A low voltage switch cabinet fault diagnosis method and system

Abstract

The present invention discloses a low-voltage switch cabinet fault diagnosis method and system, and relates to the technical field of power equipment fault detection. A low-voltage switch cabinet fault diagnosis system includes: a fault diagnosis module and a fault prediction module. The present invention can flexibly adapt to different working environments and fault modes through real-time monitoring and dynamic time interval adjustment, ensuring that potential fault information is captured in time at critical moments; dual-channel analysis of fault data and sensor monitoring data, combined with multi-scale convolution and bidirectional time series feature extraction, can deeply mine the spatial and temporal features in the data and improve the accuracy of diagnosis; using attention mechanism and residual enhancement technology, the expression ability of features is further improved, ensuring the efficiency and reliability of fault diagnosis results; through the continuously optimized low-voltage switch cabinet fault prediction model, efficient prediction of faults is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault detection, and in particular to a low-voltage switch cabinet fault diagnosis method and system. Background Art

[0002] Low-voltage switchgear is a key device for distributing, controlling and protecting electric energy in power systems and is widely used in industrial, commercial and civil power distribution systems. With the growth of power demand and the rapid development of smart grid technology, the performance and reliability requirements of low-voltage switchgear are becoming higher and higher. However, the existing low-voltage switchgear fault diagnosis technology faces several key challenges: First, the application of a single sensor often reduces the accuracy of partial discharge detection due to environmental factors, such as electromagnetic interference. Electromagnetic interference will affect the sensor signal, causing data distortion or noise, affecting the accuracy of fault diagnosis, and thus increasing the difficulty of fault detection. Second, the traditional alarm mechanism based on fixed thresholds has a high false alarm rate, which not only increases maintenance costs, but also may lead to the neglect of actual faults. Under this mechanism, the set threshold may not be able to adapt to the changing working environment, resulting in normal fluctuations being misjudged as faults, thereby triggering unnecessary alarms or missing real fault signals. Finally, the existing technology lacks the ability to predict the trend of equipment status and cannot warn potential faults in advance. Faults usually have a certain evolution process. If the trend of equipment status changes cannot be effectively monitored and predicted, it will be difficult to achieve preventive maintenance, resulting in the failure to take timely repair measures before the equipment fails.

[0003] Therefore, a fault diagnosis method is needed that can improve the accuracy of fault detection, reduce false alarms, and have trend prediction function, so as to effectively improve the operation and maintenance efficiency and reliability of low-voltage switchgear. Summary of the invention

[0004] The present invention aims to provide a low-voltage switch cabinet fault diagnosis method and system to improve the accuracy of fault detection.

[0005] A low voltage switch cabinet fault diagnosis method comprises the following steps:

[0006] When a low-voltage switch cabinet fault signal is received, current low-voltage switch cabinet fault data and a low-voltage switch cabinet sensor monitoring data set are obtained; the low-voltage switch cabinet sensor monitoring data set includes N low-voltage switch cabinet sensor monitoring data;

[0007] The current low-voltage switch cabinet fault data includes the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data includes the temperature and humidity data of the low-voltage switch cabinet, the vibration spectrum data of the low-voltage switch cabinet and the switch status data of the low-voltage switch cabinet;

[0008] The low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval. When a low-voltage switch cabinet fault signal is received, the current time node is taken as the origin to obtain the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data to form a low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received;

[0009] Based on the current low-voltage switchgear fault data, low-voltage switchgear sensor monitoring data set and low-voltage switchgear fault diagnosis model, the low-voltage switchgear fault diagnosis results are obtained; the low-voltage switchgear fault repair and subsequent operations are performed on the low-voltage switchgear according to the low-voltage switchgear fault diagnosis results; the low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis;

[0010] Among them, the dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model realizes the prediction and response to low-voltage switchgear faults by dynamically adjusting the time interval and continuously optimizing the prediction model.

[0011] As a preferred technical solution of the present invention, the low-voltage switch cabinet fault diagnosis model includes a data preprocessing layer, a dual-channel analysis layer and a result output layer;

[0012] The data preprocessing layer is used to preprocess and classify the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set to obtain the preprocessed low-voltage switch cabinet fault data and the preprocessed low-voltage switch cabinet sensor monitoring data set; at the same time, the first channel label is added to the preprocessed low-voltage switch cabinet fault data, and the second channel label is added to the preprocessed low-voltage switch cabinet sensor monitoring data set;

[0013] The dual-channel analysis layer is used to synchronously analyze the pre-processed low-voltage switchgear fault data and the pre-processed low-voltage switchgear sensor monitoring data set to obtain the low-voltage switchgear fault diagnosis results;

[0014] The result output layer is used to output the fault diagnosis results of low-voltage switchgear.

[0015] As a preferred technical solution of the present invention, the specific steps of performing synchronous analysis in the dual-channel analysis layer include:

[0016] The dual-channel analysis layer includes a data input layer, a first channel layer, a second channel layer and a feature fusion layer;

[0017] Wherein, the first channel layer and the second channel layer are parallel structures;

[0018] In the data input layer, the pre-processed low-voltage switch cabinet fault data is input into the first channel layer based on the first channel label; the pre-processed low-voltage switch cabinet sensor monitoring data set is input into the second channel layer based on the second channel label;

[0019] In the first channel layer, convolution feature extraction is performed on the pre-processed low-voltage switch cabinet fault data to obtain the convolution feature of the low-voltage switch cabinet fault;

[0020] In the second channel layer, the time series feature extraction is performed on the pre-processed low-voltage switch cabinet sensor monitoring data set to obtain the low-voltage switch cabinet monitoring time series feature;

[0021] In the feature fusion layer, the specific steps are:

[0022] The convolution feature of the low-voltage switch cabinet fault is flattened to obtain the two-dimensional convolution feature Z of the low-voltage switch cabinet fault; the low-voltage switch cabinet monitoring time series feature is mapped through the fully connected layer to obtain the two-dimensional low-voltage switch cabinet monitoring time series feature S;

[0023] Using the formula The fault diagnosis characteristic matrix A of the low-voltage switch cabinet is calculated; is the transposition of the two-dimensional low-voltage switchgear monitoring time series feature S, and C is the scaling factor;

[0024] Based on the two-dimensional low-voltage switchgear fault convolution feature Z and the low-voltage switchgear fault diagnosis feature matrix A, a bidirectional feature interaction is performed to obtain the interactive low-voltage switchgear fault convolution feature; the two-dimensional low-voltage switchgear fault convolution feature Z and the interactive low-voltage switchgear fault convolution feature are residually connected to obtain the enhanced low-voltage switchgear fault convolution feature;

[0025] Based on the two-dimensional low-voltage switchgear monitoring time series feature S and the low-voltage switchgear fault diagnosis feature matrix A, a two-dimensional low-voltage switchgear monitoring time series feature S and an interactive low-voltage switchgear monitoring time series feature are subjected to a residual connection to obtain an enhanced low-voltage switchgear monitoring time series feature.

[0026] The sigmoid function is used to perform weighted feature fusion and pooling on the enhanced low-voltage switchgear fault convolution features and the enhanced low-voltage switchgear monitoring time series features to obtain the fused low-voltage switchgear fault diagnosis feature R;

[0027] Using the formula Calculate the fault diagnosis classification results and obtain the low-voltage switchgear fault diagnosis results; where MLP is the trained multi-layer perceptron model and softmax is the activation function.

[0028] As a preferred technical solution of the present invention, the specific steps of performing convolution feature extraction in the first channel layer and performing temporal feature extraction in the second channel layer include:

[0029] The first channel layer is constructed based on the CNN model, which includes J1 3*3 convolution kernels, J2 5*5 convolution kernels and SE modules;

[0030] Among them, all convolution kernels are used to perform multi-scale convolution operations on the pre-processed low-voltage switchgear fault data, and the SE module is used to enhance important features using the attention mechanism;

[0031] The second channel layer is constructed based on the bidirectional GRU model; forward propagation and backward propagation are used to perform bidirectional feature extraction to obtain the bidirectional hidden time series of low-voltage switchgear monitoring; feature weighted aggregation is performed on the bidirectional hidden time series of low-voltage switchgear monitoring to obtain the timing characteristics of low-voltage switchgear monitoring.

[0032] As a preferred technical solution of the present invention, the low-voltage switch cabinet predictive fault analysis model includes a data analysis layer, a dynamic prediction layer, a time interval change layer and a prediction result output layer;

[0033] The data analysis layer is used to obtain the first M low-voltage switchgear sensor monitoring data to form a low-voltage switchgear prediction fault analysis data set;

[0034] The dynamic prediction layer is used to perform risk prediction analysis on the low-voltage switchgear prediction fault analysis data set to obtain the low-voltage switchgear prediction fault risk feature value;

[0035] The time interval variation layer is used to perform data analysis based on the fault risk characteristic value predicted by the low-voltage switchgear to obtain the dynamic time interval;

[0036] The prediction result output layer is used to output dynamic time intervals.

[0037] As a preferred technical solution of the present invention, the specific steps of performing data analysis in the time interval variation layer include:

[0038] Based on the low-voltage switchgear predicted fault risk characteristic value and the current dynamic time interval, K changing time interval individuals G are constructed. k , k=1, 2, ..., K; each change time interval individual contains a simulated dynamic time interval for updating; K change time interval individuals G k Combine, get the iterative population with variable time interval, and set the maximum number of iterations;

[0039] Using the formula F=W1*(X(T post ) / T post -Y(T pre ) / T pre ) 2 +W2*(T post -T pre )2 +W3*max(0,(T min -T post ) 2 Calculate the dynamically changing objective function F;

[0040] Among them, T post is the current dynamic time interval, T pre Individual G for different time intervals of change k The corresponding simulation dynamic time interval, X(T post ) represents the predicted fault risk characteristic value of the low-voltage switchgear, Y(T pre ) indicates that the dynamic time interval is updated to T pre Failure trend value after, T min is the minimum dynamic sampling time interval; W1, W2 and W3 are weight coefficients;

[0041] The change time interval individual G k The corresponding calculated dynamic change objective function F is used as the fitness D k ;

[0042] In the population iteration process, the local optimal position sharing and dynamic neighborhood partitioning mechanism are used to perform population iteration mutation on the iterative population with varying time intervals, thereby completing the update operation on the iterative population with varying time intervals.

[0043] When the maximum number of iterations is reached, the change time interval individual corresponding to the maximum fitness is output, which is the optimal change time interval individual; based on the optimal change time interval individual, the current dynamic time interval is updated to obtain the dynamic time interval.

[0044] As a preferred technical solution of the present invention, the specific steps of training the dynamic prediction layer include:

[0045] Collect several groups of fault prediction training samples, each group of fault prediction training samples contains predicted fault label values ​​and verified low-voltage switchgear fault analysis data sets; combine several groups of fault prediction training samples to obtain a fault prediction training set;

[0046] The dynamic prediction layer is trained based on the fault prediction training set to obtain the initial dynamic prediction layer; the initial dynamic prediction layer is evaluated, and if the initial dynamic prediction layer passes the model evaluation, the initial dynamic prediction layer is used as the dynamic prediction layer in the low-voltage switchgear fault prediction analysis model; otherwise, the fault prediction training set is used to continue model training.

[0047] A low voltage switch cabinet fault diagnosis system, comprising:

[0048] A fault diagnosis module includes a data acquisition unit and a fault diagnosis unit;

[0049] The data acquisition unit is used to acquire the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set when receiving the low-voltage switch cabinet fault signal; the low-voltage switch cabinet sensor monitoring data set contains N low-voltage switch cabinet sensor monitoring data; the current low-voltage switch cabinet fault data contains the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data contains the low-voltage switch cabinet temperature and humidity data, the low-voltage switch cabinet vibration spectrum data and the low-voltage switch cabinet switch state data; the low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval, and when the low-voltage switch cabinet fault signal is received, the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data are acquired with the current time node as the origin to form the low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received;

[0050] The fault diagnosis unit is used to analyze the current low-voltage switchgear fault data, the low-voltage switchgear sensor monitoring data set and the low-voltage switchgear fault diagnosis model to obtain the low-voltage switchgear fault diagnosis result; perform fault repair and subsequent operations on the low-voltage switchgear according to the low-voltage switchgear fault diagnosis result; the low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis;

[0051] A fault prediction module, including a fault prediction analysis unit;

[0052] The fault prediction analysis unit is used to update the dynamic time interval; wherein the dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model realizes the prediction and response to the low-voltage switchgear fault by dynamically adjusting the time interval and continuously optimizing the prediction model.

[0053] The present invention has the following advantages:

[0054] 1. Through real-time monitoring and dynamic time interval adjustment, the present invention can flexibly adapt to different working environments and fault modes, ensuring that potential fault information is captured in time at critical moments; dual-channel analysis of fault data and sensor monitoring data, combined with multi-scale convolution and bidirectional time series feature extraction, can deeply mine the spatial and temporal features in the data and improve the accuracy of diagnosis; the use of attention mechanism and residual enhancement technology further improves the expression ability of features and ensures the efficiency and reliability of fault diagnosis results; through the continuously optimized low-voltage switchgear fault prediction model, efficient prediction of faults is achieved, and rapid response is achieved based on the diagnosis results, which reduces the downtime of equipment failures and extends the service life of the equipment.

[0055] 2. The present invention can simultaneously process the electrical variable data and sensor monitoring data of the low-voltage switchgear through a dual-channel structure; the first channel focuses on the convolution feature extraction of the electrical variable data, and the second channel focuses on the extraction of timing features. The combination of the two can capture the key information in different types of data, thereby achieving more accurate fault diagnosis; through the convolution and timing feature extraction layers, the model can deeply analyze the different features in the low-voltage switchgear and effectively fuse the two types of features. The enhanced low-voltage switchgear fault convolution features and monitoring timing features are connected through residuals, which strengthens the expression ability of key features and avoids the loss or excessive compression of information, thereby improving the robustness and accuracy of the diagnosis; using the weighted fusion mechanism in the feature fusion layer, combined with the Sigmoid function and pooling operations, the model can effectively integrate different information from the two channels and provide accurate input for subsequent fault classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention is a schematic diagram of the structure of a low-voltage switch cabinet fault diagnosis system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0058] Embodiment 1, a low voltage switch cabinet fault diagnosis method, comprising the following steps:

[0059] When a low-voltage switch cabinet fault signal is received, the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set are obtained; the low-voltage switch cabinet sensor monitoring data set contains N low-voltage switch cabinet sensor monitoring data; the specific value of N is set by professional technicians according to actual conditions;

[0060] The low-voltage switch cabinet fault signal is a device fault or disconnection signal. The current low-voltage switch cabinet fault data includes the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data includes the low-voltage switch cabinet temperature and humidity data, the low-voltage switch cabinet vibration spectrum data and the low-voltage switch cabinet switch status data;

[0061] The electrical variable data of low-voltage switchgear includes monitoring of electrical parameters such as current, voltage, and power. These data help monitor the electrical performance of the switchgear in real time and promptly detect abnormal conditions in the electrical system, such as overload, short circuit, and ground fault. By monitoring the changes in parameters such as current and voltage, the electrical variable data can detect abnormalities (such as current fluctuations, overload, and frequency instability) before a fault occurs, providing guidance for subsequent repairs and maintenance. It helps analyze the distribution and change trends of the load, thereby optimizing load management and avoiding overload or energy efficiency issues. These electrical variable data usually come from devices such as current sensors, voltage sensors, and power sensors, and can provide real-time feedback on the electrical status inside the switchgear.

[0062] The temperature and humidity data of low-voltage switchgear are used to monitor environmental conditions to avoid damage to equipment caused by high temperature or high humidity. High temperature may cause electrical components to overheat, resulting in failures, while excessive humidity will affect the insulation performance of the equipment and increase the risk of equipment damage. The temperature and humidity sensors provide environmental data by monitoring the temperature and humidity inside and around the switchgear in real time to help ensure that the equipment operates under appropriate conditions; vibration spectrum data helps monitor the operating status of mechanical components in the switchgear, especially switches, relays, circuit breakers and other components. Abnormal vibrations may indicate that the equipment is worn, loose, or about to fail. By analyzing the vibration spectrum, predictive maintenance can be performed to detect potential problems in a timely manner and reduce equipment downtime; vibration data is usually collected through accelerometers and vibration analyzers to determine mechanical failures;

[0063] The switch status data of the low-voltage switchgear is used to monitor the operating status of the switchgear. By monitoring the open and closed status of the switch, it is possible to promptly detect whether the circuit breaker or switch has an erroneous operation or failure, thereby ensuring the normal operation of the equipment and avoiding a wider range of failures. The switch status data is usually provided by a switch status sensor or a PLC control system.

[0064] The low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval. When a low-voltage switch cabinet fault signal is received, the current time node is taken as the origin to obtain the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data to form a low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received;

[0065] Based on the current low-voltage switchgear fault data, low-voltage switchgear sensor monitoring data set and low-voltage switchgear fault diagnosis model, the low-voltage switchgear fault diagnosis results are obtained; the low-voltage switchgear fault repair and subsequent operations are performed on the low-voltage switchgear according to the low-voltage switchgear fault diagnosis results; the low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis;

[0066] By continuously monitoring the electrical variable data, temperature and humidity data, vibration spectrum data and switch status data of the low-voltage switchgear, and acquiring the data before and after the fault signal is received, the system can accurately capture the dynamic changes before and after the fault occurs. This time-node-based data collection method can improve the timeliness and accuracy of fault diagnosis.

[0067] Through the dual-channel structure, multi-scale convolution and time series feature extraction are performed on fault data and sensor monitoring data respectively, which can fully explore the deep-level rules and implicit information in the data. At the same time, combined with the attention weighted fusion and residual enhancement mechanism, the diagnostic ability of the model is improved, making the diagnosis of various types of faults more accurate and efficient;

[0068] The low-voltage switchgear fault diagnosis model includes a data preprocessing layer, a dual-channel analysis layer and a result output layer;

[0069] The data preprocessing layer is used to preprocess and classify the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set to obtain the preprocessed low-voltage switch cabinet fault data and the preprocessed low-voltage switch cabinet sensor monitoring data set; at the same time, the first channel label is added to the preprocessed low-voltage switch cabinet fault data, and the second channel label is added to the preprocessed low-voltage switch cabinet sensor monitoring data set;

[0070] The dual-channel analysis layer is used to synchronously analyze the pre-processed low-voltage switchgear fault data and the pre-processed low-voltage switchgear sensor monitoring data set to obtain the low-voltage switchgear fault diagnosis results;

[0071] The result output layer is used to output the fault diagnosis results of the low-voltage switchgear;

[0072] The specific steps for performing synchronous analysis in the dual-channel analysis layer include:

[0073] The dual-channel analysis layer includes a data input layer, a first channel layer, a second channel layer and a feature fusion layer;

[0074] Wherein, the first channel layer and the second channel layer are parallel structures;

[0075] In the data input layer, the pre-processed low-voltage switch cabinet fault data is input into the first channel layer based on the first channel label; the pre-processed low-voltage switch cabinet sensor monitoring data set is input into the second channel layer based on the second channel label;

[0076] In the first channel layer, convolution feature extraction is performed on the pre-processed low-voltage switch cabinet fault data to obtain the convolution feature of the low-voltage switch cabinet fault;

[0077] In the second channel layer, the time series feature extraction is performed on the pre-processed low-voltage switch cabinet sensor monitoring data set to obtain the low-voltage switch cabinet monitoring time series feature;

[0078] In the feature fusion layer, the specific steps are:

[0079] The convolution feature of the low-voltage switch cabinet fault is flattened to obtain the two-dimensional convolution feature Z of the low-voltage switch cabinet fault; the low-voltage switch cabinet monitoring time series feature is mapped through the fully connected layer to obtain the two-dimensional low-voltage switch cabinet monitoring time series feature S;

[0080] Using the formula The fault diagnosis characteristic matrix A of the low-voltage switch cabinet is calculated; is the transposition of the two-dimensional low-voltage switchgear monitoring time series feature S, and C is the scaling factor; the scaling factor is set by professional technicians according to actual conditions;

[0081] Based on the two-dimensional low-voltage switchgear fault convolution feature Z and the low-voltage switchgear fault diagnosis feature matrix A, a bidirectional feature interaction is performed to obtain the interactive low-voltage switchgear fault convolution feature; the two-dimensional low-voltage switchgear fault convolution feature Z and the interactive low-voltage switchgear fault convolution feature are residually connected to obtain the enhanced low-voltage switchgear fault convolution feature;

[0082] Based on the two-dimensional low-voltage switchgear monitoring time series feature S and the low-voltage switchgear fault diagnosis feature matrix A, a two-dimensional low-voltage switchgear monitoring time series feature S and an interactive low-voltage switchgear monitoring time series feature are subjected to a residual connection to obtain an enhanced low-voltage switchgear monitoring time series feature.

[0083] The sigmoid function is used to perform weighted feature fusion and pooling on the enhanced low-voltage switchgear fault convolution features and the enhanced low-voltage switchgear monitoring time series features to obtain the fused low-voltage switchgear fault diagnosis feature R;

[0084] Using the formula Calculate the fault diagnosis classification results and obtain the low-voltage switchgear fault diagnosis results; where MLP is the trained multi-layer perceptron model and softmax is the activation function;

[0085] Through the dual-channel structure, the model can simultaneously process the electrical variable data and sensor monitoring data of the low-voltage switchgear; the first channel focuses on the convolution feature extraction of the electrical variable data, and the second channel focuses on the extraction of the timing features. The combination of the two can capture the key information in different types of data, thereby achieving more accurate fault diagnosis; through the convolution and timing feature extraction layers, the model can deeply analyze the different features in the low-voltage switchgear and effectively fuse the two types of features. The enhanced low-voltage switchgear fault convolution features and monitoring timing features are connected through residuals, which strengthens the expression ability of key features and avoids the loss or excessive compression of information, thereby Improves the robustness and accuracy of diagnosis; by using the weighted fusion mechanism in the feature fusion layer, combined with the Sigmoid function and pooling operation, the model can effectively integrate different information from the two channels and provide accurate input for subsequent fault classification. This weighted fusion method ensures that the model can give priority to important features while suppressing noise and irrelevant features, thereby improving the overall diagnostic effect; through residual connections and bidirectional feature interactions, the model avoids the problem of gradient disappearance or feature degradation. This enhancement mechanism improves the adaptability and stability of the model in complex environments, enabling it to better handle various abnormal situations and make accurate diagnoses;

[0086] In the fault diagnosis of low-voltage switchgear, the training process of using the multi-layer perceptron (MLP) model includes forward propagation, loss calculation, back propagation and parameter update. First, the fault data and sensor monitoring data of the low-voltage switchgear are input into the model, and the diagnosis results are obtained through forward propagation. Then, the loss is calculated based on the true label and the predicted result, the gradient is calculated through the back propagation algorithm, and the parameters of the model are updated using the gradient descent method. After multiple iterations of optimization, the final model can effectively identify and diagnose the fault type of the low-voltage switchgear, thereby providing accurate support for subsequent fault handling.

[0087] The specific steps of performing convolution feature extraction in the first channel layer and performing temporal feature extraction in the second channel layer include:

[0088] The first channel layer is constructed based on the CNN model, which includes J1 3*3 convolution kernels, J2 5*5 convolution kernels and SE modules;

[0089] Among them, all convolution kernels are used to perform multi-scale convolution operations on the pre-processed low-voltage switchgear fault data, and the SE module is used to enhance important features using the attention mechanism;

[0090] The second channel layer is constructed based on the bidirectional GRU model; bidirectional feature extraction is performed using forward propagation and backward propagation to obtain the bidirectional hidden time series of low-voltage switchgear monitoring; feature weighted aggregation is performed on the bidirectional hidden time series of low-voltage switchgear monitoring to obtain the time series features of low-voltage switchgear monitoring;

[0091] By using convolution kernels of different sizes in the first channel layer, the model can extract detailed features from low-voltage switchgear fault data at multiple scales. This multi-scale convolution operation can more comprehensively capture the diversity of fault signals, improve the accuracy of feature extraction and the ability to identify complex fault modes; the SE module introduces an attention mechanism to dynamically adjust the importance of each feature channel, effectively enhancing the network's attention to key features and suppressing irrelevant information. This enables the model to focus on the most diagnostically valuable features when processing low-voltage switchgear fault data, thereby improving the accuracy of fault diagnosis;

[0092] The second channel layer is based on a bidirectional GRU model, which can simultaneously capture the forward and backward time dependencies in the sequence data, making up for the key information that may be missed by the unidirectional model. This bidirectional feature extraction improves the model's ability to process time series data, especially for low-voltage switchgear monitoring data with time dependencies. By performing feature weighted aggregation on the bidirectional hidden time series of low-voltage switchgear monitoring, the model can comprehensively consider the feature information at different time steps and generate a time series feature with high expressiveness. This weighted aggregation effectively integrates long-term and short-term time series dependencies, improving the accuracy and robustness of the model in time series data analysis.

[0093] The dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model achieves the prediction and response to low-voltage switchgear faults by dynamically adjusting the time interval and continuously optimizing the prediction model;

[0094] The low-voltage switchgear predictive fault analysis model includes a data analysis layer, a dynamic prediction layer, a time interval change layer and a prediction result output layer;

[0095] The data analysis layer is used to obtain the first M low-voltage switchgear sensor monitoring data to form a low-voltage switchgear prediction fault analysis data set;

[0096] The dynamic prediction layer is used to perform risk prediction analysis on the low-voltage switchgear prediction fault analysis data set to obtain the low-voltage switchgear prediction fault risk feature value;

[0097] The specific steps for training the dynamic prediction layer include:

[0098] Collect several groups of fault prediction training samples, each group of fault prediction training samples contains predicted fault label values ​​and verified low-voltage switchgear fault analysis data sets; combine several groups of fault prediction training samples to obtain a fault prediction training set;

[0099] Based on the fault prediction training set, the dynamic prediction layer is trained to obtain an initial dynamic prediction layer; the initial dynamic prediction layer is evaluated, and if the initial dynamic prediction layer passes the model evaluation, the initial dynamic prediction layer is used as the dynamic prediction layer in the low-voltage switchgear prediction fault analysis model; otherwise, the fault prediction training set is used to continue model training;

[0100] The time interval variation layer is used to perform data analysis based on the fault risk characteristic value predicted by the low-voltage switchgear to obtain the dynamic time interval;

[0101] The prediction result output layer is used to output dynamic time intervals;

[0102] By dynamically adjusting the time interval and continuously optimizing the prediction model, the prediction of low-voltage switchgear faults is not only more accurate, but also able to adapt to changes in equipment status in real time, thereby providing a faster fault response. This real-time dynamic prediction capability can greatly shorten the warning time before a fault occurs and improve the efficiency of fault prevention. The dynamic prediction layer can extract accurate fault risk feature values ​​through in-depth analysis of the low-voltage switchgear prediction fault analysis data set. Combined with these features, the model can accurately predict the risk of potential faults, thereby providing a reliable decision-making basis for operation and maintenance personnel and helping them to formulate more accurate maintenance strategies. Through the time interval change layer, the model can adjust the time interval according to the predicted risk feature value, so that data collection and analysis can reflect the operating status of the low-voltage switchgear in real time. This flexible time interval dynamic adjustment mechanism effectively improves the real-time and accuracy of data analysis and ensures that fault prediction can be carried out at the right time.

[0103] The specific steps for data analysis in the time interval change layer include:

[0104] Based on the low-voltage switchgear predicted fault risk characteristic value and the current dynamic time interval, K changing time interval individuals G are constructed. k , k=1, 2, ..., K; each change time interval individual contains a simulated dynamic time interval for updating; K change time interval individuals G k Combine, get the iterative population with variable time interval, and set the maximum number of iterations;

[0105] Using the formula F=W1*(X(T post ) / T post -Y(T pre ) / T pre ) 2 +W2*(T post -T pre ) 2 +W3*max(0,(T min -T post ) 2Calculate the dynamically changing objective function F;

[0106] Among them, T post is the current dynamic time interval, T pre Individual G for different time intervals of change k The corresponding simulation dynamic time interval, X(T post ) represents the predicted fault risk characteristic value of the low-voltage switchgear, Y(T pre ) indicates that the dynamic time interval is updated to T pre Failure trend value after, T min is the minimum dynamic sampling time interval; W1, W2 and W3 are weight coefficients, which are set by professional technicians according to actual conditions; Y() is obtained by fitting historical data; historical data is trained through regression analysis or machine learning model to establish the relationship between dynamic time interval and fault trend. Based on historical fault data, linear regression, support vector machine or neural network methods are used to learn the influence of different time intervals on fault trend, so as to obtain function Y;

[0107] The change time interval individual G k The corresponding calculated dynamic change objective function F is used as the fitness D k ;

[0108] In the population iteration process, the local optimal position sharing and dynamic neighborhood partitioning mechanism are used to perform population iteration mutation on the iterative population with varying time intervals, thereby completing the update operation on the iterative population with varying time intervals.

[0109] When the maximum number of iterations is reached, the change time interval individual corresponding to the maximum fitness is output, which is the optimal change time interval individual; based on the optimal change time interval individual, the current dynamic time interval is updated to obtain the dynamic time interval;

[0110] By constructing multiple individual time intervals and performing iterative optimization, the model can accurately calculate the optimal dynamic time interval, thereby ensuring the real-time and accuracy of low-voltage switchgear data. This time interval adjustment based on the fault risk characteristic value enables data collection and analysis to be more in line with the timing of actual fault occurrence, improving the sensitivity and accuracy of fault prediction; through multiple iterations and dynamic adjustments, the time interval variation layer can continuously optimize the time interval to adapt to different fault trends and equipment conditions. The local optimal position sharing and dynamic neighborhood division mechanism can effectively avoid local optimality during the population iteration process, improve the global optimization ability of the model, and ensure that the most appropriate time interval is finally obtained; this optimization process enables the low-voltage switchgear fault prediction model to automatically adapt to changes in different operating environments and equipment conditions, enhances adaptability, and ensures that accurate prediction results can still be provided in various complex situations;

[0111] For example, in the actual fault diagnosis process, low-voltage switchgear is used to control critical power supply. Assuming that a low-voltage switchgear sensor detects a sharp increase in temperature from 25°C to 40°C and an abnormal change in vibration frequency, the fault diagnosis process is immediately started. First, the latest 100 low-voltage switchgear sensor monitoring data are obtained, of which the first 50 data are used to analyze the current state, and the last 50 data are used to compare and predict possible fault modes; risk prediction is performed through the dynamic time interval optimization model; assuming that the current dynamic time interval is 5 minutes, the formula is used to calculate the dynamic change objective function, considering the fault risk characteristic value and time interval difference, and finally determining the optimal dynamic time interval to be 3 minutes. This adjustment makes fault diagnosis more accurate and can quickly reflect the real-time status of the equipment; output a new dynamic time interval of 3 minutes according to the optimal time interval, and update the fault prediction model in real time. In subsequent operations, the frequency of data collection and analysis is accelerated according to the updated time interval to improve the response speed of fault prediction.

[0112] Example 2, a low voltage switch cabinet fault diagnosis system, see Figure 1 As shown, including:

[0113] A fault diagnosis module includes a data acquisition unit and a fault diagnosis unit;

[0114] The data acquisition unit is used to acquire the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set when receiving the low-voltage switch cabinet fault signal; the low-voltage switch cabinet sensor monitoring data set contains N low-voltage switch cabinet sensor monitoring data; the current low-voltage switch cabinet fault data contains the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data contains the low-voltage switch cabinet temperature and humidity data, the low-voltage switch cabinet vibration spectrum data and the low-voltage switch cabinet switch state data; the low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval, and when the low-voltage switch cabinet fault signal is received, the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data are acquired with the current time node as the origin to form the low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received;

[0115] The fault diagnosis unit is used to analyze the current low-voltage switchgear fault data, the low-voltage switchgear sensor monitoring data set and the low-voltage switchgear fault diagnosis model to obtain the low-voltage switchgear fault diagnosis result; perform fault repair and subsequent operations on the low-voltage switchgear according to the low-voltage switchgear fault diagnosis result; the low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis;

[0116] A fault prediction module, including a fault prediction analysis unit;

[0117] The fault prediction analysis unit is used to update the dynamic time interval; wherein the dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model realizes the prediction and response to the low-voltage switchgear fault by dynamically adjusting the time interval and continuously optimizing the prediction model.

[0118] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A low voltage switch cabinet fault diagnosis method, characterized in that: The following steps are involved: When a low-voltage switch cabinet fault signal is received, current low-voltage switch cabinet fault data and low-voltage switch cabinet sensor monitoring data set are obtained; The low-voltage switch cabinet sensor monitoring data set contains N low-voltage switch cabinet sensor monitoring data; The current low-voltage switch cabinet fault data includes the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data includes the temperature and humidity data of the low-voltage switch cabinet, the vibration spectrum data of the low-voltage switch cabinet and the switch status data of the low-voltage switch cabinet; The low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval. When a low-voltage switch cabinet fault signal is received, the current time node is taken as the origin to obtain the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data to form a low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received; Based on the current low-voltage switchgear fault data, the low-voltage switchgear sensor monitoring data set and the low-voltage switchgear fault diagnosis model, the low-voltage switchgear fault diagnosis result is obtained; According to the fault diagnosis results of the low-voltage switchgear, the low-voltage switchgear is repaired and subsequently operated. The low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis. The dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model achieves the prediction and response to low-voltage switchgear faults by dynamically adjusting the time interval and continuously optimizing the prediction model; The low-voltage switchgear fault diagnosis model includes a data preprocessing layer, a dual-channel analysis layer and a result output layer; The data preprocessing layer is used to preprocess and classify the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set to obtain the preprocessed low-voltage switch cabinet fault data and the preprocessed low-voltage switch cabinet sensor monitoring data set; at the same time, the first channel label is added to the preprocessed low-voltage switch cabinet fault data, and the second channel label is added to the preprocessed low-voltage switch cabinet sensor monitoring data set; The dual-channel analysis layer is used to synchronously analyze the pre-processed low-voltage switchgear fault data and the pre-processed low-voltage switchgear sensor monitoring data set to obtain the low-voltage switchgear fault diagnosis results; The result output layer is used to output the fault diagnosis results of low-voltage switchgear.

2. A low voltage switch cabinet fault diagnosis method according to claim 1, characterized in that: The specific steps for performing synchronous analysis in the dual-channel analysis layer include: The dual-channel analysis layer includes a data input layer, a first channel layer, a second channel layer and a feature fusion layer; Wherein, the first channel layer and the second channel layer are parallel structures; In the data input layer, the pre-processed low-voltage switch cabinet fault data is input into the first channel layer based on the first channel label; the pre-processed low-voltage switch cabinet sensor monitoring data set is input into the second channel layer based on the second channel label; In the first channel layer, convolution feature extraction is performed on the pre-processed low-voltage switch cabinet fault data to obtain the convolution feature of the low-voltage switch cabinet fault; In the second channel layer, the time series feature extraction is performed on the pre-processed low-voltage switch cabinet sensor monitoring data set to obtain the low-voltage switch cabinet monitoring time series feature; In the feature fusion layer, the specific steps are: The convolution feature of the low-voltage switch cabinet fault is flattened to obtain the two-dimensional convolution feature Z of the low-voltage switch cabinet fault; the low-voltage switch cabinet monitoring time series feature is mapped through the fully connected layer to obtain the two-dimensional low-voltage switch cabinet monitoring time series feature S; Using the formula The fault diagnosis characteristic matrix A of the low-voltage switch cabinet is calculated; is the transposition of the two-dimensional low-voltage switchgear monitoring time series feature S, and C is the scaling factor; Based on the two-dimensional low-voltage switchgear fault convolution feature Z and the low-voltage switchgear fault diagnosis feature matrix A, a bidirectional feature interaction is performed to obtain the interactive low-voltage switchgear fault convolution feature; the two-dimensional low-voltage switchgear fault convolution feature Z and the interactive low-voltage switchgear fault convolution feature are residually connected to obtain the enhanced low-voltage switchgear fault convolution feature; Based on the two-dimensional low-voltage switchgear monitoring time series feature S and the low-voltage switchgear fault diagnosis feature matrix A, a two-dimensional low-voltage switchgear monitoring time series feature S and an interactive low-voltage switchgear monitoring time series feature are subjected to a residual connection to obtain an enhanced low-voltage switchgear monitoring time series feature. The sigmoid function is used to perform weighted feature fusion and pooling on the enhanced low-voltage switchgear fault convolution features and the enhanced low-voltage switchgear monitoring time series features to obtain the fused low-voltage switchgear fault diagnosis feature R; Using the formula Calculate the fault diagnosis classification results and obtain the low-voltage switchgear fault diagnosis results; where MLP is the trained multi-layer perceptron model and softmax is the activation function.

3. A low voltage switch cabinet fault diagnosis method according to claim 2, characterized in that: The specific steps of performing convolution feature extraction in the first channel layer and performing temporal feature extraction in the second channel layer include: The first channel layer is constructed based on the CNN model, which includes J1 3*3 convolution kernels, J2 5*5 convolution kernels and SE modules; Among them, all convolution kernels are used to perform multi-scale convolution operations on the pre-processed low-voltage switchgear fault data, and the SE module is used to enhance important features using the attention mechanism; The second channel layer is constructed based on the bidirectional GRU model; forward propagation and backward propagation are used to perform bidirectional feature extraction to obtain the bidirectional hidden time series of low-voltage switchgear monitoring; feature weighted aggregation is performed on the bidirectional hidden time series of low-voltage switchgear monitoring to obtain the timing characteristics of low-voltage switchgear monitoring.

4. A low voltage switch cabinet fault diagnosis method according to claim 3, characterized in that: The low-voltage switchgear predictive fault analysis model includes a data analysis layer, a dynamic prediction layer, a time interval change layer and a prediction result output layer; The data analysis layer is used to obtain the first M low-voltage switchgear sensor monitoring data to form a low-voltage switchgear prediction fault analysis data set; The dynamic prediction layer is used to perform risk prediction analysis on the low-voltage switchgear prediction fault analysis data set to obtain the low-voltage switchgear prediction fault risk feature value; The time interval variation layer is used to perform data analysis based on the fault risk characteristic value predicted by the low-voltage switchgear to obtain the dynamic time interval; The prediction result output layer is used to output dynamic time intervals.

5. A low voltage switch cabinet fault diagnosis method according to claim 4, characterized in that: The specific steps for data analysis in the time interval change layer include: Based on the low-voltage switchgear predicted fault risk characteristic value and the current dynamic time interval, K changing time interval individuals G are constructed. k , k=1, 2, ..., K; each change time interval individual contains a simulated dynamic time interval for updating; K change time interval individuals G k Combine, get the iterative population with variable time interval, and set the maximum number of iterations; Using the formula Calculate the dynamically changing objective function F; Among them, T post is the current dynamic time interval, T pre Individual G for different time intervals of change k The corresponding simulation dynamic time interval, X(T post ) represents the predicted fault risk characteristic value of the low-voltage switchgear, Y(T pre ) indicates that the dynamic time interval is updated to T pre Failure trend value after, T min is the minimum dynamic sampling time interval; W1, W2 and W3 are weight coefficients; The change time interval individual G k The corresponding calculated dynamic change objective function F is used as the fitness D k ; In the population iteration process, the local optimal position sharing and dynamic neighborhood partitioning mechanism are used to perform population iteration mutation on the iterative population with varying time intervals, thereby completing the update operation on the iterative population with varying time intervals. When the maximum number of iterations is reached, the change time interval individual corresponding to the maximum fitness is output, which is the optimal change time interval individual; based on the optimal change time interval individual, the current dynamic time interval is updated to obtain the dynamic time interval.

6. A low voltage switch cabinet fault diagnosis method according to claim 5, characterized in that: The specific steps for training the dynamic prediction layer include: Collect several groups of fault prediction training samples, each group of fault prediction training samples contains predicted fault label values ​​and verified low-voltage switchgear fault analysis data sets; combine several groups of fault prediction training samples to obtain a fault prediction training set; The dynamic prediction layer is trained based on the fault prediction training set to obtain the initial dynamic prediction layer; the initial dynamic prediction layer is evaluated, and if the initial dynamic prediction layer passes the model evaluation, the initial dynamic prediction layer is used as the dynamic prediction layer in the low-voltage switchgear fault prediction analysis model; otherwise, the fault prediction training set is used to continue model training.

7. A low voltage switch cabinet fault diagnosis system, characterized in that: The system applies a low-voltage switch cabinet fault diagnosis method as described in any one of claims 1 to 6, comprising: A fault diagnosis module includes a data acquisition unit and a fault diagnosis unit; The data acquisition unit is used to acquire the current low-voltage switch cabinet fault data and the low-voltage switch cabinet sensor monitoring data set when receiving the low-voltage switch cabinet fault signal; the low-voltage switch cabinet sensor monitoring data set contains N low-voltage switch cabinet sensor monitoring data; the current low-voltage switch cabinet fault data contains the electrical variable data of the low-voltage switch cabinet; the low-voltage switch cabinet sensor monitoring data contains the low-voltage switch cabinet temperature and humidity data, the low-voltage switch cabinet vibration spectrum data and the low-voltage switch cabinet switch state data; the low-voltage switch cabinet sensor monitoring data is continuously monitored according to the dynamic time interval, and when the low-voltage switch cabinet fault signal is received, the first N / 2 low-voltage switch cabinet sensor monitoring data and the last N / 2 low-voltage switch cabinet sensor monitoring data are acquired with the current time node as the origin to form the low-voltage switch cabinet sensor monitoring data set; the current time node is the time when the low-voltage switch cabinet fault signal is received; The fault diagnosis unit is used to analyze the current low-voltage switchgear fault data, the low-voltage switchgear sensor monitoring data set and the low-voltage switchgear fault diagnosis model to obtain the low-voltage switchgear fault diagnosis result; perform fault repair and subsequent operations on the low-voltage switchgear according to the low-voltage switchgear fault diagnosis result; the low-voltage switchgear fault diagnosis model performs multi-scale convolution and bidirectional time series feature extraction on the fault data and sensor monitoring data in a dual-channel structure, and combines attention weighted fusion and residual enhancement to achieve fault diagnosis; A fault prediction module, including a fault prediction analysis unit; The fault prediction analysis unit is used to update the dynamic time interval; wherein the dynamic time interval is determined in real time based on the low-voltage switchgear predictive fault analysis model; the low-voltage switchgear predictive fault analysis model realizes the prediction and response to the low-voltage switchgear fault by dynamically adjusting the time interval and continuously optimizing the prediction model.

Images