Isolation switch jam state diagnosis method, device, equipment, medium and product

Abstract

The embodiment of the invention provides an isolation switch jam state diagnosis method and device, equipment, a medium and a product. The method comprises the steps of monitoring a current signal of a target isolation switch, obtaining an angle signal, a torque signal and a current signal of the target isolation switch when the current signal is higher than a preset threshold value, and processing the angle signal, the torque signal and the current signal to obtain multi-modal fusion data, and inputting the multi-modal fusion data into a pre-trained isolation switch jam state diagnosis model to obtain a jam state diagnosis result of the target isolation switch. The method is used for achieving the effects of enhancing the capability of judging complex faults and improving the accuracy of a fault diagnosis result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power equipment state detection, and in particular to a disconnecting switch jam state diagnosis method, device, equipment, medium and product. BACKGROUND

[0002] The disconnecting switch in the power system is a key device of the substation and the power transmission and distribution network, and is mainly used for electrical isolation, maintenance disconnection and switching operation. Due to its long-term exposure to outdoor environment and frequent mechanical opening and closing operation, it is easy to cause mechanical jamming, movement retardation, contact misplacement and other faults due to mechanical wear, lubrication failure, transmission mechanism deformation or environmental corrosion. If not found in time, it may further cause an increase in contact resistance, local overheating, insulation deterioration and even discharge breakdown, which seriously threatens the safety of the power grid.

[0003] At present, the state monitoring of the disconnecting switch mainly relies on a single sensor such as vibration, current or infrared temperature measurement, and fault diagnosis is based on fixed thresholds or manual analysis.

[0004] However, the single sensor is susceptible to environmental noise interference, and the fault diagnosis is based on fixed thresholds or manual analysis, which lacks adaptive ability, resulting in the problem of poor fault diagnosis result precision of the existing method. SUMMARY

[0005] The embodiments of the present application provide a disconnecting switch jam state diagnosis method, device, equipment, medium and product to enhance the discrimination ability for complex faults and improve the fault diagnosis result precision.

[0006] In a first aspect, the embodiments of the present application provide a disconnecting switch jam state diagnosis method, comprising:

[0007] Monitoring a current signal of a target disconnecting switch.

[0008] When the current signal is higher than a preset threshold, obtaining an angle signal, a torque signal and the current signal of the target disconnecting switch.

[0009] Processing the angle signal, the torque signal and the current signal to obtain multi-modal fusion data.

[0010] Inputting the multi-modal fusion data into a pre-trained disconnecting switch jam state diagnosis model to obtain a jam state diagnosis result of the target disconnecting switch.

[0011] In a possible implementation, in combination with the first aspect, processing the angle signal, the torque signal and the current signal to obtain the multi-modal fusion data comprises:

[0012] The angle signal, the torque signal and the current signal are sequentially aligned to obtain a multi-channel signal sequence with a preset length.

[0013] The multi-channel signal sequence is normalized to obtain multi-modal fusion data.

[0014] In a possible implementation, in combination with the first aspect, the training process of the isolator sticking state diagnosis model comprises:

[0015] A training sample set is obtained, and the training sample set comprises a plurality of groups of training sample data.

[0016] The training sample data is input into the isolator sticking state diagnosis model to obtain a sticking state prediction result.

[0017] A loss value between the sticking state prediction result and a state label in the training sample data is calculated according to a preset loss function with a weight.

[0018] A preset optimizer is used to update parameters of the isolator sticking state diagnosis model according to the loss value, and based on the updated parameters, the input of the training sample data into the isolator sticking state diagnosis model to obtain the sticking state prediction result and subsequent processing are repeatedly performed until a preset condition is reached, and a trained isolator sticking state diagnosis model is obtained.

[0019] In a possible implementation, in combination with the first aspect, the training sample set is obtained, and the training sample set comprises a plurality of groups of training sample data, which comprises:

[0020] Sample angle signals, sample torque signals, sample current signals and corresponding state labels of a sample isolator are obtained.

[0021] The sample angle signals, the sample torque signals and the sample current signals are processed to obtain sample multi-modal fusion data.

[0022] The training sample set is constructed according to the sample multi-modal fusion data and the corresponding state labels, and the training sample set comprises a plurality of groups of training sample data with a preset batch size.

[0023] In a possible implementation, in combination with the first aspect, the multi-modal fusion data is input into the pre-trained isolator sticking state diagnosis model to obtain a sticking state diagnosis result of a target isolator, which comprises:

[0024] The multi-modal fusion data is feature-extracted through a convolution layer of the isolator sticking state diagnosis model to obtain convolution fusion features corresponding to the multi-modal fusion data.

[0025] The pooling layer of the isolator sticking state diagnosis model performs down-sampling processing on the convolution fusion feature to obtain down-sampled features corresponding to the convolution fusion feature.

[0026] The regularization layer of the isolator sticking state diagnosis model performs regularization processing on the down-sampled features to obtain regularization features corresponding to the down-sampled features.

[0027] The bidirectional long short-term memory network layer of the isolator sticking state diagnosis model processes the regularization features to obtain bidirectional time sequence features.

[0028] The output layer of the isolator sticking state diagnosis model processes the bidirectional time sequence features to obtain the sticking state diagnosis result of the target isolator.

[0029] In a possible implementation, in combination with the first aspect, the sticking state diagnosis result of the target isolator includes: the target isolator has a sticking abnormality and a probability of having the sticking abnormality, and / or the target isolator does not have a sticking abnormality and a probability of not having the sticking abnormality.

[0030] The second aspect, an embodiment of the present application provides an isolator sticking state diagnosis device, comprising:

[0031] The monitoring module is configured to monitor a current signal of a target isolator.

[0032] The acquisition module is configured to acquire an angle signal, a torque signal, and the current signal of the target isolator when the current signal is higher than a preset threshold, and to process the angle signal, the torque signal, and the current signal to obtain multi-modal fusion data.

[0033] The prediction module is configured to input the multi-modal fusion data into a pre-trained isolator sticking state diagnosis model to obtain a sticking state diagnosis result of the target isolator.

[0034] In a possible implementation, in combination with the second aspect, the acquisition module processes the angle signal, the torque signal, and the current signal to obtain the multi-modal fusion data, and is specifically configured to:

[0035] Perform sequence alignment processing on the angle signal, the torque signal, and the current signal to obtain a multi-channel signal sequence of a preset length.

[0036] Perform normalization processing on the multi-channel signal sequence to obtain the multi-modal fusion data.

[0037] In a possible implementation, in combination with the second aspect, an isolator sticking state diagnosis device further includes a training module configured to perform a training process of an isolator sticking state diagnosis model, and is specifically configured to:

[0038] The training sample set is obtained, and the training sample set includes multiple groups of training sample data.

[0039] The training sample data is input into the disconnecting switch jam state diagnosis model to obtain a jam state prediction result.

[0040] A loss value between the jam state prediction result and a state label in the training sample data is calculated according to a preset loss function with a weight.

[0041] A preset optimizer is used to update parameters of the disconnecting switch jam state diagnosis model according to the loss value, and based on the updated parameters, the training sample data is repeatedly input into the disconnecting switch jam state diagnosis model to obtain the jam state prediction result and subsequent processing until a preset condition is reached, thereby obtaining a trained disconnecting switch jam state diagnosis model.

[0042] In a possible implementation, in combination with the second aspect, the training module obtains a training sample set, and the training sample set includes multiple groups of training sample data, and is specifically used for:

[0043] Sample angle signals, sample torque signals, sample current signals and corresponding state labels of a sample disconnecting switch are obtained.

[0044] The sample angle signals, the sample torque signals and the sample current signals are processed to obtain sample multi-modal fusion data.

[0045] The training sample set is constructed according to the sample multi-modal fusion data and the corresponding state labels, and the training sample set includes multiple groups of training sample data with a preset batch size.

[0046] In a possible implementation, in combination with the second aspect, the prediction module inputs the multi-modal fusion data into the pre-trained disconnecting switch jam state diagnosis model to obtain a jam state diagnosis result of a target disconnecting switch, and is specifically used for:

[0047] The multi-modal fusion data is feature-extracted through a convolution layer of the disconnecting switch jam state diagnosis model to obtain convolution fusion features corresponding to the multi-modal fusion data.

[0048] The convolution fusion features are down-sampled through a pooling layer of the disconnecting switch jam state diagnosis model to obtain down-sampling features corresponding to the convolution fusion features.

[0049] The down-sampling features are regularized through a regularization layer of the disconnecting switch jam state diagnosis model to obtain regularization features corresponding to the down-sampling features.

[0050] The regularization features are processed through a bidirectional long short-term memory network layer of the disconnecting switch jam state diagnosis model to obtain bidirectional time sequence features.

[0051] The output layer of the isolator sticking state diagnosis model processes the bidirectional time sequence features to obtain a sticking state diagnosis result of the target isolator.

[0052] In a possible implementation, in combination with the second aspect, the sticking state diagnosis result of the target isolator includes: the target isolator has a sticking abnormality and a probability of having the sticking abnormality, and / or the target isolator does not have a sticking abnormality and a probability of not having the sticking abnormality.

[0053] In a third aspect, an embodiment of the present application provides an electronic device, including: a processor, and a memory connected with the processor in communication.

[0054] The memory stores computer execution instructions.

[0055] The processor executes the computer execution instructions stored in the memory, so that the processor executes the first aspect and / or various possible implementations of the first aspect.

[0056] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the first aspect and / or various possible implementations of the first aspect.

[0057] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and the computer program is executed by the processor to implement the first aspect and / or various possible implementations of the first aspect.

[0058] The isolator sticking state diagnosis method, device, equipment, medium and product provided by the embodiments of the present application can monitor the current signal of the target isolator, acquire the angle signal, the torque signal and the current signal of the target isolator when the current signal is higher than a preset threshold, process the angle signal, the torque signal and the current signal, obtain the multi-modal fusion data, input the multi-modal fusion data into the pre-trained isolator sticking state diagnosis model, and obtain the sticking state diagnosis result of the target isolator, so as to enhance the discrimination ability for complex faults and improve the accuracy of the fault diagnosis result. BRIEF DESCRIPTION OF DRAWINGS

[0059] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate the embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0060] Figure 1 A scene schematic diagram of the isolator sticking state diagnosis method provided by the present application;

[0061] Figure 2A flowchart illustrating a method for diagnosing the stuck state of a disconnecting switch provided in this application. Figure 1 ;

[0062] Figure 3 A flowchart illustrating a method for diagnosing the stuck state of a disconnecting switch provided in this application. Figure 2 ;

[0063] Figure 4 A schematic diagram of the model training process for the diagnostic model of the disconnector switch jamming state provided in this application;

[0064] Figure 5 A specific example diagram of a method for diagnosing the stuck state of a disconnecting switch provided in this application;

[0065] Figure 6 A flowchart for data preprocessing and dataset creation provided in this application;

[0066] Figure 7 This is a schematic diagram of the diagnostic model for the stuck state of the disconnecting switch provided in this application.

[0067] Figure 8 The classification results of the training sample set and test sample set provided for this application;

[0068] Figure 9 A schematic diagram of a diagnostic device for the stuck state of a disconnecting switch provided in this application;

[0069] Figure 10 A schematic diagram of the structure of the electronic device provided in this application.

[0070] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0071] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0072] First, the terms used in this application will be explained:

[0073] Convolutional Neural Network, commonly known as CNN, is a deep learning model that extracts local features through convolutional layers. Its working principle is based on convolution operations, using trainable filters to slide over the input data and capture key features through weighted summation of local regions.

[0074] 1D Convolutional Neural Network, commonly known as 1D CNN, is a variant of CNN specifically designed for processing one-dimensional sequence data. Its core principle is similar to that of two-dimensional CNN, but the convolution operation is only performed in a single dimension, usually the time dimension.

[0075] Bidirectional Long Short-Term Memory, commonly known as BiLSTM, is an extension of Long Short-Term Memory (LSTM) and is specifically designed to capture bidirectional temporal dependencies in sequence data. Its core structure consists of two independent LSTM units: one processes input data in the forward time order (from the beginning to the end of the sequence) to capture "past to present" temporal features; the other processes input data in the reverse time order (from the end to the beginning of the sequence) to capture "future to present" temporal features. Finally, the outputs of the two directions are concatenated or fused to form a feature representation that contains both past and future context information.

[0076] Max Pooling Layer, commonly known as MaxPooling layer, is the most commonly used type of pooling layer. Its main function is to filter and reduce the dimension of the feature map output by the convolutional layer.

[0077] Dropout layer is a commonly used regularization layer in deep learning, which can prevent model overfitting.

[0078] Secondly, the application background of the embodiments of the present application is explained:

[0079] The isolator in the power system is a key device of the substation, power transmission and distribution network, and is mainly used for electrical isolation, maintenance disconnection and switching operation. Due to its long-term exposure to outdoor environment and frequent mechanical opening and closing operation, it is easy to cause mechanical jamming, movement retardation, contact failure and other faults due to mechanical wear, lubrication failure, transmission mechanism deformation or environmental corrosion. If not found in time, it may further cause an increase in contact resistance, local overheating, insulation deterioration and even discharge breakdown, which seriously threatens the safety of the power grid.

[0080] At present, the state monitoring of the isolator mainly relies on a single sensor such as vibration, current or infrared temperature measurement, and fault diagnosis is based on fixed threshold or manual analysis. However, a single sensor is easily disturbed by environmental noise and is difficult to accurately represent complex fault characteristics. Fault diagnosis based on fixed threshold or manual analysis lacks adaptive ability, resulting in the problem of poor accuracy of fault diagnosis results of the existing method.

[0081] In view of the above problem, the inventors studied whether the jamming state diagnosis result of the isolator can be obtained by fusing the angle, torque, current and other multi-modal signal data of the isolator, and inputting the fused data into a trained model. The inventors proposed an isolator jamming state diagnosis method, which acquires the angle signal, torque signal and current signal of the target isolator when the current signal is higher than a preset threshold by monitoring the current signal of the target isolator. The angle signal, torque signal and current signal are processed to obtain multi-modal fusion data. The multi-modal fusion data is input into a pre-trained isolator jamming state diagnosis model to obtain the jamming state diagnosis result of the target isolator, thereby enhancing the discrimination ability of complex faults and improving the accuracy of fault diagnosis results.

[0082] Taking the intelligent inspection system of the substation as an example, combined with Figure 1 , the specific application scenario of the isolator jamming state diagnosis method provided by the present application is illustrated. As shown in Figure 1 , the specific application scenario of the present application includes an isolator 101, a multi-modal sensor 102, an edge intelligent gateway 103 and a power grid monitoring platform 104. The multi-modal sensor 102 is deployed on the isolator 101 and can collect the running data of the isolator 101 in real time, and is in communication connection with the edge intelligent gateway 103. The edge intelligent gateway 103 is in communication connection with the power grid monitoring platform 104.

[0083] When the disconnector 101 performs the switching operation, the multi-modal sensor 102 (angle, torque, current sensor) collects the operation data in real time and sends the operation data to the edge intelligent gateway 103; the edge intelligent gateway 103 calls the carding state diagnosis method of the disconnector 101, obtains the carding state diagnosis result, and sends the carding state diagnosis result to the power grid monitoring platform 104 in real time to assist the staff to quickly locate the fault position.

[0084] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0085] Figure 2 The flowchart of a disconnector carding state diagnosis method provided by the present application Figure 1 As shown in Figure 2 , the method comprises:

[0086] S201, monitoring the current signal of the target disconnector.

[0087] In this step, a non-intrusive sensing technology is used to sense the change of the current signal of the target disconnector in real time through the current transformer arranged at the key position of the target disconnector.

[0088] It should be noted that other types of sensors are also arranged for sensing various physical change signals in subsequent steps. Specifically, torque sensors and angle sensors can be arranged for sensing torque signals and angle signals in subsequent steps.

[0089] In a possible implementation, the current data of the disconnector is monitored and collected by a three-phase alternating current transducer integrated in the motor drive circuit.

[0090] S202, when the current signal is higher than the preset threshold, acquiring the angle signal, the torque signal and the current signal of the target disconnector.

[0091] In this step, when it is sensed that the current signal of the target disconnector is higher than the preset threshold, data interception is triggered, and various physical change signals of the target disconnector are acquired through various sensors, so as to realize automatic interception of various physical change signals corresponding to the effective opening and closing action segment based on the current threshold, and reduce invalid data interference. The preset threshold can be determined based on historical data analysis or can be determined according to the rated parameters of the equipment.

[0092] It should be understood that when the current signal is lower than the preset threshold, the acquisition of the various physical change signals is stopped.

[0093] In a possible implementation, the three-phase current of the motor is collected by the current transformer and synchronized to the controller, and a threshold (such as a current peak value) is set in the controller to trigger data interception, that is, when the current signal of the target disconnecting switch is monitored to be higher than the preset threshold, the controller sends a synchronous collection instruction to the angle sensor and the torque sensor, the angle signal is obtained by monitoring the opening and closing angle change of the disconnecting switch through the angle sensor, and the torque signal is obtained by collecting the mechanical torque output by the driving motor through the torque sensor. At the same time, the controller will collect the angle signal, the torque signal and the current signal during data interception and package them into a JSON structure containing time stamp, physical quantity unit, sampling channel and synchronization identifier in a unified format.

[0094] In a possible implementation, the angle, torque and current signals of the disconnecting switch are collected by the multi-modal sensor, a current threshold is set as an action trigger criterion, and when the monitored current exceeds the set threshold, the data collection process is automatically started, data collection and recording are performed at a preset sampling frequency, and in each operation period, the effective waveform data of the angle, torque and current in the corresponding time period is intercepted, the collected multi-source data is uniformly formatted and packaged into a JSON structure containing time stamp, physical quantity unit, sampling channel and synchronization identifier, and subsequent analysis and processing are facilitated.

[0095] S203, processing the angle, torque and current signals to obtain multi-modal fusion data.

[0096] In this step, the angle signal, the torque signal and the current signal are subjected to sequence alignment processing to obtain a multi-channel signal sequence of a preset length. The multi-channel signal sequence is subjected to normalization processing to obtain multi-modal fusion data.

[0097] Specifically, the angle signal, the torque signal and the current signal data of different lengths are truncated or zero-padded to be uniformly fixed to a preset length to obtain a multi-channel signal sequence to adapt to the input requirements of the disconnecting switch sticking state diagnosis model. The multi-channel signal sequence is subjected to normalization processing, for example, a Min-Max normalization strategy can be used to map the multi-channel signal sequence to the interval [0, 1].

[0098] In a possible implementation, the current signal output by the current transformer, the angle signal output by the angle sensor and the torque signal output by the torque sensor are all analog signals, which need to be converted into digital quantities first, then packaged into serial port signals based on a serial port protocol format, and finally transmitted to a data processing device, such as an edge intelligent gateway, through an RS485 interface, and subjected to sequence alignment processing and normalization processing in the data processing device to obtain multi-modal fusion data.

[0099] S204, input the multi-modal fusion data into the pre-trained disconnecting switch stickiness state diagnosis model to obtain a stickiness state diagnosis result of the target disconnecting switch.

[0100] In this step, the stickiness state diagnosis result of the target disconnecting switch can include: the target disconnecting switch has a stickiness abnormality and a probability of having the stickiness abnormality, and / or the target disconnecting switch does not have a stickiness abnormality and a probability of not having the stickiness abnormality.

[0101] The disconnecting switch stickiness state diagnosis method provided by the embodiment of the application includes the following steps:

[0102] Figure 3 The flowchart of the disconnecting switch stickiness state diagnosis method provided by the embodiment of the application Figure 2 As shown in the flowchart, the embodiment of the application is based on the embodiment of the application and specifically explains a disconnecting switch stickiness state diagnosis method. Figure 3 The embodiment of the application is based on the embodiment of the application and specifically explains a disconnecting switch stickiness state diagnosis method. Figure 2 The embodiment of the application is based on the embodiment of the application and specifically explains a disconnecting switch stickiness state diagnosis method.

[0103] S301, when the current signal is higher than the preset threshold, the angle signal, the torque signal and the current signal of the target disconnecting switch are acquired.

[0104] In this step, the current signal of the target disconnecting switch is monitored, and when the current signal is higher than the preset threshold, the angle signal, the torque signal and the current signal of the target disconnecting switch are acquired.

[0105] S302, the angle signal, the torque signal and the current signal are subjected to sequence alignment processing to obtain a multi-channel signal sequence with a preset length.

[0106] In this step, the unequal length angle signal, torque signal and current signal data are truncated or zero-padded to a fixed preset length to obtain a multi-channel signal sequence to adapt to the input requirements of the disconnecting switch stickiness state diagnosis model. The multi-channel signal sequence size is [batch number, feature dimension, time step], wherein the time step is the preset length and is fixed at 99; the feature dimension is 3, corresponding to the angle channel, torque channel and current channel.

[0107] S303, normalizing the multi-channel signal sequence to obtain multi-modal fusion data.

[0108] In this step, the multi-channel signal sequence is normalized, for example, a Min-Max Normalization strategy can be used to map the multi-channel signal sequence to the [0, 1] interval to obtain multi-modal fusion data.

[0109] S304, inputting the multi-modal fusion data to the input layer of the pre-trained diagnosis model.

[0110] In this step, the multi-modal fusion data is input to the input layer of the pre-trained disconnecting switch stickiness state diagnosis model. The diagnosis model is a cascaded neural network structure that combines the local perception ability of a one-dimensional convolutional neural network (1D CNN) and the time series modeling ability of a bidirectional long short-term memory network (BiLSTM).

[0111] S305, extracting features from the multi-modal fusion data through the convolutional layer of the diagnosis model to obtain convolutional fusion features.

[0112] In this step, the multi-modal fusion data is extracted through the convolutional layer of the disconnecting switch stickiness state diagnosis model to obtain convolutional fusion features corresponding to the multi-modal fusion data. The convolutional layer can be a one-dimensional convolutional layer used in a one-dimensional convolutional neural network (1D CNN), and the convolutional kernel size of the convolutional layer is configured as 5, the input channel number is 3, and the output channel number is configured as 64. The 64 output channels correspond to extracting 64 different local time series features (such as signal mutations in different frequency bands, multi-physical quantity collaborative change patterns, etc.) from the original 3 input channel physical quantities, providing a richer feature basis for subsequent time series modeling.

[0113] In one possible implementation, the mathematical expression corresponding to the convolution operation is:

[0114]

[0115] wherein, is the convolutional fusion feature output after the convolution operation; represents the input channel number, which is 3; denotes the convolution kernel width, and has a value of 5; is a sample batch number; is a weight parameter of the convolution kernel, is a trainable parameter in the model, and is randomly assigned at the beginning, for example, the Xavier initialization method can be used for assignment, and the assignment is continuously adjusted in the training process; is input training sample data; is a convolution kernel position; is a time step number; is a bias term, corresponding to the number of output channels, and there are 64 bias values (1 for each output channel), which is used to adjust the baseline of the convolution calculation result, avoid the influence of the overall offset of the input data on feature extraction, and enhance the model fitting capability.

[0116] S306, the pooling layer of the diagnostic model is used to perform down-sampling processing on the convolution fusion feature to obtain a down-sampled feature.

[0117] In this step, the pooling layer of the isolation switch card stick state diagnostic model is used to perform down-sampling processing on the convolution fusion feature to obtain a down-sampled feature corresponding to the convolution fusion feature. The pooling layer can be a maximum pooling layer, and the kernel size of the pooling layer is configured as 2.

[0118] In one possible implementation, the mathematical expression of the pooling calculation is as follows:

[0119]

[0120] wherein, is the down-sampled feature output after pooling; denotes the pooling window width, and has a value of 2; is the input of the pooling layer, which directly inherits the numerical value and dimension information of the output feature of the convolution layer.

[0121] S307, the regularization layer of the diagnostic model is used to perform regularization processing on the down-sampled feature to obtain a regularization feature.

[0122] In this step, the regularization layer of the isolation switch card stick state diagnostic model is used to perform regularization processing on the down-sampled feature to obtain a regularization feature corresponding to the down-sampled feature. The regularization layer can be a Dropout layer, and the probability is configured as 0.3. Regularization is realized by randomly discarding neurons to prevent model overfitting.

[0123] S308, the bidirectional long short-term memory network layer of the diagnostic model is used to process the regularization feature to obtain a bidirectional time sequence feature.

[0124] In this step, the bidirectional long short-term memory network (BiLSTM) layer of the isolating switch jam state diagnosis model is used to process the regularization features to obtain bidirectional time sequence features. The number of hidden units of the bidirectional long short-term memory network layer is configured as 64.

[0125] In one possible implementation, the mathematical expression corresponding to the forget gate in the BiLSTM layer is as follows:

[0126]

[0127] wherein, is the output value of the forget gate; is the weight matrix of the forget gate; is the hidden state of the LSTM unit at the previous moment, containing the time sequence feature information at the previous moment; is the input feature of the LSTM unit at the current moment, which is the down-sampling feature output by the regularization layer; is the bias term of the forget gate, used to adjust the output baseline of the forget gate and enhance the adaptability of the model to different data distributions.

[0128] The mathematical expression corresponding to the input gate and the candidate memory is as follows:

[0129]

[0130] The mathematical expression corresponding to the cell state update is as follows:

[0131]

[0132] wherein, is the output value of the input gate; is the candidate memory, containing new information of the input feature at the current moment; is the weight matrix of the input gate; is the weight matrix of the candidate memory; and are the bias terms of the input gate and the candidate memory, respectively; is the state of the LSTM unit at the current moment.

[0133] The mathematical expression corresponding to the output gate and the hidden state is as follows:

[0134]

[0135] wherein, is the output value of the output gate; is the hidden state of the LSTM unit at the current moment; is the weight matrix of the output gate; is the bias term of the output gate; is used to update the cell state The value of the state information of the unit is normalized to the interval of -1 to 1.

[0136] The mathematical expression corresponding to the BiLSTM bidirectional output formula is as follows:

[0137]

[0138] wherein, is the bidirectional hidden state of the BiLSTM layer at time t; is the hidden state of the forward LSTM unit at time t, and the time series data is processed from front to back in time sequence, capturing historical time series dependency; is the hidden state of the backward LSTM unit at time t, and the time series data is processed from back to front in time reverse order, capturing future time series dependency.

[0139] S309, processing the bidirectional time series features through the output layer of the diagnosis model to obtain the stickiness state diagnosis result of the target disconnecting switch.

[0140] In this step, the bidirectional time series features are processed through the output layer of the disconnecting switch stickiness state diagnosis model to obtain the stickiness state diagnosis result of the target disconnecting switch. The output layer includes a fully connected layer and a Sigmoid activation function, and a threshold decision method is used to obtain the stickiness state diagnosis result of the target disconnecting switch.

[0141] Specifically, the bidirectional time series features are first linearly mapped through the fully connected layer to convert the high-dimensional time series features into single-dimensional continuous numerical values; then the Sigmoid activation function layer is used for non-linear conversion to map the probability values between 0 and 1, and the probability value directly represents the probability of the target disconnecting switch having a stickiness anomaly. Since stickiness and normal are mutually exclusive and complete in the binary classification task, the probability of normal can be calculated by subtracting the probability of having a stickiness anomaly from 1. Finally, threshold judgment is performed, and 0.5 is set as the binary classification threshold: if the probability of having a stickiness anomaly is greater than 0.5, it is determined that there is a stickiness (at this time, the probability of normal is less than or equal to 0.5); otherwise, it is determined that there is no stickiness (at this time, the probability of normal is greater than or equal to 0.5), thereby completing the binary classification diagnosis of the disconnecting switch state.

[0142] In one possible implementation, the mapping formula of the fully connected layer is as follows:

[0143]

[0144] The formula for mapping to the binary classification output through the Sigmoid activation function is as follows:

[0145]

[0146] wherein, is the output of the first fully connected layer; is a rectified linear unit activation function with input of the fully connected layer linear output; is the weight matrix of the first fully connected layer, which is randomly assigned initially; is the bias term of the first fully connected layer; is the binary classification prediction output of the model; is a Sigmoid activation function used to output the probability prediction value of the sample belonging to the sticking class; is the weight matrix of the second mapping; is the bias term of the second mapping.

[0147] In one possible implementation, the isolator sticking state diagnosis model is a cascaded neural network structure that fuses the local perception ability of a one-dimensional convolutional neural network (1D CNN) and the time series modeling ability of a bidirectional long short-term memory network (BiLSTM), including a local time series feature extraction module, a sequence compression and regularization module, a global time series modeling module, and a state determination module. Among them, the local time series feature extraction module takes multi-modal sequence data (angle, torque, current, and other multi-modal fusion data) as input, based on 1D CNN, slides to extract local change patterns, and captures mutation features and inter-modal correlations; the sequence compression and regularization module introduces a max pooling layer (MaxPooling layer) to compress the time dimension and reduce redundant calculations, and uses a Dropout layer random inactivation mechanism to improve the model's robustness to local overfitting; the global time series modeling module sends the regularized output to a bidirectional long short-term memory network (BiLSTM), which models the time dependence relationship from both forward and backward directions, enhancing the perception ability of the sticking behavior evolution process; the state determination module uses a linear fully connected layer to map the bidirectional long short-term memory network (BiLSTM) output vector to the diagnosis space, and combines a Sigmoid activation function to generate a binary classification probability for identifying whether there is a sticking anomaly.

[0148] The application embodiment provides a kind of isolator jam state diagnostic method, by when current signal is higher than preset threshold, the angle signal, torque signal and current signal of target isolator are acquired, angle signal, torque signal and current signal are carried out sequence alignment processing and, obtain the multi-channel signal sequence of preset length, then carry out normalization processing, obtain multi-modal fusion data, input multi-modal fusion data to the input layer of pre-trained diagnostic model, by the convolution layer, pooling layer, regularization layer, bidirectional long short-term memory network layer and output layer of diagnostic model, multi-modal fusion data is carried out feature extraction, down-sampling, regularization, bidirectional time series modeling, conversion output etc. Series of processing, obtain the jam state diagnostic result of target isolator, realize the discrimination ability of enhancement to complex fault, improve the sensitivity to early fault, finally improve the effect of fault diagnosis result precision.

[0149] Figure 4 The model training process diagram of the isolator jam state diagnostic model in the isolator jam state diagnostic method provided by the application is as shown in Figure 4 The model training process in the isolator jam state diagnostic method is described in detail based on any one of the above embodiments, including:

[0150] S401, obtain training sample set, and the training sample set includes multiple groups of training sample data.

[0151] In this step, sample angle signal, sample torque signal, sample current signal and corresponding state label of sample isolator are acquired. Sample angle signal, sample torque signal and sample current signal are processed to obtain sample multi-modal fusion data. According to sample multi-modal fusion data and its corresponding state label, training sample set is constructed; multiple groups of training sample data with preset batch size are included in training sample set.

[0152] Specifically, the data in training sample set is represented in the form of tensor, and its size is [batch number, feature dimension, time step]. The time step is fixed at 99, and the feature dimension is 3, corresponding to angle, torque and current channels. The preset batch size of multiple groups of training sample data is 2.

[0153] In a possible implementation, the training sample set obtained in this step is established by the following process: the angle, torque and current signals of the disconnecting switch during opening and closing are collected by a multi-modal sensor, the collected multi-source data is uniformly formatted and packaged as a JSON structure containing timestamp, physical quantity unit, sampling channel and synchronization identifier, and the like, and is stored in a specified directory path. When the training sample set is to be obtained, the above stored file is read and parsed, the read data is subjected to sequence alignment processing, and then combined into a tensor structure according to a uniform format, and labeled as normal or sticking according to the category, and then the channel feature values in the data are normalized to establish the training sample set.

[0154] S402, input the training sample data into the disconnecting switch sticking state diagnosis model to obtain a sticking state prediction result.

[0155] S403, calculate a loss value between the sticking state prediction result and a state label in the training sample data according to a preset loss function with a weight.

[0156] In this step, the preset loss function is a BCEWithLogitsLoss loss function, which adopts a binary cross entropy form with a weight, and the positive sample weight is determined by the ratio of the number of positive samples to the number of negative samples in the training sample set, so as to balance the class imbalance problem.

[0157] In a possible implementation, the mathematical expression corresponding to the loss function of the model is as follows:

[0158]

[0159] wherein, is the loss value of the model, used to measure the error between the prediction value of the model and the real label ; , represents the weight coefficient of the positive class sample (sticking), is the number of negative class samples (normal), is the number of positive class samples; is the preset weight coefficient of the negative class sample; is the real label of the sample, if the sample is sticking, the value is 1, and if the sample is normal, the value is 0.

[0160] S404, update the parameters of the disconnecting switch sticking state diagnosis model according to the loss value by using a preset optimizer.

[0161] In this step, the preset optimizer is an Adam optimizer (Adaptive Moment Estimation, Adam optimizer for short), and the learning rate is 1e-3.

[0162] S405, judge whether the preset condition is reached, if not, execute S402-S405, if yes, execute S406.

[0163] In this step, based on the updated parameters, the training sample data is repeatedly input into the disconnecting switch jam state diagnosis model to obtain the jam state prediction result and subsequent processing until the preset condition is reached, and a trained disconnecting switch jam state diagnosis model is obtained.

[0164] Specifically, the model training can adopt a small batch gradient descent strategy, the batch size is set to 2, and the preset condition is the total number of training rounds, which can be set to 44 rounds.

[0165] S406, obtain a trained disconnecting switch jam state diagnosis model.

[0166] In a possible implementation, while obtaining the training sample set, a test sample set with the same structure but different data is also obtained, after obtaining the trained model based on the training sample set, the performance of the model is verified based on the test sample set, the classification performance of the model is quantitatively evaluated from indicators such as confusion matrix, accuracy, recall rate, and F1 score, and the performance change curve of the training and verification process can be visually output. Specifically, the prediction results of the training sample set and the test sample set can be compared with the real labels, a confusion matrix can be constructed and plotted, and a color picture can be used for visual presentation. The numbers on the color scale bar in the figure represent the number of samples, and the color from light to deep corresponds to the number from less to more. The vertical coordinate represents the actual class to which the test sample belongs, and the horizontal coordinate represents the classification result of the test sample. The samples with the same actual class and predicted class correspond to the elements at the diagonal line position of the confusion matrix, which represent correct classification.

[0167] The disconnecting switch jam state diagnosis method provided by the embodiment of the application comprises the following steps: obtaining a training sample set, the training sample set comprising a plurality of groups of training sample data; inputting the training sample data into a disconnecting switch jam state diagnosis model to obtain a jam state prediction result; calculating a loss value between the jam state prediction result and a state label in the training sample data according to a preset loss function with a weight; updating parameters of the disconnecting switch jam state diagnosis model according to the loss value by using a preset optimizer based on small batch training; repeatedly inputting the training sample data into the disconnecting switch jam state diagnosis model based on the updated parameters to obtain the jam state prediction result and subsequent processing until a preset condition is reached, and obtaining a trained disconnecting switch jam state diagnosis model, thereby improving the stability and generalization ability of the model and improving the accuracy of the fault diagnosis result.

[0168] On the basis of any one of the above embodiments, the following will be combined Figure 5A kind of isolating switch jam state diagnosis method is described in detail by specific example.

[0169] S501, data acquisition.

[0170] In this step, non-invasive sensing technology is used to place angle sensors, torque sensors and current transformers at key positions of the isolating switch, to sense various physical change signals in the opening and closing operation process in real time. Set the current threshold as the action trigger criterion, and automatically start the data acquisition process when the monitored current exceeds the set threshold, to ensure that the dynamic information at the key action moment is captured.

[0171] In each operation cycle, the effective waveform data of angle, torque and current in the corresponding time period is intercepted to avoid redundant information interference. The collected multi-source data is uniformly formatted and packaged as a JSON structure containing timestamp, physical quantity unit, sampling channel and synchronization identifier, etc. information, to facilitate subsequent analysis and processing. The above structured data is stored in the specified directory path, and a unified and standardized data input interface is constructed.

[0172] S502, data set establishment.

[0173] In this step, as shown in Figure 6 , Figure 6 is the data preprocessing and data set establishment flowchart, and the data set is established through a plurality of steps as shown in Figure 6 .

[0174] First, data reading. Read and parse the data file of structured data stored in S501, containing the time series information of angle, torque and current signals of the isolating switch during one split action.

[0175] Second, sequence alignment processing. The read data is subjected to sequence alignment processing, and the data segment longer than the preset threshold is truncated, and the data shorter than the preset threshold is zero-padded, so that the length of all data samples is kept consistent, and sample multi-modal fusion data is obtained.

[0176] Third, tensor construction and label binding. The sample multi-modal fusion data is combined into a tensor structure according to a unified format, and is labeled according to the category, including normal or jam.

[0177] Fourth, feature normalization. The feature values in the sample multi-modal fusion data are normalized, and the Min-Max Normalization strategy is adopted to improve the numerical stability in the training process.

[0178] Step 5: Dataset Partitioning. After processing all multimodal fusion data, label combinations, and feature normalization, the dataset is divided into training and testing sets using stratified sampling for subsequent model training and performance evaluation. Within the training set, the data is further divided into multiple training sample sets with a preset batch size of 2.

[0179] Step 6: Output the standardized dataset.

[0180] S503. Establish and train a diagnostic model for the stuck state of the isolation switch.

[0181] In this step, the diagnostic model for the stuck state of the disconnector switch is a cascaded neural network structure that integrates the local perception capability of a one-dimensional convolutional neural network (1D CNN) and the temporal modeling capability of a bidirectional long short-term memory network (BiLSTM). During the model training phase, multiple sets of training sample data (all three-channel time-series data) from S502 are used as training samples. After inputting them into the model, the parameters are optimized through a loss function to achieve model training. The three-channel time-series data records the temporal characteristics of the angle, torque, and current during the opening and closing of the disconnector switch.

[0182] Specifically, such as Figure 7 As shown, Figure 7 This is a schematic diagram of the diagnostic model for the stuck state of a disconnecting switch. The processing procedure of the diagnostic model for the stuck state of a disconnecting switch on the input data includes:

[0183] The first step is to input the data. A set of training sample data is input into the model. The batch size of the training sample data is 2. After being received by the input layer of the model, the training sample data (which are three-channel input signals of angle, torque and current) is sent into the CNN layer in time series form, that is, the convolutional layer in the CNN that performs convolution operations.

[0184] The second step is the CNN layer. The CNN layer performs convolutional fusion feature extraction on the training sample data. It is configured as a one-dimensional convolutional layer with a kernel width of 5, 3 input channels, and 64 output channels. Its convolutional operation on the training sample data can be represented as:

[0185]

[0186] The above formula is the same as the mathematical expression corresponding to the convolution operation in S305. The parameters in the formula have been explained in S305 and will not be repeated here.

[0187] Third step, MaxPooling layer. The MaxPooling layer receives the convolution fusion features output by the convolution layer, and performs pooling and dimension reduction processing on the convolution fusion features, time domain compression, enhances the robustness of the model to local distortion, and is used to reduce the probability of overfitting. The pooling calculation expression is:

[0188]

[0189] The above formula is the same as the mathematical expression corresponding to the pooling calculation in S306. The parameters in the formula have been explained and described in S306, and will not be repeated here.

[0190] Fourth step, Dropout layer. The Dropout layer receives the down-sampled features output by the pooling layer, and performs a regularization layer processing on the down-sampled features to output regularized features. The Dropout layer forces the model to learn more robust features by randomly discarding a portion of neurons during training, preventing model overfitting. The dropout probability is configured to be 0.3.

[0191] Fifth step, BiLSTM layer. The BiLSTM layer receives the regularized features output by the regularization layer. This layer has bidirectional information propagation capability and can capture both historical and future context timing features. The basic update formula of the unidirectional long short-term memory network (LSTM) includes:

[0192] Forget gate:

[0193]

[0194] Input gate and candidate memory:

[0195]

[0196] Cell state update:

[0197]

[0198] Output gate and hidden state:

[0199]

[0200] In the BiLSTM layer, the output bidirectional timing features are obtained by concatenating the hidden states of the forward and backward long short-term memory networks. The BiLSTM bidirectional output formula can be expressed as:

[0201]

[0202] The above formulas correspond to the mathematical expressions of the forgetting gate, the input gate, the candidate memory, the unit state update, the output gate, and the hidden state in S308, and the BiLSTM bidirectional output formula. The parameters in the formulas have been explained in S308 and will not be repeated here.

[0203] Step 6, output layer. The output layer extracts the bidirectional time sequence features output by the BiLSTM layer at the last time (time T) . The output layer includes two processing links: first, linear mapping of bidirectional time sequence features through a fully connected layer, and then nonlinear conversion through a Sigmoid activation function layer, finally outputting the predicted value. The processing link of the output layer on the bidirectional time sequence features can be represented as follows:

[0204] First fully connected layer:

[0205]

[0206] Second mapping to binary output:

[0207]

[0208] The above formulas correspond to the mapping formula of the fully connected layer in S309 and the formula of the binary output. The parameters in the formulas have been explained in S309 and will not be repeated here.

[0209] Finally, the sample imbalance processing and loss function design of the model are explained. To alleviate the training bias caused by the imbalance of positive and negative sample ratio, the loss function of the model adopts the weighted binary cross entropy form, which is expressed as follows:

[0210]

[0211] The above formula corresponds to the mathematical expression of the loss function in S403. The parameters in the formula have been explained in S403 and will not be repeated here.

[0212] S504, verify the performance of the trained isolator stick state diagnosis model.

[0213] In this step, the prediction results of the training sample set and the test sample set are compared with the true labels, a confusion matrix is constructed and plotted, and the classification accuracy and confusion relationship of different states are visually presented. After classifying the original isolator data, the classification results as shown in Figure 8 are obtained, Figure 8The image shows the classification results for the training and testing sample sets. Background density ranges from sparse to dense, corresponding to an increasing number of samples. The vertical axis represents the actual class of the test sample, while the horizontal axis represents the classification result. Samples whose actual and predicted classes match are represented by the elements on the diagonal of the confusion matrix, indicating correct classification.

[0214] In one possible implementation, the classification results can be visualized using a color image, where the numbers on the color scale represent the number of samples, with lighter to darker colors corresponding to a larger number of samples. The vertical axis represents the actual class of the test sample, while the horizontal axis represents the classification result of the test sample. Samples whose actual class matches the predicted class correspond to the elements on the diagonal of the confusion matrix, indicating correct classification.

[0215] It should be noted that, in Figure 5 The processing steps S501-S504 shown in the embodiments do not constitute a specific limitation on a method for diagnosing the stuck state of a disconnecting switch. In other embodiments of this application, a method for diagnosing the stuck state of a disconnecting switch may include... Figure 5 The embodiments may include more or fewer steps; for example, a method for diagnosing the stuck state of an isolating switch may include... Figure 5 Some steps in the embodiments, or, Figure 5 Some steps in the embodiments can be replaced by steps with the same function, or, Figure 5 Some steps in the embodiments can be broken down into multiple steps, etc.

[0216] Figure 9 This application provides a schematic diagram of the structure of a diagnostic device for the jamming state of a disconnecting switch. Figure 9 As shown, the disconnector switch jamming status diagnostic device 90 provided in this embodiment includes:

[0217] Monitoring module 901 is used to monitor the current signal of the target disconnect switch.

[0218] The acquisition module 902 is used to acquire the angle signal, torque signal and current signal of the target disconnect switch when the current signal is higher than a preset threshold; and to process the angle signal, torque signal and current signal to obtain multimodal fusion data.

[0219] The prediction module 903 is used to input multimodal fusion data into a pre-trained disconnector switch jamming state diagnosis model to obtain the jamming state diagnosis result of the target disconnector switch.

[0220] In one possible implementation, the acquisition module 902 processes the angle signal, torque signal, and current signal to obtain multimodal fusion data, specifically for:

[0221] The angle signal, the torque signal and the current signal are sequentially aligned to obtain a multi-channel signal sequence with a preset length.

[0222] The multi-channel signal sequence is normalized to obtain multi-modal fusion data.

[0223] In a possible implementation, the isolator sticking state diagnosis apparatus further includes a training module 904 configured to perform a training process of the isolator sticking state diagnosis model, and specifically configured to:

[0224] The training sample set is obtained, and the training sample set includes a plurality of groups of training sample data.

[0225] The training sample data is input into the isolator sticking state diagnosis model to obtain a sticking state prediction result.

[0226] The loss value between the sticking state prediction result and a state label in the training sample data is calculated according to a preset loss function with a weight.

[0227] A preset optimizer is used to update the parameters of the isolator sticking state diagnosis model according to the loss value, and based on the updated parameters, the inputting of the training sample data into the isolator sticking state diagnosis model to obtain the sticking state prediction result and the subsequent processing are repeatedly performed until a preset condition is reached, and a trained isolator sticking state diagnosis model is obtained.

[0228] In a possible implementation, the training module 904 obtains the training sample set, and the training sample set includes a plurality of groups of training sample data, and specifically configured to:

[0229] The sample angle signal, the sample torque signal, the sample current signal and the corresponding state label of the sample isolator are obtained.

[0230] The sample angle signal, the sample torque signal and the sample current signal are processed to obtain sample multi-modal fusion data.

[0231] The training sample set is constructed according to the sample multi-modal fusion data and the corresponding state label, and the training sample set includes a plurality of groups of training sample data with a preset batch size.

[0232] In a possible implementation, the prediction module 903 inputs the multi-modal fusion data into the pre-trained isolator sticking state diagnosis model to obtain the sticking state diagnosis result of the target isolator, and specifically configured to:

[0233] The multi-modal fusion data is feature extracted through a convolution layer of the isolator sticking state diagnosis model to obtain a convolution fusion feature corresponding to the multi-modal fusion data.

[0234] The pooling layer of the isolator sticking state diagnosis model performs down-sampling processing on the convolution fusion feature to obtain down-sampled features corresponding to the convolution fusion feature.

[0235] The regularization layer of the isolator sticking state diagnosis model performs regularization processing on the down-sampled features to obtain regularization features corresponding to the down-sampled features.

[0236] The bidirectional long short-term memory network layer of the isolator sticking state diagnosis model processes the regularization features to obtain bidirectional time sequence features.

[0237] The output layer of the isolator sticking state diagnosis model processes the bidirectional time sequence features to obtain a sticking state diagnosis result of the target isolator.

[0238] In a possible implementation, the sticking state diagnosis result of the target isolator includes: the target isolator has a sticking abnormality and a probability of having the sticking abnormality, and / or the target isolator does not have a sticking abnormality and a probability of not having the sticking abnormality.

[0239] The isolator sticking state diagnosis apparatus provided in this embodiment can execute the method provided in the method embodiments, and has similar implementation principles and technical effects, which will not be described here again.

[0240] Figure 10 The structure of the electronic device provided in this application is shown in FIG. 1. Figure 10 As shown in FIG. 1, the electronic device 100 provided in this embodiment includes at least one processor 1001 and a memory 1002. Optionally, the device 100 further includes a communication component 1003. The processor 1001, the memory 1002 and the communication component 1003 are connected through a bus 1004.

[0241] In the specific implementation process, the at least one processor 1001 executes the computer execution instructions stored in the memory 1002, so that the at least one processor 1001 executes the method described above.

[0242] The specific implementation process of the processor 1001 can refer to the method embodiments described above, and has similar implementation principles and technical effects, which will not be described here again.

[0243] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0244] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), such as at least one disk memory.

[0245] The bus can be an industry standard architecture (ISA) bus, a peripheral component (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0246] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above method.

[0247] The present application also provides a computer readable storage medium, which stores computer execution instructions, and when a processor executes the computer execution instructions, the above method is implemented.

[0248] The above readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0249] An example readable storage medium is coupled to the processor such that the processor can read information from the readable storage medium and can write information to the readable storage medium. Of course, the readable storage medium can also be a part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.

[0250] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0251] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0252] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0253] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0254] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the foregoing storage medium includes various media capable of storing program codes, such as ROM, RAM, magnetic disk, or optical disk.

[0255] Finally, it should be noted that other embodiments of the present application will readily occur to those skilled in the art upon consideration of the specification and practice of the present application disclosed herein. The present application is intended to include all such variations as fall within the general scope of the application, and includes the generic principles disclosed and the best mode known to the inventors to be currently practiced as well as variations thereof, without departing from the scope of the present application as defined by the claims. The specification and examples give the best application of the present application as known to at least one of the inventors at the time of the filing of this application. It is to be understood that since numerous modifications and changes will readily occur to those skilled in the art, the application is not to be limited to the exact construction and operation as illustrated and described. Accordingly, all such variations are intended to be included within the scope of the present application as defined in the claims. The application is to be limited only by the claims.

Claims

1. A method of diagnosing a stuck state of a disconnector, characterized by, The method comprises the following steps: monitoring a current signal of a target disconnecting switch; when the current signal is higher than a preset threshold, acquiring an angle signal, a torque signal and the current signal of the target disconnecting switch; processing the angle signal, the torque signal and the current signal to obtain multi-modal fusion data; inputting the multi-modal fusion data into a pre-trained disconnecting switch sticking state diagnosis model to obtain a sticking state diagnosis result of the target disconnecting switch.

2. The method of claim 1, wherein, The processing of the angle signal, the torque signal and the current signal to obtain multi-modal fusion data comprises: performing sequence alignment processing on the angle signal, the torque signal and the current signal to obtain a multi-channel signal sequence with a preset length; performing normalization processing on the multi-channel signal sequence to obtain the multi-modal fusion data.

3. The method of claim 1, wherein, The training process of the disconnecting switch sticking state diagnosis model comprises: acquiring a training sample set, wherein the training sample set comprises a plurality of training sample data; inputting the training sample data into the disconnecting switch sticking state diagnosis model to obtain a sticking state prediction result; calculating a loss value between the sticking state prediction result and a state label in the training sample data according to a preset loss function with a weight; updating parameters of the disconnecting switch sticking state diagnosis model according to the loss value by using a preset optimizer, and repeatedly performing the inputting of the training sample data into the disconnecting switch sticking state diagnosis model to obtain a sticking state prediction result and subsequent processing based on the updated parameters until a preset condition is met, thereby obtaining a trained disconnecting switch sticking state diagnosis model.

4. The method of claim 3, wherein, The acquiring of the training sample set, wherein the training sample set comprises a plurality of training sample data, comprises: acquiring sample angle signals, sample torque signals, sample current signals and corresponding state labels of a sample disconnecting switch; processing the sample angle signals, the sample torque signals and the sample current signals to obtain sample multi-modal fusion data; constructing the training sample set according to the sample multi-modal fusion data and the corresponding state labels, wherein the training sample set comprises a plurality of training sample data with a preset batch size.

5. The method of claim 1, wherein, The inputting of the multi-modal fusion data into the pre-trained disconnecting switch sticking state diagnosis model to obtain the sticking state diagnosis result of the target disconnecting switch comprises: extracting features of the multi-modal fusion data through a convolution layer of the disconnecting switch sticking state diagnosis model to obtain convolution fusion features corresponding to the multi-modal fusion data; performing down-sampling processing on the convolution fusion features through a pooling layer of the disconnecting switch sticking state diagnosis model to obtain down-sampling features corresponding to the convolution fusion features; performing regularization processing on the down-sampling features through a regularization layer of the disconnecting switch sticking state diagnosis model to obtain regularization features corresponding to the down-sampling features; processing the regularization features through a bidirectional long short-term memory network layer of the disconnecting switch sticking state diagnosis model to obtain bidirectional time sequence features. The output layer of the isolator jam state diagnosis model processes the bidirectional time sequence features to obtain the jam state diagnosis result of the target isolator.

6. The method of claim 1, wherein, The jam state diagnosis result of the target isolator includes: the target isolator has a jam abnormality and a probability of having a jam abnormality, and / or the target isolator does not have a jam abnormality and a probability of not having a jam abnormality.

7. A diagnostic device for the jammed state of a disconnecting switch, characterized in that, The method comprises: a monitoring module configured to monitor a current signal of a target isolator; an acquisition module configured to, when the current signal is higher than a preset threshold, acquire an angle signal, a torque signal and the current signal of the target isolator, and process the angle signal, the torque signal and the current signal to obtain multi-modal fusion data; a prediction module configured to input the multi-modal fusion data into a pre-trained isolator jam state diagnosis model to obtain a jam state diagnosis result of the target isolator.

8. An electronic device, comprising: The method comprises: a processor, and a memory connected to the processor in communication; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, so that the processor executes the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method of any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 6.

Images