosine similarity can be used. According to the similarity score, all fault modes are sorted to obtain a sorted list of candidate fault modes, and the higher the similarity score of the fault mode in the list, the greater its possibility.
[0137] Step S45: performing similarity confidence calculation according to the candidate fault mode ranking list to obtain a fault mode confidence vector; performing fault similarity vector generation on the candidate fault mode ranking list and the fault mode confidence vector to obtain a fault similarity vector;
[0138] In the embodiment of the present invention, the confidence of each fault mode is calculated according to the similarity score of each fault mode in the candidate fault mode ranking list. For example, the similarity score can be converted into a probability distribution using a softmax function to form a fault mode confidence vector. The candidate fault mode ranking list and the fault mode confidence vector are integrated to generate a fault similarity vector, which is used to characterize the similarity between the current fault and each fault mode.
[0139] Preferably, step S44 includes the following steps:
[0140] Step S441: performing fault mode similarity calculation on the fault feature vector and the fault mode feature description library to obtain a fault mode similarity matrix;
[0141] Step S442: performing fault mode recognition weight allocation on the fault mode similarity matrix to obtain fault mode recognition weight data;
[0142] Step S443: performing similarity weighted fusion on the fault mode similarity matrix according to the fault mode recognition weight data to obtain a weighted fault mode similarity matrix;
[0143] Step S444: screening candidate fault modes of the weighted fault mode similarity matrix using a preset fault mode similarity threshold to obtain a candidate fault mode set;
[0144] Step S445: sorting the candidate fault mode set to obtain a candidate fault mode sorting list.
[0145] In an embodiment of the present invention, a suitable distance measurement method, such as Euclidean distance, Manhattan distance, cosine similarity, etc., is used to calculate the similarity between the fault feature vector extracted in step S3 and each fault mode feature vector in the fault mode feature description library. The calculated similarity values form a matrix, called a fault mode similarity matrix, in which each row represents a fault mode and each column represents a feature dimension. According to the prior knowledge or historical data of each fault mode in the fault mode feature description library, an identification weight is assigned to each fault mode. For example, different weights can be assigned according to factors such as the frequency and severity of each fault mode in the historical data. The higher the weight, the more important the fault mode is, and more attention should be paid to it in the identification process. The fault mode identification weight data obtained in step S442 is applied to the fault mode similarity matrix. For example, the similarity value corresponding to each fault mode can be multiplied by the identification weight of the fault mode to obtain a weighted similarity value. Finally, a weighted fault mode similarity matrix is obtained, which comprehensively considers the similarity calculation and the prior knowledge of the fault mode. A fault mode similarity threshold is set to screen fault modes with high similarity to the current fault feature vector. The fault modes corresponding to the similarity values greater than or equal to the threshold in the weighted fault mode similarity matrix are screened out to form a candidate fault mode set. According to the weighted similarity values corresponding to each fault mode in the candidate fault mode set, the candidate fault modes are sorted in descending order to obtain a candidate fault mode sorting list. In the sorting list, the higher the ranking of the fault mode, the higher the similarity with the current fault feature vector, and the greater the possibility of being the current fault.
[0146] Preferably, step S5 comprises the following steps:
[0147] Step S51: collecting fault-related factor data of the distribution box to obtain a fault-related factor data set, wherein the fault-related factor data set includes load rate data and equipment status data;
[0148] Step S52: performing a failure mode probability assessment based on historical distribution box failure case data and a failure correlation factor data set to obtain a calibrated failure mode probability vector;
[0149] Step S53: performing load factor probability correction on the calibration fault mode probability vector according to the load rate data to obtain a load corrected fault mode probability vector;
[0150] Step S54: performing device state probability correction on the load correction fault mode probability vector according to the device state data to obtain a device state correction fault mode probability vector;
[0151] Step S55: performing probability normalization processing on the equipment state correction fault mode probability vector to obtain a normalized fault mode probability vector; generating a fault probability distribution according to the normalized fault mode probability vector to obtain fault probability distribution data.
[0152] In an embodiment of the present invention, the load rate data of the distribution box is collected in real time, for example, the current, voltage and other information of each circuit are obtained through a smart meter or a current sensor, and the load rate is calculated. At the same time, the status data of each device in the distribution box is collected, such as the circuit breaker status (on / off), the sensor status (normal / abnormal), the protection device action information, etc. The load rate data and the equipment status data are integrated to form a fault correlation factor data set. The historical distribution box failure case data and the fault correlation factor data set are used to build a fault mode probability assessment model, such as a Bayesian network, a support vector machine, etc. The model is used to perform a probability assessment on each fault mode in the candidate fault mode sorting list obtained in step S4 to obtain an initial fault mode probability vector. Then, the initial probability vector is calibrated according to expert experience or historical data to obtain a calibrated fault mode probability vector to make it more in line with the actual situation. According to the relationship between the load rate data and the fault mode, the calibrated fault mode probability vector is corrected. For example, some fault modes are more likely to occur under high load rate conditions, then the probability values of these fault modes can be correspondingly increased according to the current load rate level to obtain a load correction fault mode probability vector. According to the relationship between the equipment status data and the fault mode, the load correction fault mode probability vector is further corrected. For example, if a certain equipment fails, the probability value of the fault mode related to the equipment should be increased accordingly to obtain the equipment status correction fault mode probability vector. The equipment status correction fault mode probability vector is normalized to ensure that the sum of the probability values of all fault modes is 1, and the normalized fault mode probability vector is obtained. Finally, according to the normalized fault mode probability vector, the fault probability distribution data is generated to intuitively display the possibility of each fault mode.
[0153] Preferably, step S52 includes the following steps:
[0154] Step S521: performing Bayesian network structure learning based on historical distribution box failure case data to obtain a Bayesian network structure diagram;
[0155] Step S522: using the fault correlation factor data set to perform network parameter learning on the Bayesian network structure diagram to obtain a parameterized Bayesian network;
[0156] Step S523: inputting the fault similarity vector into the parameterized Bayesian network to update the node state probability, thereby obtaining an updated Bayesian network;
[0157] Step S524: performing network reasoning and probability propagation calculation on the updated Bayesian network to obtain a posterior probability vector of the fault mode;
[0158] Step S525: Perform a posteriori probability calibration on the fault mode posterior probability vector to obtain a calibrated fault mode probability vector.
[0159] In an embodiment of the present invention, historical distribution box fault case data is analyzed to extract key variables such as fault mode, load rate, and equipment status, and a Bayesian network structure learning algorithm, such as a K2 algorithm, a PC algorithm, etc., is used to learn the dependency between variables from the data and construct a Bayesian network structure diagram. The structure diagram reflects the causal relationship or association relationship between different variables. The fault correlation factor data set obtained in step S51 is used to perform parameter learning on the Bayesian network structure diagram obtained in step S521. For example, the conditional probability table (CPT) or other parameterized forms of each node in the network can be learned using methods such as maximum likelihood estimation and Bayesian estimation to obtain a parameterized Bayesian network. The network can quantitatively describe the probability relationship between variables. The fault similarity vector obtained in step S4 is input as evidence into the parameterized Bayesian network to update the node state probability associated with the fault mode. For example, if the fault similarity vector shows that a certain fault mode is more likely, the state probability of the node corresponding to the fault mode can be set to a higher value. According to the updated Bayesian network, the posterior probability of each fault mode under the current evidence (fault similarity vector) is calculated using the Bayesian network inference algorithm, such as the joint tree algorithm, the variable elimination method, etc., to obtain the fault mode posterior probability vector. This vector reflects the probability of occurrence of each fault mode after considering the fault similarity information. Due to the influence of historical data, model assumptions and other factors, the fault mode posterior probability vector obtained in step S524 may be biased. Therefore, it is necessary to calibrate the posterior probability vector based on expert experience, domain knowledge or other reliable information. For example, the Platt calibration method, rank-preserving regression and other methods can be used to obtain a more accurate and reliable calibrated fault mode probability vector.
[0160] Preferably, step S6 comprises the following steps:
[0161] Step S61: sorting the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list;
[0162] Step S62: acquiring the distribution box circuit topology data; performing a fault mode circuit association analysis according to the fault mode sorting list and the distribution box circuit topology data to obtain a fault mode circuit association table;
[0163] Step S63: Screen candidate circuits according to the fault mode circuit association table and the fault mode sorting list to obtain a candidate fault circuit set; calculate circuit characteristic indexes for the candidate fault circuit set to obtain a circuit characteristic index variation matrix;
[0164] Step S64: updating the loop fault probability according to the loop characteristic index variation matrix to obtain an updated loop fault probability vector; performing loop topology analysis on the distribution box loop topology data to obtain a loop topology relationship diagram;
[0165] Step S65: locating the fault area according to the updated loop fault probability vector and the loop topology relationship diagram to obtain fault area locating result data;
[0166] Step S66: Perform fault cause analysis based on the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
[0167] As an example of the present invention, refer to Figure 3 As shown, in this example, step S6 includes:
[0168] Step S61: sorting the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list;
[0169] In the embodiment of the present invention, the fault probability distribution data obtained in step S5 is integrated with the candidate fault mode ranking list obtained in step S4, and the candidate fault modes are re-ranked according to the fault probability value to obtain a new fault mode ranking list, wherein the higher the probability value, the higher the ranking.
[0170] Step S62: acquiring the distribution box circuit topology data; performing a fault mode circuit association analysis according to the fault mode sorting list and the distribution box circuit topology data to obtain a fault mode circuit association table;
[0171] In the embodiment of the present invention, the circuit topology data of the distribution box, such as the primary wiring diagram, the equipment list, etc., is obtained to understand the connection relationship between the circuits, the protection configuration and other information. Then, according to the characteristics of each fault mode in the fault mode sorting list and the circuit topology data, the circuits that may be affected by the fault mode are analyzed, and the association relationship between the fault mode and the circuit is established to form a fault mode circuit association table.
[0172] Step S63: Screen candidate circuits according to the fault mode circuit association table and the fault mode sorting list to obtain a candidate fault circuit set; calculate circuit characteristic indexes for the candidate fault circuit set to obtain a circuit characteristic index variation matrix;
[0173] In the embodiment of the present invention, according to the fault mode loop association table, loops related to high probability fault modes are screened out to form a candidate fault loop set. Then, for each loop in the candidate fault loop set, its characteristic index, such as current, voltage, power, harmonics, etc., is calculated, and compared with the reference value under normal operating conditions, and the characteristic index change amount is calculated to form a loop characteristic index change amount matrix.
[0174] Step S64: updating the loop fault probability according to the loop characteristic index variation matrix to obtain an updated loop fault probability vector; performing loop topology analysis on the distribution box loop topology data to obtain a loop topology relationship diagram;
[0175] In an embodiment of the present invention, the failure probability of the candidate faulty circuit is updated according to the circuit characteristic index change matrix, combined with expert experience or a machine learning model, to obtain an updated circuit failure probability vector. The larger the change, the higher the possibility of failure of the circuit. At the same time, the circuit topology data of the distribution box is analyzed, for example, using a graph theory method to construct a circuit topology relationship diagram for visually displaying the connection relationship between the circuits.
[0176] Step S65: locating the fault area according to the updated loop fault probability vector and the loop topology relationship diagram to obtain fault area locating result data;
[0177] In an embodiment of the present invention, the fault area is located by combining the updated loop fault probability vector and the loop topology relationship diagram, using a fault propagation model or a graph search algorithm, such as a depth-first search, a breadth-first search, etc., to identify the loop or device that is most likely to fail, and obtain fault area location result data.
[0178] Step S66: Analyze the cause of the fault according to the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection;
[0179] In an embodiment of the present invention, based on the fault area location result data, combined with the fault cause, fault phenomenon and other information related to the fault mode in the fault mode feature description library, the fault cause is analyzed, and a distribution box fault cause analysis report is generated to provide a reference basis for maintenance.
[0180] Preferably, step S62 includes the following steps:
[0181] Step S621: define component failure modes for the failure mode sorting list to obtain a component failure mode library;
[0182] Step S622: acquiring the distribution box circuit topology data; performing a fault mode propagation path analysis according to the component fault mode library and the distribution box circuit topology data to obtain a fault mode propagation path diagram;
[0183] Step S623: identifying loop sensitive components according to the fault mode propagation path diagram to obtain a fault mode sensitive component table;
[0184] Step S624: constructing a fault mode component impact feature vector for the fault mode sorting list according to the fault mode sensitive component table to obtain a fault mode component impact feature vector;
[0185] Step S625: constructing loop element feature vectors according to the distribution box loop topology data to obtain a loop element feature vector table;
[0186] Step S626: performing fault mode-loop correlation calculation on the fault mode component influence feature vector and the loop component feature vector table to obtain a fault mode-loop correlation matrix;
[0187] Step S627: setting a correlation threshold according to the fault mode-loop correlation matrix to obtain a correlation threshold; screening the fault mode-loop correlation matrix for fault mode correlation loops according to the correlation threshold to obtain a fault mode loop correlation table.
[0188] In an embodiment of the present invention, according to the characteristic description of each fault mode in the fault mode sorting list, the component faults that may cause the fault mode are analyzed, and these component fault modes are defined and classified to establish a component fault mode library. For example, for the overcurrent fault mode, component fault modes such as circuit breaker short circuit fault, line overload fault, motor stall fault, etc. can be defined. The distribution box loop topology data is obtained, including information such as a primary wiring diagram, protection setting, and equipment parameters. Then, according to the characteristics of each component fault mode in the component fault mode library, combined with the loop topology data, the propagation path of the fault mode in the distribution box loop, such as the propagation path of current, voltage, signal, etc., is analyzed, and visualized in the form of a graph to obtain a fault mode propagation path diagram. According to the fault mode propagation path diagram, the key components that may be affected by each fault mode on the propagation path, such as circuit breakers, contactors, sensors, etc., are analyzed, and these components are identified as loop sensitive components, and a fault mode sensitive component table is established. According to the fault mode sensitive component table, a component impact feature vector is constructed for each fault mode. This vector contains the degree of influence of the fault mode on each component. For example, 0 or 1 can be used to indicate whether the fault mode will affect the component, or a more precise value can be used to indicate the degree of influence. According to the distribution box loop topology data, a loop element feature vector is constructed for each loop. This vector describes which components are contained in the loop. For example, 0 or 1 can be used to indicate whether the component exists in the loop. The similarity between each fault mode element influence feature vector and each loop element feature vector is calculated. For example, indicators such as Jaccard similarity and cosine similarity can be used to obtain the fault mode-loop association matrix. Each element in the matrix represents the degree of association between the fault mode and the loop. The larger the value, the higher the degree of association. According to the fault mode-loop association matrix, a correlation threshold is set to filter loops with a high degree of association with the fault mode. The fault modes and loops corresponding to the elements in the matrix that are greater than or equal to the threshold are filtered out to establish a fault mode loop association table.
[0189] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.
[0190] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.
Claims
1. A method for detecting faults in a distribution box, It is characterized in that The following steps are involved: Step S1: collecting sensor data from the distribution box through temperature and current sensors and performing multi-dimensional data preprocessing to obtain a multi-dimensional time series data set; Step S2: constructing a high-dimensional feature vector for the multi-dimensional time series data set to obtain a high-dimensional feature vector set; Dynamic feature extraction is performed on the high-dimensional feature vector set to obtain a multi-dimensional feature matrix, wherein step S2 includes the following steps: Step S21: performing time series segmentation on the multidimensional time series data set to obtain a time series segment set; performing preliminary extraction of univariate features of temperature and current data on each time segment in the time series segment set to obtain a univariate feature set; Step S22: performing time series feature enhancement on the time series segment set according to the univariate feature set to obtain an enhanced time domain feature set; performing fast Fourier transform on each time segment in the enhanced time domain feature set to obtain a frequency domain feature set; Step S23: constructing a high-dimensional feature vector for the univariate feature set, the enhanced time domain feature set, and the frequency domain feature set to obtain a high-dimensional feature vector set; Step S24: performing feature correlation analysis on the high-dimensional feature vector set to obtain a feature correlation matrix; sorting the feature importance of the high-dimensional feature vector set according to the feature correlation matrix to obtain a key feature sorting set; Step S25: performing feature dimensionality reduction on the key feature sorting set to obtain a reduced-dimensional feature set; performing feature time series reconstruction on the reduced-dimensional feature set to obtain a fault feature time series; performing multidimensional feature matrix generation on the fault feature time series to obtain a multidimensional feature matrix; Step S3: Obtain historical distribution box failure case data; perform difference analysis of different feature subsets on the multidimensional feature matrix to obtain a feature difference statistic matrix; extract fault feature fragments and reconstruct fault feature vectors based on the feature difference statistic matrix and historical distribution box failure case data to obtain a fault feature vector, wherein step S3 includes the following steps: Step S31: performing time series pattern clustering analysis on the time series in the multidimensional feature matrix to obtain a time series pattern cluster set; performing intra-cluster feature association analysis on the time series pattern cluster set to obtain an intra-cluster feature association matrix set; Step S32: performing key feature subset identification according to the intra-cluster feature association matrix set to obtain the key feature subset; performing difference analysis on different feature subsets in the key feature subset to obtain a feature difference statistics matrix; Step S33: Calculate feature weights according to the feature difference statistics matrix to obtain a feature weight vector; perform multi-feature weighted fusion on the key feature subset according to the feature weight vector to obtain a fused feature time series; Step S34: Acquire historical distribution box failure case data; locate the failure time window according to the distribution box historical failure data in the historical distribution box failure case data to obtain failure time window data; Step S35: extracting fault feature segments from the fused feature time series according to the fault time window data to obtain fault feature segment data; reconstructing fault feature vectors from the fault feature segment data to obtain fault feature vectors; Step S4: performing case failure mode feature description analysis based on historical distribution box failure case data to obtain a failure mode feature description library; performing failure mode identification based on the failure feature vector and the failure mode feature description library to obtain a candidate failure mode ranking list; generating a failure similarity vector based on the candidate failure mode ranking list to obtain a failure similarity vector, wherein step S4 includes the following steps: Step S41: extract features from historical distribution box fault case data and perform correlation analysis to obtain a case distribution box fault feature vector; construct a preliminary fault case library for the case distribution box fault feature vector to obtain a preliminary fault case library; Step S42: performing a fault case cluster analysis on the fault feature vectors of the distribution box in the preliminary fault case library to obtain a fault mode cluster; Step S43: performing a fault mode feature description on the fault mode cluster to obtain a fault mode feature description library; Step S44: performing fault mode identification according to the fault feature vector and the fault mode feature description library to obtain a candidate fault mode ranking list, wherein step S44 includes the following steps: Step S441: performing fault mode similarity calculation on the fault feature vector and the fault mode feature description library to obtain a fault mode similarity matrix; Step S442: performing fault mode recognition weight allocation on the fault mode similarity matrix to obtain fault mode recognition weight data; Step S443: performing similarity weighted fusion on the fault mode similarity matrix according to the fault mode recognition weight data to obtain a weighted fault mode similarity matrix; Step S444: screening candidate fault modes of the weighted fault mode similarity matrix using a preset fault mode similarity threshold to obtain a candidate fault mode set; Step S445: sorting the candidate fault mode set to obtain a candidate fault mode sorting list; Step S45: performing similarity confidence calculation according to the candidate fault mode ranking list to obtain a fault mode confidence vector; performing fault similarity vector generation on the candidate fault mode ranking list and the fault mode confidence vector to obtain a fault similarity vector; Step S5: Collect the data of fault-related factors of the distribution box to obtain a data set of fault-related factors; perform a fault mode probability assessment based on the historical distribution box fault case data and the data set of fault-related factors to obtain a calibrated fault mode probability vector; perform a fault probability correction on the calibrated fault mode probability vector based on the data set of fault-related factors to obtain fault probability distribution data; Step S6: Sort the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list; obtain the distribution box loop topology data; perform fault mode loop association analysis according to the fault mode sorting list and the distribution box loop topology data to obtain a fault mode loop association table; locate the fault area according to the fault mode loop association table and the fault mode sorting list to obtain fault area location result data; perform fault cause analysis according to the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
2. The distribution box fault detection method according to claim 1, It is characterized in that Step S1 includes the following steps: Step S11: collecting sensor data from the distribution box through temperature and current sensors to obtain sensor data; Step S12: Cleaning the sensor data to obtain sensor cleaning data; Step S13: performing time-space alignment on the sensor cleaning data to obtain time-space alignment data; Step S14: performing data integration on the spatiotemporal aligned data to obtain a multi-dimensional time series data set.
3. The distribution box fault detection method according to claim 1, It is characterized in that Step S5 includes the following steps: Step S51: collecting fault-related factor data of the distribution box to obtain a fault-related factor data set, wherein the fault-related factor data set includes load rate data and equipment status data; Step S52: performing a failure mode probability assessment based on historical distribution box failure case data and a failure correlation factor data set to obtain a calibrated failure mode probability vector; Step S53: performing load factor probability correction on the calibration fault mode probability vector according to the load rate data to obtain a load corrected fault mode probability vector; Step S54: performing device state probability correction on the load correction fault mode probability vector according to the device state data to obtain a device state correction fault mode probability vector; Step S55: performing probability normalization processing on the equipment state correction fault mode probability vector to obtain a normalized fault mode probability vector; generating a fault probability distribution according to the normalized fault mode probability vector to obtain fault probability distribution data.
4. The method for detecting a fault in a distribution box according to claim 3, It is characterized in that Step S52 includes the following steps: Step S521: performing Bayesian network structure learning based on historical distribution box failure case data to obtain a Bayesian network structure diagram; Step S522: using the fault correlation factor data set to perform network parameter learning on the Bayesian network structure diagram to obtain a parameterized Bayesian network; Step S523: inputting the fault similarity vector into the parameterized Bayesian network to update the node state probability, thereby obtaining an updated Bayesian network; Step S524: performing network reasoning and probability propagation calculation on the updated Bayesian network to obtain a posterior probability vector of the fault mode; Step S525: Perform a posteriori probability calibration on the fault mode posterior probability vector to obtain a calibrated fault mode probability vector.
5. The distribution box fault detection method according to claim 1, It is characterized in that Step S6 includes the following steps: Step S61: sorting the candidate fault mode sorting list by fault probability according to the fault probability distribution data to obtain a fault mode sorting list; Step S62: acquiring the distribution box circuit topology data; performing a fault mode circuit association analysis according to the fault mode sorting list and the distribution box circuit topology data to obtain a fault mode circuit association table; Step S63: Screen candidate circuits according to the fault mode circuit association table and the fault mode sorting list to obtain a candidate fault circuit set; calculate circuit characteristic indexes for the candidate fault circuit set to obtain a circuit characteristic index variation matrix; Step S64: updating the loop fault probability according to the loop characteristic index variation matrix to obtain an updated loop fault probability vector; performing loop topology analysis on the distribution box loop topology data to obtain a loop topology relationship diagram; Step S65: locating the fault area according to the updated loop fault probability vector and the loop topology relationship diagram to obtain fault area locating result data; Step S66: Perform fault cause analysis based on the fault area location result data and the fault mode feature description library to obtain a distribution box fault cause analysis report to implement distribution box fault detection.
6. The method for detecting a distribution box fault according to claim 5, It is characterized in that Step S62 includes the following steps: Step S621: define component failure modes for the failure mode sorting list to obtain a component failure mode library; Step S622: acquiring the distribution box circuit topology data; performing a fault mode propagation path analysis according to the component fault mode library and the distribution box circuit topology data to obtain a fault mode propagation path diagram; Step S623: identifying loop sensitive components according to the fault mode propagation path diagram to obtain a fault mode sensitive component table; Step S624: constructing a fault mode component impact feature vector for the fault mode sorting list according to the fault mode sensitive component table to obtain a fault mode component impact feature vector; Step S625: constructing loop element feature vectors according to the distribution box loop topology data to obtain a loop element feature vector table; Step S626: performing fault mode-loop correlation calculation on the fault mode component influence feature vector and the loop component feature vector table to obtain a fault mode-loop correlation matrix; Step S627: setting a correlation threshold according to the fault mode-loop correlation matrix to obtain a correlation threshold; screening the fault mode-loop correlation matrix for fault mode correlation loops according to the correlation threshold to obtain a fault mode loop correlation table.
Citation Information
Patent Citations
Intelligent power distribution network fault positioning method and system
CN116559598A
Intelligent diagnosis and isolation device and method for line fault of power distribution network
CN118607390A
Cited By
Distribution box fault positioning method and system based on multi-dimensional time-frequency characteristics
CN120831609A
