Power distribution cabinet fault detection method and system

Abstract

The invention discloses a power distribution cabinet fault detection method and system, and relates to the technical field of power equipment state monitoring, and the method comprises the steps: laying a multi-modal sensor array, constructing a multi-dimensional data collection network, collecting various types of sensing data in real time, and carrying out the data preprocessing; establishing a multi-modal data fusion model by using a deep learning algorithm, and forming a fault feature mapping relation through feature extraction, data correlation analysis and cross validation to realize collaborative perception of a data level; and constructing a multi-dimensional fault diagnosis decision tree based on a collaborative perception result, accurately positioning a specific device position where a fault occurs in combination with weight distribution and confidence evaluation of data of each sensor, and giving out evaluation results of a fault type and severity. The system effectively reduces the equipment failure rate, prolongs the service life, reduces the maintenance cost, improves the power supply reliability, promotes the development of an electric power system towards the intelligent operation and maintenance direction, and is of great significance to the improvement of the operation reliability and maintenance efficiency of a power distribution system.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment status monitoring, and in particular to a distribution cabinet fault detection method and system. Background Art

[0002] The current intelligent operation and maintenance of power distribution systems faces many technical challenges. Traditional methods for fault diagnosis and life prediction of distribution cabinets mainly rely on a single type of sensor data for analysis, which makes it difficult to fully reflect the operating status of the equipment. In the existing technology, although the monitoring method based on electrical parameters can reflect the basic operating status of the equipment, it cannot effectively identify mechanical faults and early degradation characteristics; the method based on infrared temperature measurement has advantages in temperature anomaly detection, but is easily affected by environmental factors and difficult to analyze the root cause of the fault; the method based on acoustic detection is more sensitive to partial discharge faults, but is prone to misjudgment in complex environments; the monitoring method based on mechanical characteristics can reflect the changes in the mechanical properties of the equipment, but it is difficult to achieve a comprehensive status assessment when used alone.

[0003] At the same time, existing fault diagnosis methods often use simple threshold judgments or single machine learning models, lacking the ability to deeply integrate and analyze multi-source heterogeneous data, resulting in insufficient accuracy and reliability of diagnostic results. In terms of life prediction, traditional methods mainly rely on statistical analysis and empirical models, failing to fully utilize the multimodal data information generated during the operation of the equipment, making it difficult to accurately assess the health status of the equipment and predict the remaining service life. In addition, existing technologies generally have problems such as low accuracy and poor interpretability in fault location and severity assessment, making it difficult to provide effective support for maintenance decisions. Therefore, it is urgent to develop a distribution cabinet status monitoring and life prediction method that can comprehensively utilize multimodal data and has intelligent analysis and precise diagnosis capabilities to improve the operating reliability and maintenance efficiency of the distribution system. Summary of the invention

[0004] In view of the problems existing in the existing distribution cabinet fault detection method, the present invention proposes a distribution cabinet fault detection method and system.

[0005] Therefore, the problem to be solved by the present invention is how to intelligently analyze and accurately diagnose the status of the distribution cabinet and predict the life span so as to improve the operating reliability and maintenance efficiency of the distribution system.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, an embodiment of the present invention provides a distribution cabinet fault detection method, which includes: deploying a multimodal sensor array, constructing a multidimensional data acquisition network, collecting various types of sensor data in real time, including temperature data, current data, vibration acceleration data, and partial discharge signal data, and preprocessing the collected data; using a deep learning algorithm to establish a multimodal data fusion model, and through feature extraction, data correlation analysis and cross-validation, forming a fault feature mapping relationship to achieve collaborative perception at the data level; based on the collaborative perception results, constructing a multidimensional fault diagnosis decision tree, combining the weight allocation and confidence evaluation of each sensor data, accurately locating the specific device location where the fault occurs, and giving an evaluation result of the fault type and severity.

[0007] As a preferred solution of the power distribution cabinet fault detection method of the present invention, wherein: The temperature sensors are evenly arranged along the busbar inside the power distribution cabinet, and the distance between adjacent temperature sensors is Satisfy the formula ,in, is the total length of the busbar, is the number of temperature sensors, collecting the temperature data ; Vibration sensors are respectively set at the positions of the temperature sensors to collect the vibration acceleration data , vibration acceleration data Sampling frequency Satisfy the formula ,in, The Hall current sensor is set at the incoming line end of the power distribution cabinet to collect the current data. , current data Satisfy the formula ,in, is the sensitivity coefficient of the Hall element, is the Hall voltage, the current data With the temperature data Synchronous sampling; an ultrasonic partial discharge sensor is set on the top of the distribution cabinet to collect the partial discharge signal D, the partial discharge signal The sampling frequency Satisfy the formula .

[0008] As a preferred solution of the power distribution cabinet fault detection method of the present invention, the preprocessing includes data synchronization and noise reduction processing; the data synchronization is based on timestamp alignment, and a linear interpolation method is used to process data with inconsistent sampling frequencies: , Where Y represents any sensor data, are adjacent sampling moments, is the time to be interpolated; the sensor data matrix M is generated based on the synchronized data, and the sensor data matrix M satisfies: , in, is the temperature data time series, is the vibration acceleration data time series, is the current data time series, is the time series of partial discharge signal, and the sampling frequency is 4 times of other signals; is the number of sampling points; perform db4 wavelet basis decomposition on each row of the signal sequence in the sensor data matrix M to satisfy the formula: , in, is the input signal sequence, is the db4 wavelet basis function, To decompose the scale, is the translation parameter; three-layer wavelet decomposition is used to obtain the scale coefficients, which are expressed as: , , in, For the Layer low frequency coefficients, is the high frequency coefficient of the j+1th layer, is the low-pass filter coefficient, is the high-pass filter coefficient, is the filter length of the db4 wavelet; soft threshold processing is performed on the high-frequency coefficients, and the threshold function is: , in, is the adaptive threshold, Represents the high-frequency coefficients obtained by wavelet decomposition of each layer; the adaptive threshold is obtained by maximum likelihood estimation , expressed as: , in, is the noise standard deviation, is the signal length, is the high frequency coefficient of the first layer, is the median operator; further, the processed coefficients are used for wavelet reconstruction to obtain the denoised signal: , in, To reconstruct the low-pass filter coefficients, To reconstruct the high-pass filter coefficients, Represents rounding down; reassemble the reconstructed row signals to obtain the preprocessed sensor data matrix M1: , in, , , , They are the denoised temperature, vibration, current and partial discharge signal sequences respectively.

[0009] As a preferred solution of the distribution cabinet fault detection method of the present invention, the feature extraction formula is expressed as: , Among them, M1 is the pre-processed sensor data matrix, which contains four types of signals: temperature, vibration, current, and partial discharge; is the generalized time-frequency kernel matrix; further, the kernel matrix It is expressed as: , in, , , are the wavelet basis functions of temperature, vibration and partial discharge signals respectively. represents the conjugate complex number of the wavelet basis function, , , is the corresponding scale parameter; is the time variable, is the frequency variable, is the integral variable; according to the difference in contribution of different modal features to fault diagnosis, adaptive feature fusion is performed through the attention mechanism, which is expressed as: , , in, For the The query vector is The attention weight value of the key vector, is the query vector, representing the focus of the target feature, For the The key vector of features represents the content of the features to be fused. For the The attention weight of each modality feature, For the The characteristics of a mode, is the feature dimension, is the number of modes, is the feature representation after fusion; in order to ensure the training stability and feature transfer efficiency of the deep network, the residual learning mechanism is introduced: , Among them, R is the feature after residual enhancement, is the network mapping function, which includes multiple layers of convolution and activation operations. The input features are directly passed to the output through the skip connection.

[0010] As a preferred solution of the distribution cabinet fault detection method of the present invention, a multi-task learning architecture is constructed based on the extracted data features, and the fault diagnosis and life prediction tasks are optimized at the same time. The joint loss function is expressed as: , , , in, is the total loss function, is the cross entropy loss, is the mean square error loss, used for life prediction tasks, is the true label of the fault type, is the predicted probability of failure type, is the actual remaining life, To predict the remaining life, is the task weight coefficient, determined by cross-validation, is the number of fault categories, is the number of samples; further, the correlation strength between the feature pattern and the fault type is described by the feature mapping measured by similarity, which is expressed as: , , , Where M is the fault feature mapping matrix, Indicates Class characteristics and The mapping strength of class faults, is the feature weight coefficient, reflecting the importance of the feature, For the Class characteristics Quantity, For the The feature template of the class fault, σ is the Gaussian kernel parameter, which controls the mapping smoothness, and K is the feature dimension.

[0011] As a preferred solution of the distribution cabinet fault detection method of the present invention, based on the multimodal feature fusion matrix F and the attention weight matrix As input, by calculating the feature importance score Establish the basis for feature selection and construct the feature importance ranking set T = { , ..., }; Based on the feature importance ranking set T, the information gain ratio criterion IG is used to determine the optimal split feature: ,in, is the information entropy of feature set T, Features Conditional entropy under the condition; further, the feature with the maximum information gain ratio is selected as the node splitting condition to construct a multi-dimensional fault diagnosis decision tree D. Each internal node Include feature splitting threshold and weight , leaf nodes Storage Failure Type and failure probability , for internal nodes in a decision tree , based on the feature vector and split threshold Perform binary split: When the eigenvalue Less than or equal to the split threshold When , the sample is assigned to the left child node; when the eigenvalue Greater than the split threshold , the sample is assigned to the right child node.

[0012] As a preferred solution of the distribution cabinet fault detection method of the present invention, wherein: based on the decision tree D and the optimized sensor weights, the device fault location matrix M is calculated to establish a fault location mapping relationship; wherein the fault location matrix M is expressed as: , in, Indicates the device index, Indicates the fault type index, is the failure probability of the leaf node, is the optimized sensor weight; combined with the fault location matrix M and the residual feature R, a diagnostic confidence evaluation model is constructed to provide a quantitative indicator of the reliability of the diagnostic result, which is expressed as: , in, and is the balance coefficient, represents the L2 norm of the residual feature; based on the fault location matrix M and the confidence C, the fault severity evaluation function E is constructed, which is expressed as: , in, is the severity coefficient corresponding to the fault type, and outputs the fault severity level.

[0013] In the second aspect, an embodiment of the present invention provides a distribution cabinet fault detection system, which includes: a collection module, which is used to deploy a multimodal sensor array, build a multidimensional data acquisition network, collect various types of sensor data in real time and perform data preprocessing; an analysis module, which establishes a multimodal data fusion model by utilizing a deep learning algorithm, forms a fault feature mapping relationship through feature extraction, data correlation analysis and cross-validation, and realizes collaborative perception at the data level; a detection module, which constructs a multidimensional fault diagnosis decision tree through the collaborative perception results, combines the weight allocation and confidence assessment of each sensor data, accurately locates the specific device position where the fault occurs, and gives an assessment result of the fault type and severity.

[0014] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, any step of the above-mentioned distribution cabinet fault detection method is implemented.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned distribution cabinet fault detection method is implemented.

[0016] The beneficial effect of the present invention is that the present invention realizes a complete technical chain from data collection, feature learning to fault diagnosis, and establishes a set of systematic distribution cabinet status evaluation and life prediction methods. It has significant engineering application value, can accurately identify potential equipment failures, and warn of possible operating risks in advance, providing a scientific basis for preventive maintenance. It can effectively reduce the equipment failure rate, extend the service life, reduce maintenance costs, improve power supply reliability, and promote the development of power systems towards intelligent operation and maintenance. It is of great significance to improve the operating reliability and maintenance efficiency of the distribution system, and can provide new technical ideas and solutions for the field of power equipment status monitoring and intelligent diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 The figure is a flow chart of a method for detecting faults in a power distribution cabinet. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0021] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and provides a method for detecting a fault in a power distribution cabinet, comprising: S1: Deploy a multimodal sensor array, build a multi-dimensional data acquisition network, collect various sensor data in real time and perform data preprocessing.

[0022] Will The temperature sensors are evenly arranged along the busbar inside the power distribution cabinet, and the distance between adjacent temperature sensors is Satisfy the formula ,in, is the total length of the busbar, Is the number of temperature sensors, collecting temperature data ; Vibration sensors are set at the locations of temperature sensors to collect vibration acceleration data. , vibration acceleration data Sampling frequency Satisfy the formula ,in, is the sampling time interval, The same sampling frequency of 2kHz is used as the temperature data T; The Hall current sensor is set at the incoming line end of the power distribution cabinet to collect the current data , current data Satisfy the formula ,in, is the sensitivity coefficient of the Hall element, Hall voltage and current data With temperature data Synchronous sampling; An ultrasonic partial discharge sensor is installed on the top of the distribution cabinet to collect the partial discharge signal D. The sampling frequency of the partial discharge signal D is Satisfy the formula ,in, is the sampling frequency of temperature data T, = 8kHz; The temperature data T, vibration acceleration data V, current data I and partial discharge signal D are sampled synchronously. Data synchronization is based on timestamp alignment, and linear interpolation is used to process data with inconsistent sampling frequencies: , Where Y represents any sensor data, are adjacent sampling moments, is the time to be interpolated.

[0023] The sensor data matrix M is generated based on the synchronized data, and the sensor data matrix M satisfies: , in, is the temperature data time series, is the vibration acceleration data time series, is the current data time series, is the time series of partial discharge signal, and the sampling frequency is 4 times of other signals; is the number of sampling points.

[0024] Perform db4 wavelet basis decomposition on each row of the signal sequence in the sensor data matrix M to satisfy the formula: , in, is the input signal sequence, is the db4 wavelet basis function, To decompose the scale, is the translation parameter.

[0025] Using 3-layer wavelet decomposition, we can obtain the scale coefficients, which are expressed as: , , in, For the Layer low frequency coefficients, is the high frequency coefficient of the j+1th layer, is the low-pass filter coefficient, is the high-pass filter coefficient, is the filter length of the db4 wavelet.

[0026] Soft threshold processing is performed on high-frequency coefficients, and the threshold function is: , in, is the adaptive threshold, Represents the high-frequency coefficients obtained by wavelet decomposition of each layer.

[0027] Adaptive thresholding using maximum likelihood estimation , expressed as: , in, is the noise standard deviation, is the signal length, is the high frequency coefficient of the first layer, is the median operator.

[0028] Furthermore, the processed coefficients are used for wavelet reconstruction to obtain the denoised signal: , in, To reconstruct the low-pass filter coefficients, To reconstruct the high-pass filter coefficients, Indicates rounding down.

[0029] Reassemble the reconstructed row signals to obtain the preprocessed sensor data matrix M1: , in, , , , They are the denoised temperature, vibration, current and partial discharge signal sequences respectively.

[0030] S2: Use deep learning algorithms to establish a multimodal data fusion model. Through feature extraction, data correlation analysis and cross-validation, a fault feature mapping relationship is formed to achieve collaborative perception at the data level.

[0031] Faults in power distribution cabinets often manifest themselves in multiple sensor signals simultaneously. For example, poor contact faults can cause temperature anomalies and vibration anomalies at the same time, while insulation degradation is mainly reflected in partial discharge signals. In order to fully capture these multi-dimensional fault characteristics, a unified feature extraction based on generalized time-frequency transform is used, and the time domain evolution characteristics and frequency domain distribution characteristics of the signal are analyzed at the same time, providing rich feature information for subsequent fault identification.

[0032] The formula for feature extraction is expressed as: , Among them, M1 is the pre-processed sensor data matrix, which contains four types of signals: temperature, vibration, current, and partial discharge; It is a generalized time-frequency kernel matrix, and the independent extraction of each modal feature is achieved through block diagonal design.

[0033] Furthermore, the kernel matrix It is expressed as: , in, , , are the wavelet basis functions of temperature, vibration and partial discharge signals respectively. Morlet wavelet is selected to provide good time-frequency positioning performance. represents the complex conjugate of the wavelet basis function. , , is the corresponding scale parameter, which is determined according to the characteristic frequency range of each type of signal; is a time variable, reflecting the time domain distribution of the feature, is a frequency variable, reflecting the frequency domain distribution of the feature, is the integration variable.

[0034] Through the kernel matrix The independence of feature extraction of different modes is ensured to avoid mutual interference between modes. Customized wavelet basis functions are used for different types of signals such as temperature and vibration to improve the pertinence of feature extraction.

[0035] Since the fault characteristics of the distribution cabinet are manifested at different time and frequency scales, feature extraction is performed through a multi-scale parallel convolution structure, which is expressed as: , in, For the Layer feature map, which represents the high-level semantic information of fault features. are convolution kernels of different sizes, used to capture fine-grained, medium-scale, and large-scale features. Output of the previous layer to ensure hierarchical extraction of features. is a bias term used to adjust the feature activation threshold, is the ReLU activation function, which introduces nonlinear transformation capability, and K is the number of parallel convolution branches.

[0036] In actual operation, there are significant differences in the contribution of different sensor signals to fault diagnosis, and this difference changes dynamically with the fault type. For example, for poor contact faults, the temperature signal has a higher diagnostic value; while for mechanical faults, the vibration signal is more indicative. Based on the difference in the contribution of different modal features to fault diagnosis, adaptive feature fusion is performed through the attention mechanism, which is expressed as: , , in, For the The query vector is The attention weight value of the key vector, is the query vector, representing the focus of the target feature, For the The key vector of features represents the content of the features to be fused. For the The attention weight of each modality feature, For the The characteristics of a mode, is the feature dimension, is the number of modes, is the feature representation after fusion.

[0037] In order to capture the temporal law of fault development, the LSTM network is used to build a temporal feature extractor to extract temporal features. Specifically, the gating mechanism is used to control the flow of information to solve the gradient vanishing problem in long sequence modeling; the cell state is used to maintain long-term memory to capture the fault development trend; the input gate and the forget gate are used to achieve selective information update; the output gate is used to control the degree of information exposure to provide a more stable feature representation.

[0038] Deep neural networks are prone to gradient vanishing and degradation problems when extracting high-level semantic features. In order to ensure the training stability and feature transfer efficiency of deep networks, a residual learning mechanism is introduced: , Among them, R is the feature after residual enhancement, is the network mapping function, which includes multiple layers of convolution and activation operations. The input features are directly passed to the output through the skip connection.

[0039] Based on the extracted time series rules, the life prediction of the distribution cabinet is carried out. Since fault diagnosis and life prediction are interrelated tasks, and through the inherent correlation, a multi-task learning architecture is constructed to optimize the fault diagnosis and life prediction tasks at the same time. The joint loss function is expressed as: , , , in, is the total loss function, is the cross entropy loss, used for fault diagnosis tasks, is the mean square error loss, used for life prediction tasks, is the true label of the fault type, is the predicted probability of failure type, is the actual remaining life, To predict the remaining life, is the task weight coefficient, determined by cross-validation, is the number of fault categories, is the sample size.

[0040] Through the multi-task learning architecture, auxiliary tasks are set to provide additional supervision information for the main task, and the coordinated optimization between tasks is ensured through the weight balance of the loss function.

[0041] In order to establish the correspondence between features and fault types, the feature mapping measured by similarity describes the association strength between feature patterns and fault types, which is expressed as: , , , Where M is the fault feature mapping matrix, Indicates Class characteristics and The mapping strength of class faults, is the feature weight coefficient, reflecting the importance of the feature, For the Class characteristics Quantity, For the The feature template of the class fault, σ is the Gaussian kernel parameter, which controls the mapping smoothness, and K is the feature dimension.

[0042] The mapping matrix provides an interpretable representation of the feature-fault relationship and supports rapid adaptation of new fault types, thus achieving traceability of the fault diagnosis process.

[0043] Furthermore, to ensure the reliability of feature mapping, a verification method based on consistency indicators and association maps is constructed, which is expressed as: , , in, is the mapping consistency indicator, represents the trace of the matrix, represents the Frobenius norm, is the feature-fault association map, is the feature category, is the fault type, is the association threshold and is used to filter significant mapping relationships.

[0044] The consistency index is used to provide a quantitative evaluation of the mapping quality, and the corresponding relationship between features and faults is intuitively displayed through the association map, and the threshold screening ensures the extraction of key mapping relationships. By adaptively extracting and fusing multimodal data features, a reliable fault mapping relationship is established, providing strong support for the intelligent operation and maintenance of power distribution equipment.

[0045] S3: Based on the collaborative sensing results, a multi-dimensional fault diagnosis decision tree is constructed. Combined with the weight allocation and confidence assessment of each sensor data, the specific device location where the fault occurs is accurately located, and the assessment results of the fault type and severity are given.

[0046] Based on the multimodal feature fusion matrix F and the attention weight matrix As input, by calculating the feature importance score Establish the basis for feature selection. Then construct the feature importance ranking set T = { , ..., The calculation of feature importance scores comprehensively considers the dynamic weight of the attention mechanism and the information content of the original features, making feature selection more targeted.

[0047] Based on the feature importance ranking set T, the information gain ratio criterion IG is used to determine the optimal split feature: .in, is the information entropy of feature set T, Features Conditional entropy under certain conditions. The distinguishing ability of features is evaluated by the change in information entropy, and the most distinguishing feature is selected as the node splitting condition. It can adaptively construct decision rules to improve the rationality of the tree structure and the diagnostic efficiency.

[0048] Furthermore, a multi-dimensional fault diagnosis decision tree D is constructed, where each internal node Include feature splitting threshold and weight , leaf nodes Storage Failure Type and failure probability For internal nodes in a decision tree , based on the feature vector and split threshold To perform a binary split: When the characteristic value Less than or equal to the split threshold When , the sample is assigned to the left child node; When the characteristic value Greater than the split threshold , the sample is assigned to the right child node.

[0049] The path from the root node to the leaf node corresponds to a complete fault judgment chain, which not only ensures the interpretability of the diagnosis process but also maintains a high computational efficiency.

[0050] Based on the optimized sensor weights and decision tree D, the fault location mapping relationship is established by calculating the device fault location matrix M. The fault location matrix M is expressed as: , in, Indicates the device index, Indicates the fault type index, is the failure probability of the leaf node, is the optimized sensor weight.

[0051] The fault location matrix M provides a probability distribution diagram of device-faults, which can intuitively reflect the possible location and type of faults and provide maintenance personnel with accurate fault location information.

[0052] Combining the fault location matrix M and the residual feature R in S2, a diagnosis confidence evaluation model is constructed, which is expressed as: , in, and is the balance coefficient, Represents the L2 norm of the residual feature.

[0053] The confidence evaluation comprehensively considers the fault probability and residual characteristics, and provides a quantitative indicator of the reliability of the diagnosis result. The influence of the residual characteristics on the confidence is made nonlinear, which is more in line with the characteristics of actual fault diagnosis.

[0054] Based on the fault location matrix M and the confidence C, the fault severity evaluation function E is constructed, which is expressed as: , in, The severity coefficient corresponding to the fault type outputs the fault severity level, including minor, moderate and severe.

[0055] Based on the fault severity evaluation function E, a first threshold is set and the second threshold Determine the fault level, and < , If E ≤ T1, it is judged as a minor fault, which means that the fluctuation of core function parameters is within the allowable range, the performance of key components has not declined significantly, the insulation performance remains stable, the temperature fluctuation amplitude is less than the specified threshold, and the partial discharge is within the reference value range. It is necessary to record the fault characteristics and continuously monitor, regularly check the fault-related components, and maintain the regular maintenance cycle.

[0056] If T1 < E ≤ T2, it is judged as a moderate fault, indicating that the core function parameters have fluctuated significantly, the performance of key components has begun to decline, the insulation performance has shown a decay trend, the partial discharge has exceeded the warning value, and the mechanical characteristic parameters have shifted. It is necessary to increase the fault monitoring frequency, formulate a special maintenance plan, and prepare necessary spare parts replacement.

[0057] If E > T2, it is judged as a serious fault, indicating that the core function parameters seriously deviate from the normal values, the performance of key components has significantly deteriorated, the insulation performance has dropped sharply, the temperature has continued to rise abnormally, and the local discharge has reached a dangerous level. It is necessary to immediately activate the emergency response plan, arrange emergency maintenance or replacement, and evaluate the necessity of shutdown maintenance.

[0058] In summary, the present invention realizes a complete technical chain from data collection, feature learning to fault diagnosis, and establishes a systematic method for distribution cabinet status evaluation and life prediction. It has significant engineering application value, can accurately identify potential equipment failures, and warn of possible operating risks in advance, providing a scientific basis for preventive maintenance. It can effectively reduce equipment failure rate, extend service life, reduce maintenance costs, improve power supply reliability, and promote the development of power systems towards intelligent operation and maintenance. It is of great significance to improve the operational reliability and maintenance efficiency of the distribution system, and can provide new technical ideas and solutions for the field of power equipment status monitoring and intelligent diagnosis.

[0059] This embodiment further provides a distribution cabinet fault detection system, including: The collection module is used to deploy a multimodal sensor array, build a multi-dimensional data collection network, collect various sensor data in real time, and perform data preprocessing; The analysis module uses deep learning algorithms to establish a multimodal data fusion model, and forms a fault feature mapping relationship through feature extraction, data correlation analysis and cross-validation to achieve collaborative perception at the data level; The detection module builds a multi-dimensional fault diagnosis decision tree through collaborative perception results, combines the weight distribution and confidence assessment of each sensor data, accurately locates the specific device location where the fault occurs, and gives an assessment result of the fault type and severity.

[0060] This embodiment also provides a computer device, which is suitable for the distribution cabinet fault detection method, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the distribution cabinet fault detection method proposed in the above embodiment.

[0061] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0062] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for detecting faults in a power distribution cabinet as proposed in the above embodiment is implemented.

[0063] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0064] Embodiment 2: This embodiment provides a method for detecting faults in a power distribution cabinet. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0065] In order to verify the effectiveness of the method of the present invention, 100 10kV distribution cabinets with an operating time of more than 10 years in a substation were selected as experimental objects, and their operating data within one year were collected for analysis.

[0066] The experimental setup is as follows: Operating temperature: -10℃~45℃; Relative humidity: 20% to 90%; Data sampling frequency: electrical characteristics 100Hz, temperature characteristics 1Hz, acoustic characteristics 44.1kHz, mechanical characteristics 10Hz; The ratio of training set to test set is 7:3; The comparison methods include: Method A: Traditional threshold diagnostic method using only electrical characteristics; Method B: BP neural network method based on temperature and electrical characteristics; Method C: Multi-feature fusion method using a single deep learning model; The method of the present invention: an intelligent diagnosis method of multimodal data fusion.

[0067] The simulated fault types include: poor contact, insulation degradation, mechanical looseness, partial discharge and other typical faults. The experimental results are shown in Table 1: Table 1 Performance comparison results of different fault diagnosis methods

[0068] Under different fault types, the recognition accuracy of each method is shown in Table 2: Table 2 Identification accuracy of different fault types

[0069] From Table 1, it can be found that the method of the present invention is significantly superior to traditional methods and other intelligent diagnosis methods in key indicators such as fault detection rate, false alarm rate, and missed detection rate. In particular, the fault detection rate reached 96.8%, an increase of 18.5 percentage points compared with method A; the false alarm rate and missed detection rate dropped to 3.5% and 4.2% respectively, reflecting a high diagnostic reliability. In terms of fault location accuracy, the method of the present invention achieved an excellent performance of 94.3%, and the early warning can be as long as 8-12 days, leaving sufficient preparation time for equipment maintenance.

[0070] Although the diagnosis time is slightly higher than that of the simple threshold judgment method, it is significantly lower than that of method C, showing good real-time performance. In addition, the confidence evaluation accuracy of the method of the present invention reaches 92.7%, providing a reliable credibility guarantee for the fault diagnosis results.

[0071] Furthermore, combined with the contents of Table 2, from the perspective of the recognition effect of different fault types, the recognition accuracy of the method of the present invention for various typical faults is maintained at above 93%, among which the recognition effect for temperature anomaly and poor contact is the most outstanding, with accuracy rates reaching 96.2% and 95.8% respectively. In contrast, the traditional method A performed the worst in mechanical loose fault recognition, only 68.4%, indicating that a single feature is difficult to effectively identify complex faults. Although methods B and C have improved the recognition effect by introducing multi-feature analysis, they still cannot reach the performance level of the method of the present invention.

[0072] This result fully proves that the multimodal data fusion method proposed in this invention has significant advantages in complex fault identification and can comprehensively and accurately identify various types of distribution cabinet faults.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting a fault in a power distribution cabinet, characterized in that: include: Deploy a multi-modal sensor array, build a multi-dimensional data acquisition network, collect various sensor data in real time, including temperature data, current data, vibration acceleration data, and partial discharge signal data, and pre-process the collected data; A multimodal data fusion model is established using deep learning algorithms. Through feature extraction, data correlation analysis and cross-validation, a fault feature mapping relationship is formed to achieve collaborative perception at the data level. Based on the collaborative sensing results, a multi-dimensional fault diagnosis decision tree is constructed. Combined with the weight distribution and confidence assessment of each sensor data, the specific device location where the fault occurs is accurately located, and the assessment results of the fault type and severity are given.

2. The method for detecting a fault in a power distribution cabinet according to claim 1, wherein: By The temperature sensors are evenly arranged along the busbar inside the power distribution cabinet, and the distance between adjacent temperature sensors is Satisfy the formula ,in, is the total length of the busbar, is the number of temperature sensors, collecting the temperature data ; Vibration sensors are respectively set at the positions of the temperature sensors to collect the vibration acceleration data , vibration acceleration data Sampling frequency Satisfy the formula ,in, is the sampling time interval; The Hall current sensor is set at the incoming line end of the power distribution cabinet to collect the current data , current data Satisfy the formula ,in, is the sensitivity coefficient of the Hall element, is the Hall voltage, the current data With the temperature data Synchronous sampling; An ultrasonic partial discharge sensor is arranged on the top of the distribution cabinet to collect the partial discharge signal D. The sampling frequency Satisfy the formula .

3. The method for detecting a fault in a power distribution cabinet according to claim 2, wherein: The preprocessing includes data synchronization and noise reduction processing; The data synchronization is based on timestamp alignment, and uses a linear interpolation method to process data with inconsistent sampling frequencies: ， Where Y represents any sensor data, are adjacent sampling moments, is the time to be interpolated; The sensor data matrix M is generated based on the synchronized data, and the sensor data matrix M satisfies: ， in, is the temperature data time series, is the vibration acceleration data time series, is the current data time series, is the time series of partial discharge signal, and the sampling frequency is 4 times of other signals; is the number of sampling points; Perform db4 wavelet basis decomposition on each row of the signal sequence in the sensor data matrix M to satisfy the formula: ， in, is the input signal sequence, is the db4 wavelet basis function, To decompose the scale, is the translation parameter; Using 3-layer wavelet decomposition, we can obtain the coefficients of each scale, which can be expressed as: ， ， in, For the Layer low frequency coefficients, is the high frequency coefficient of the j+1th layer, is the low-pass filter coefficient, is the high-pass filter coefficient, is the filter length of db4 wavelet; Soft threshold processing is performed on high-frequency coefficients, and the threshold function is: ， in, is the adaptive threshold, Represents the high-frequency coefficients obtained by wavelet decomposition of each layer; Adaptive thresholding using maximum likelihood estimation , expressed as: ， in, is the noise standard deviation, is the signal length, is the high frequency coefficient of the first layer, is the median operator; Furthermore, the processed coefficients are used for wavelet reconstruction to obtain the denoised signal: ， in, To reconstruct the low-pass filter coefficients, To reconstruct the high-pass filter coefficients, Indicates rounding down; Reassemble the reconstructed row signals to obtain the preprocessed sensor data matrix M1: ， in, , , , They are the denoised temperature, vibration, current and partial discharge signal sequences respectively.

4. The method for detecting a fault in a power distribution cabinet according to claim 3, wherein: The feature extraction formula is expressed as: ， Among them, M1 is the pre-processed sensor data matrix, which contains four types of signals: temperature, vibration, current, and partial discharge; is the generalized time-frequency kernel matrix; Furthermore, the kernel matrix It is expressed as: ， in, , , are the wavelet basis functions of temperature, vibration and partial discharge signals respectively. represents the conjugate complex number of the wavelet basis function, , , is the corresponding scale parameter; is the time variable, is the frequency variable, is the integration variable; In view of the difference in the contribution of different modal features to fault diagnosis, adaptive feature fusion is performed through the attention mechanism, which is expressed as: ， ， in, For the The query vector is The attention weight value of the key vector, is the query vector, representing the focus of the target feature, For the The key vector of features represents the content of the features to be fused. For the The attention weight of each modality feature, For the The characteristics of a mode, is the feature dimension, is the number of modes, is the feature representation after fusion; In order to ensure the training stability and feature transfer efficiency of the deep network, the residual learning mechanism is introduced: ， Among them, R is the feature after residual enhancement, is the network mapping function, which includes multiple layers of convolution and activation operations. The input features are directly passed to the output through the skip connection.

5. The method for detecting a fault in a power distribution cabinet according to claim 4, wherein: Based on the extracted data features, a multi-task learning architecture is constructed to optimize the fault diagnosis and life prediction tasks simultaneously. The joint loss function is expressed as: ， ， ， in, is the total loss function, is the cross entropy loss, is the mean square error loss, used for life prediction tasks, is the true label of the fault type, is the predicted probability of failure type, is the actual remaining life, To predict the remaining life, is the task weight coefficient, determined by cross-validation, is the number of fault categories, is the sample size; Furthermore, the correlation strength between the feature pattern and the fault type is described by the feature mapping of similarity measurement, which is expressed as: ， ， ， Where M is the fault feature mapping matrix, Indicates Class characteristics and The mapping strength of class faults, is the feature weight coefficient, reflecting the importance of the feature, For the Class characteristics Quantity, For the The feature template of the class fault, σ is the Gaussian kernel parameter, which controls the mapping smoothness, and K is the feature dimension.

6. The method for detecting a fault in a power distribution cabinet according to claim 5, wherein: Based on the multimodal feature fusion matrix F and the attention weight matrix As input, by calculating the feature importance score Establish the basis for feature selection and construct the feature importance ranking set T = { , ..., }; Based on the feature importance ranking set T, the information gain ratio criterion IG is used to determine the optimal split feature: ,in, is the information entropy of feature set T, Features Conditional entropy under conditions; Furthermore, the feature with the maximum information gain ratio is selected as the node splitting condition to construct a multi-dimensional fault diagnosis decision tree D. Each internal node Include feature splitting threshold and weight , leaf nodes Storage Failure Type and failure probability , for internal nodes in a decision tree , based on the feature vector and split threshold To perform a binary split: When the characteristic value Less than or equal to the split threshold When , the sample is assigned to the left child node; When the characteristic value Greater than the split threshold , the sample is assigned to the right child node.

7. The method for detecting a fault in a power distribution cabinet according to claim 6, wherein: Based on the decision tree D and the optimized sensor weights, a fault location mapping relationship is established by calculating the device fault location matrix M; wherein the fault location matrix M is expressed as: ， in, Indicates the device index, Indicates the fault type index, is the failure probability of the leaf node, is the optimized sensor weight; Combining the fault location matrix M and the residual feature R, a diagnosis confidence evaluation model is constructed to provide a quantitative indicator of the reliability of the diagnosis result, which is expressed as: ， in, and is the balance coefficient, Represents the L2 norm of the residual feature; Based on the fault location matrix M and the confidence C, the fault severity evaluation function E is constructed, which is expressed as: ， in, is the severity coefficient corresponding to the fault type, and outputs the fault severity level.

8. A distribution cabinet fault detection system, based on the distribution cabinet fault detection method according to any one of claims 1 to 7, characterized in that: include: The collection module is used to deploy a multimodal sensor array, build a multi-dimensional data collection network, collect various sensor data in real time, and perform data preprocessing; The analysis module uses deep learning algorithms to establish a multimodal data fusion model, and forms a fault feature mapping relationship through feature extraction, data correlation analysis and cross-validation to achieve collaborative perception at the data level; The detection module builds a multi-dimensional fault diagnosis decision tree through collaborative perception results, combines the weight distribution and confidence assessment of each sensor data, accurately locates the specific device location where the fault occurs, and gives an assessment result of the fault type and severity.

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

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

Images