Disconnecting link state monitoring method and system, terminal equipment and storage medium

Abstract

The invention discloses a disconnecting link state monitoring method and system, terminal equipment and a storage medium, and the method comprises the steps: carrying out the short-term state prediction operation of a disconnecting link through a three-dimensional convolutional neural network model, obtaining the short-term prediction state characteristics of the disconnecting link, carrying out the long-term state prediction operation of the short-term prediction state characteristics of the disconnecting link through a TCN-diffusion model, and obtaining the short-term prediction state characteristics of the disconnecting link. And finally, fault probability analysis is carried out on the long-term prediction state characteristics of the disconnecting link through a probability-based Bayesian anomaly detection model, and the fault probability of the disconnecting link is obtained. The state of the disconnecting link is predicted through the combination of the short-term state prediction model and the long-term state prediction model, so that the long-term state of the disconnecting link is predicted, the accuracy of the predicted state of the disconnecting link is improved, a guarantee is provided for subsequent fault analysis of the predicted state of the disconnecting link, and then the accuracy of fault analysis of the disconnecting link is improved.

Description

Technical Field

[0001] The present invention relates to the field of knife switch state monitoring, and in particular to a knife switch state monitoring method, system, terminal equipment and storage medium. Background Art

[0002] The status monitoring of knife switches is of vital importance to ensure the reliable operation of power systems and prevent equipment failures. At present, the monitoring of knife switch status mainly relies on traditional sensor technologies, including vibration sensors, temperature sensors and visual detection equipment. These technologies can provide basic information about whether the knife switch is in the on or off state, as well as some abnormal indicators during operation.

[0003] The existing switch state prediction technology usually uses the CNN neural convolutional network model to predict the state. The obtained switch state prediction data is the switch state at the next moment. However, the switch state prediction result at the next moment only reflects the short-term state of the switch in the future, and cannot reflect the long-term state of the switch in the future. Therefore, it will lead to the problem of low accuracy of the switch state prediction result, which will lead to an increase in the error of fault analysis based on the switch prediction state. Summary of the invention

[0004] The present invention provides a switch state monitoring method, system, terminal equipment and storage medium, which can improve the accuracy of switch fault analysis.

[0005] In order to solve the above technical problem, an embodiment of the present invention provides a switch state monitoring method, comprising:

[0006] Obtaining knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data;

[0007] Inputting the current switch state data into the trained short-term prediction model, so that the short-term prediction model generates the switch short-term prediction state data according to the current switch state data;

[0008] Input the short-term prediction state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term prediction state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term prediction state data of the knife switch;

[0009] The long-term predicted status data of the knife switch is input into the trained probability-based Bayesian anomaly detection model to calculate the predicted probability of knife switch failure. The knife switch status result is obtained based on the predicted probability and the preset failure probability threshold.

[0010] Further, the inputting of the current switch state data into the trained short-term prediction model so that the short-term prediction model generates the switch short-term prediction state data according to the current switch state data includes:

[0011] Preprocessing and fusing the image data, vibration data, temperature data and sound data of the knife switch state data at the current moment to obtain the knife switch fusion state data at the current moment;

[0012] The current switch fusion state data is obtained through the input layer of the short-term prediction model, and the convolution layer is used to extract the features of the current switch fusion state data. The features extracted by the convolution layer are enhanced using the attention mechanism to obtain enhanced switch fusion state features.

[0013] The enhanced knife switch fusion state features are predicted in the short term through the fully connected layer to obtain the knife switch short-term prediction state data.

[0014] Further, the image data, vibration data, temperature data and sound data of the knife switch state data at the current moment are preprocessed and fused to obtain the knife switch fusion state data at the current moment, including:

[0015] Using linear interpolation method, the image data, vibration data, temperature data and sound data with different time stamps are aligned to the same time axis to obtain the image data, vibration data, temperature data and sound data on the same time axis;

[0016] The vibration data on the same time axis are denoised using a low-pass filter to obtain denoised vibration data;

[0017] The temperature data and the sound data on the same time axis are smoothed and denoised using a median filter to obtain denoised temperature data and sound data;

[0018] The image data, denoised vibration data, denoised temperature data and sound data are calibrated and feature vectors are extracted, and the extracted feature vectors are weighted fused to obtain the switch fusion state data at the current moment.

[0019] Furthermore, the model training of the TCN-diffusion model includes:

[0020] Obtain the switch status data at each historical moment, and perform preprocessing and fusion processing on the switch status data at each historical moment to obtain the switch fusion status data at each historical moment;

[0021] The feature data of the knife switch fusion state at each historical moment are extracted through the time series convolution layer to obtain the feature data of the knife switch fusion state at each historical moment;

[0022] Through the positive diffusion process in the diffusion layer of the TCN-diffusion model, positive diffusion noise is added to the knife switch fusion state feature data at each historical moment to obtain the positive diffusion knife switch state data;

[0023] Select the state data of the positive diffusion knife switch at a certain moment, set a fixed time length, and make a long-term prediction of the state data of the positive diffusion knife switch at the selected moment for a fixed time length through the reverse diffusion process in the diffusion layer of the TCN-diffusion model to obtain the long-term prediction state data of the knife switch for a fixed time length;

[0024] According to the long-term predicted state data of the knife switch of a fixed time length and the real knife switch state data, the loss value is calculated using the mean square error loss function; wherein the real knife switch state data is the knife switch fusion state data at the historical moment corresponding to the fixed time length;

[0025] According to the loss value, the Adam optimization algorithm is used to optimize the model parameters of the TCN-diffusion model;

[0026] Repeat the model training until the loss value converges to obtain the trained TCN-diffusion model.

[0027] Furthermore, the model training of the probability-based Bayesian anomaly detection model includes:

[0028] Obtain historical fault data and historical normal data of the knife switch, and package the historical fault data and historical normal data of the knife switch and divide them into a training set and a verification set;

[0029] Perform data preprocessing and feature extraction on the historical fault data and historical normal data of the training set to obtain the historical fault feature data and historical normal feature data of the switch status;

[0030] Calculate the fault feature probability and prior probability of the switch state based on the historical fault data, historical normal data, historical fault feature data and historical normal feature data of the switch state;

[0031] The fault feature probability, normal feature probability and fault prior probability of the switch state are used as model parameters of the Bayesian anomaly detection model to build a Bayesian anomaly detection model;

[0032] The accuracy of the Bayesian anomaly detection model is evaluated through the validation set to obtain the accuracy evaluation result;

[0033] According to the accuracy evaluation results, the model parameters of the Bayesian anomaly detection model are adjusted and the accuracy is re-evaluated until the accuracy reaches the preset accuracy, thereby obtaining a trained Bayesian anomaly detection model.

[0034] Furthermore, after inputting the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculating the predicted probability of the knife switch failure, and determining the knife switch state result based on the predicted probability and the preset failure probability threshold, the method further includes:

[0035] Obtain the knife switch status result. When the knife switch status result is a fault, input the knife switch long-term prediction status data into the trained fault classification model based on support vector machine, so that the fault classification model based on support vector machine performs fault classification and identification according to the knife switch long-term prediction status data to obtain the fault type of the knife switch long-term prediction status.

[0036] Based on the above method embodiment, the present invention provides a corresponding system embodiment;

[0037] An embodiment of the present invention provides a switch state monitoring system, including: a data acquisition module, a short-term prediction module, a long-term prediction module and a fault probability analysis module;

[0038] The data acquisition module is used to obtain the knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data;

[0039] The short-term prediction module is used to input the knife switch state data at the current moment into the trained short-term prediction model, so that the short-term prediction model generates the knife switch short-term prediction state data according to the knife switch state data at the current moment;

[0040] The long-term prediction module is used to input the short-term prediction state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term prediction state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and the diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term prediction state data of the knife switch;

[0041] The fault probability analysis module is used to input the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculate the predicted probability of the knife switch failure, and obtain the knife switch state result based on the predicted probability and the preset fault probability threshold.

[0042] Furthermore, after the fault probability analysis module, it also includes: a fault type classification module;

[0043] The fault type classification module is used to obtain the knife switch status result. When the knife switch status result is a fault, the long-term predicted state data of the knife switch is input into the trained fault classification model based on the support vector machine, so that the fault classification model based on the support vector machine performs fault classification and identification according to the long-term predicted state data of the knife switch to obtain the fault type of the long-term predicted state of the knife switch.

[0044] Based on the above-mentioned method embodiment, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, a knife switch status monitoring method as described in the present invention is implemented.

[0045] Based on the above method embodiment, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute a knife switch status monitoring method as described in the present invention when the computer program is running.

[0046] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0047] The present invention firstly performs a short-term state prediction operation on the knife switch through a three-dimensional convolutional neural network model to obtain the short-term predicted state characteristics of the knife switch, then performs a long-term state prediction operation on the short-term predicted state characteristics of the knife switch through a TCN-diffusion model to obtain the long-term predicted state characteristics of the knife switch, and finally performs a fault probability analysis on the long-term predicted state characteristics of the knife switch through a probability-based Bayesian anomaly detection model to obtain the probability of the knife switch failure. The state prediction of the knife switch is performed by combining the short-term state prediction model and the long-term state prediction model to achieve the prediction of the long-term state of the knife switch, improve the accuracy of the predicted state of the knife switch, provide a guarantee for the subsequent fault analysis of the predicted state of the knife switch, and thus improve the accuracy of the fault analysis of the knife switch. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 : A flow chart of the steps of a switch state monitoring method provided by an embodiment of the present invention;

[0049] Figure 2 : A system structure diagram of a knife switch status monitoring system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] In the description of the present invention, it should be understood that the terms “first” and “second” are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0052] Embodiment 1:

[0053] Please refer to Figure 1 , is a flow chart of a method for monitoring a knife switch state provided by an embodiment of the present invention, the method at least comprising the following steps:

[0054] Step S1: obtaining the knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data;

[0055] In this embodiment, a light field camera, a vibration sensor, a temperature sensor and a sound sensor are deployed near the knife switch to obtain image data, vibration data, temperature data and sound data of the knife switch state at the current moment and each historical moment; wherein the light field camera may include but is not limited to installing a metalens array light field camera at a key position of the knife switch, and setting the acquisition frequency of the metalens array light field camera to 10 frames per second to collect image data of the knife switch state; the image data of the knife switch state is represented by the following formula: L(u,v,s,t)=∫I(x,y,θ,φ)dΩ wherein (u,v) represents the plane coordinates of the light field image, (s,t) represents the angular coordinates of the light field image, I(x,y,θ,φ) represents the intensity of light at the spatial position (x,y) and direction (θ,φ), and dΩ is the solid angle element.

[0056] Step S2: inputting the current switch state data into the trained short-term prediction model, so that the short-term prediction model generates the switch short-term prediction state data according to the current switch state data;

[0057] In this embodiment, the knife switch status data at the current moment is input into the input layer of the short-term prediction model, and several convolution layers of the short-term prediction model use a 3D convolution layer with a convolution kernel size of 3*3*3 to extract features of the knife switch status data at the current moment in the input layer of the short-term prediction model. The extracted features are enhanced through the attention mechanism module of each convolution layer to obtain enhanced features, and the enhanced features are input into the fully connected layer for short-term prediction processing, and the short-term prediction status data of the knife switch is output.

[0058] In this embodiment, the model training of the short-term prediction model includes:

[0059] Obtain the switch status data at each historical moment, and perform data preprocessing and fusion processing on the switch status data at each historical moment to obtain the fused switch status data at each historical moment;

[0060] The fused knife switch status data at each historical moment is divided into a training set and a validation set, and the training set is input into the short-term prediction model. The convolution layer of the short-term prediction model is used to extract features from the input data, and the attention mechanism module is used to enhance the features extracted by the convolution layer. The fully connected layer is used to perform short-term prediction processing on the enhanced features output by the attention mechanism module to obtain the knife switch short-term prediction status data.

[0061] According to the short-term prediction state data of the knife switch and the real state data of the knife switch at the historical moment, the loss value of the short-term prediction model is calculated using the weighted cross entropy loss function;

[0062] According to the loss value, the Adam optimization algorithm is used to adjust the model parameters of the short-term prediction model, and the short-term prediction model training is repeated until the loss value converges to obtain a converged short-term prediction model;

[0063] The validation set is used to evaluate the model performance of the converged short-term prediction model to obtain a first performance evaluation result, and the hyperparameters of the short-term prediction model are adjusted according to the first performance evaluation result to obtain a trained short-term prediction model.

[0064] In this embodiment, the weighted cross entropy loss function is specifically:

[0065]

[0066] Among them, w i is the category weight, y i is the true label, p i is the predicted probability.

[0067] In this embodiment, the Adam optimization algorithm is specifically:

[0068]

[0069] in, and are momentum estimation and quadratic momentum estimation respectively, and ∈ is the smoothing coefficient.

[0070] Step S3: input the short-term predicted state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term predicted state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term predicted state data of the knife switch;

[0071] In this embodiment, the model training of the TCN-diffusion model includes:

[0072] Obtain the switch status data at each historical moment, and perform preprocessing and fusion processing on the switch status data at each historical moment to obtain the switch fusion status data at each historical moment;

[0073] The feature data of the knife switch fusion state at each historical moment are extracted through the time series convolution layer to obtain the feature data of the knife switch fusion state at each historical moment;

[0074] Through the positive diffusion process in the diffusion layer of the TCN-diffusion model, positive diffusion noise is added to the knife switch fusion state feature data at each historical moment to obtain the positive diffusion knife switch state data;

[0075] Select the state data of the positive diffusion knife switch at a certain moment, set a fixed time length, and make a long-term prediction of the state data of the positive diffusion knife switch at the selected moment for a fixed time length through the reverse diffusion process in the diffusion layer of the TCN-diffusion model to obtain the long-term prediction state data of the knife switch for a fixed time length;

[0076] According to the long-term predicted state data of the knife switch of a fixed time length and the real knife switch state data, the loss value is calculated using the mean square error loss function; wherein the real knife switch state data is the knife switch fusion state data at the historical moment corresponding to the fixed time length;

[0077] According to the loss value, the Adam optimization algorithm is used to optimize the model parameters of the TCN-diffusion model;

[0078] Repeat the model training until the loss value converges to obtain the trained TCN-diffusion model.

[0079] In this embodiment, the time series convolution layer uses 1D convolution to process time series data, the convolution kernel size is k, the step size is s, and the feature C(t) is extracted, which is specifically expressed as follows:

[0080]

[0081] Among them, W i represents the convolution kernel weight, and b represents the bias value.

[0082] In this embodiment, the forward diffusion process of the diffusion layer of the TCN-diffusion model is expressed as follows:

[0083]

[0084] The reverse diffusion process of the diffusion layer of the TCN-diffusion model is expressed as follows:

[0085]

[0086] Among them, x t represents the state at time t, β t represents the diffusion coefficient, ∈ t represents the noise term.

[0087] In this embodiment, the mean square error loss function is specifically:

[0088]

[0089] Among them, y i Indicates the real status data of the knife switch. Represents the long-term predicted status data of the knife switch.

[0090] Step S4: input the long-term predicted state data of the knife switch into the trained probability-based Bayesian anomaly detection model, calculate the predicted probability of the knife switch failure, and obtain the knife switch state result based on the predicted probability and the preset failure probability threshold.

[0091] In this embodiment, the model training of the probability-based Bayesian anomaly detection model includes:

[0092] Obtain historical fault data and historical normal data of the knife switch, and package the historical fault data and historical normal data of the knife switch and divide them into a training set and a verification set;

[0093] Perform data preprocessing and feature extraction on the historical fault data and historical normal data of the training set to obtain the historical fault feature data and historical normal feature data of the switch status;

[0094] Calculate the fault feature probability and prior probability of the switch state based on the historical fault data, historical normal data, historical fault feature data and historical normal feature data of the switch state;

[0095] The fault feature probability, normal feature probability and fault prior probability of the switch state are used as model parameters of the Bayesian anomaly detection model to build a Bayesian anomaly detection model;

[0096] The accuracy of the Bayesian anomaly detection model is evaluated through the validation set to obtain the accuracy evaluation result;

[0097] According to the accuracy evaluation results, the model parameters of the Bayesian anomaly detection model are adjusted and the accuracy is re-evaluated until the accuracy reaches the preset accuracy, thereby obtaining a trained Bayesian anomaly detection model.

[0098] In this embodiment, the data preprocessing and fusion processing include:

[0099] Using linear interpolation method, the image data, vibration data, temperature data and sound data with different time stamps are aligned to the same time axis to obtain the image data, vibration data, temperature data and sound data on the same time axis;

[0100] The vibration data on the same time axis are denoised using a low-pass filter to obtain denoised vibration data;

[0101] The temperature data and the sound data on the same time axis are smoothed and denoised using a median filter to obtain denoised temperature data and sound data;

[0102] The image data, denoised vibration data, denoised temperature data and sound data are calibrated and feature vectors are extracted, and the extracted feature vectors are weighted fused to obtain the switch fusion state data.

[0103] In this embodiment, the linear interpolation method is used to align the image data, vibration data, temperature data and sound data of different time stamps to the same time axis to obtain the image data, vibration data, temperature data and sound data of the same time axis. Specifically, the linear interpolation method is used to align the image data of different time stamps to the same time axis T, which is set as the timestamp of the i-th sensor, t i =t i-1 +Δt i Among them, Δt i is the sampling interval of the i-th sensor, the sampling interval of the same time axis T is ΔT, and linear interpolation is performed on each timestamp t_i:

[0104]

[0105] Among them, X(t i ) indicates that the sensor is at timestamp t i Collected data.

[0106] In this embodiment, the vibration data on the same time axis are denoised using a low-pass filter to obtain denoised vibration data, specifically:

[0107]

[0108] Where h(t) is the impulse response function of the filter, usually a Gaussian filter: σ is the standard deviation of the filter.

[0109] In this embodiment, the temperature data and sound data on the same time axis are smoothed and denoised using a median filter to obtain denoised temperature data and sound data, specifically:

[0110] X filtered (t)=median{X(tk),..,X(t),...,X(t+k)}

[0111] Where k is the filter window size.

[0112] In this embodiment, the image data, denoised vibration data, denoised temperature data and sound data are calibrated and feature vectors are extracted, and the extracted feature vectors are weighted fused to obtain the switch fusion state data, specifically:

[0113] The calibration includes zero drift calibration and gain calibration; the zero drift calibration formula is: is the average value of the sensor under static conditions, determined by static testing; the gain calibration formula is G is the gain of the sensor, obtained by calibration using a standard reference signal.

[0114] The feature vector extraction includes image data feature vector extraction, vibration spectrum feature vector extraction, temperature change rate feature vector extraction and sound pressure feature vector extraction;

[0115] The vibration spectrum feature vector extraction is to obtain the vibration spectrum feature by performing frequency domain conversion on the vibration data through fast Fourier transform;

[0116] The sound pressure feature vector extraction is to obtain the sound pressure feature by performing frequency domain conversion on the sound data through short-time Fourier transform;

[0117] The image data feature vector, vibration spectrum feature vector, temperature change rate feature vector and pressure feature vector are weightedly added to obtain the switch fusion state data.

[0118] In this embodiment, after inputting the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculating the predicted probability of the knife switch failure, and determining the knife switch state result according to the predicted probability and the preset failure probability threshold, the method further includes:

[0119] Obtain the knife switch status result. When the knife switch status result is a fault, input the knife switch long-term prediction status data into the trained fault classification model based on support vector machine, so that the fault classification model based on support vector machine performs fault classification and identification according to the knife switch long-term prediction status data to obtain the fault type of the knife switch long-term prediction status.

[0120] In this embodiment, the fault types of the long-term predicted state of the knife switch include mechanical faults, electrical faults, temperature faults, vibration faults and acoustic faults; wherein the mechanical faults include but are not limited to knife switch jamming, mechanical parts wear and device loosening; the electrical faults include but are not limited to poor contact, insulation damage and arc faults; the temperature faults include but are not limited to knife switch overheating and temperature sensor failure; the vibration faults include but are not limited to abnormal vibration and vibration sensor failure; the acoustic faults include but are not limited to abnormal noise and acoustic sensor failure.

[0121] In this embodiment, the model training of the fault classification model based on the support vector machine includes:

[0122] Obtain the historical fault data and historical normal data of the knife switch;

[0123] Performing data preprocessing and feature extraction on the historical fault data and historical normal data of the knife switch to obtain the historical fault feature data and historical normal feature data of the knife switch;

[0124] The historical fault feature data and historical normal feature data of the switch are packaged and divided into a training set and a validation set;

[0125] The training set is input into the fault classification model based on support vector machine to classify the faults and obtain the fault type of the long-term prediction state of the switch;

[0126] According to the fault classification results and the actual fault classification results, the fault classification function is used to calculate the loss value, and the model parameters of the fault classification model are optimized according to the loss value. The model training is repeated until the loss value converges to obtain a converged fault classification model;

[0127] The converged fault classification model is evaluated for model performance through a validation set to obtain a second performance evaluation result. According to the second performance evaluation result, the hyperparameters of the fault classification model are adjusted to obtain a trained fault classification model based on a support vector machine.

[0128] In this embodiment, after the fault classification model based on the support vector machine performs fault classification and identification according to the long-term predicted state data of the knife switch to obtain the fault type of the long-term predicted state of the knife switch, it also includes:

[0129] Inputting the fault type of the long-term predicted state of the knife switch into the decision tree model, so that the decision function of the decision tree model generates a decision value for the fault type of the long-term predicted state of the knife switch, and determining the maintenance measures of the knife switch according to the decision value;

[0130] The maintenance measures are prioritized according to the severity of the fault, the scope of impact and the difficulty of repair using a priority scoring function to obtain the optimal maintenance measure;

[0131] The operator collects feedback information based on the optimal maintenance measures and optimizes the decision tree model by updating the feedback mechanism based on the feedback information;

[0132] The decision function is specifically: x is the input feature vector, M is the number of decision tree nodes, and I is the indicator function to determine whether the input feature belongs to region R j , a j For region R j The decision value of

[0133] The priority scoring function is specifically: S = w1·S severity +w2·S impact +w3·S dificulty , S severity Score the fault severity, S impact Score the impact range, S difficulty is the repair difficulty score, w1, w2 and w3 are weight parameters;

[0134] The update feedback mechanism is specifically: t+1 =αF t +(1-α)O t , F t+1 is the updated feedback, α is the smoothing coefficient, O t This is the current feedback information.

[0135] In this embodiment, the switch state detection method further includes: establishing a data visualization interactive interface; wherein the data visualization interactive interface includes: user interface settings, data display, playback and analysis, fault information display, maintenance measures display, maintenance measures sorting, touch screen interaction, voice control and custom settings;

[0136] The user interface setting includes setting a multi-pane user interface to organically integrate various functional modules of the switch status monitoring; the multi-pane user interface is divided into four main areas: a real-time status display area, a historical data analysis area, a fault warning area, and a maintenance suggestion area;

[0137] The data display includes displaying the switch status data at the current moment and at each historical moment; the switch status data includes real-time graphs of light field images, vibration, temperature and sound data, and uses color coding and icons to identify different states (such as normal, warning, fault)

[0138] The playback and analysis include that the user can select a specific time period to view the change trend of the knife switch status, and can use the interactive chart to zoom in, zoom out and drag the chart to view detailed data;

[0139] The fault information display includes displaying the current fault warning information, including the fault type, occurrence time, impact range and maintenance measures, and providing a warning detail viewing function, where the user can click on the warning information to view detailed sensor data and fault analysis results;

[0140] The maintenance measure display includes displaying maintenance measures; the maintenance measures include but are not limited to maintenance steps, tools required for maintenance and estimated maintenance time;

[0141] The maintenance measures sorting includes sorting the maintenance measures according to priority, so that users can quickly identify and handle emergency faults;

[0142] The touch screen interaction includes providing touch screen operation, and the user can adjust the chart and data view through the gesture control interface; the gesture operation includes but is not limited to sliding, clicking, double-clicking and long pressing;

[0143] The voice control includes controlling the interface using integrated voice recognition technology according to the user's voice command, and can prompt the user of the current operation status and results through voice; wherein the control includes but is not limited to querying the real-time status, historical data, fault warning and maintenance measures of the knife switch;

[0144] The customized settings include supporting the customization of the size and position of each area of ​​the multi-pane user interface, and supporting the customization of warning thresholds and maintenance suggestion templates to meet the personalized needs of different users.

[0145] In this embodiment, the knife switch state detection method further includes: regularly collecting user feedback data and knife switch state data, and optimizing each model and data visualization interaction interface according to the user feedback data and knife switch state data;

[0146] In this embodiment, the periodic collection of user feedback data and the optimization of the interactive interface of each model and data visualization according to the user feedback data and the switch status data include:

[0147] Collect user feedback data and switch status data, and summarize and store them in the database. Use the timestamp synchronization mechanism to align the timestamps of the collected feedback data with the real-time monitoring data to obtain aligned feedback data and switch status data. Perform data cleaning and data fusion on the aligned feedback data and switch status data to obtain fused feedback data. The data cleaning formula is: X(t) is the original data, μ x and σ x are the mean and standard deviation of the data respectively;

[0148] The prediction accuracy and fault recognition rate of the probability-based Bayesian anomaly detection model are calculated by using the prediction accuracy calculation formula and the fault recognition rate calculation formula to obtain a model performance report; wherein the prediction accuracy formula is: The fault recognition rate calculation formula is:

[0149] The fused feedback data is input into the probability-based Bayesian anomaly detection model for re-model training, and the model parameters are updated using the gradient descent method and the learning rate adjustment formula; wherein the gradient descent method is specifically: θ t is the model parameter, η is the learning rate, L(θ t ) is the loss function; the learning rate adjustment formula is specifically: λ is the learning rate decay factor;

[0150] According to the user feedback data, a user feedback report is generated using a user feedback quantification formula; wherein the user feedback quantification formula is specifically: r i is the user feedback rating, e i is the system output score, N is the number of feedbacks;

[0151] According to the user feedback report and model performance report, the interactive interface of data visualization is optimized using the interface parameter adjustment formula; wherein the interface parameter adjustment formula is specifically: θ new =θ old +δθ, δθ is the parameter change adjusted according to user feedback.

[0152] In this embodiment, the knife switch status monitoring method further includes: encrypting user feedback data and knife switch status data, and encrypting data transmission.

[0153] In this embodiment, the step of encrypting the user feedback data and the switch status data and encrypting the data transmission includes:

[0154] The user feedback data and the switch status data are encrypted using the encryption formula of the Advanced Encryption Standard (AES-256) and the key management system (KMS) to obtain the encrypted data and the encryption key, and the encrypted data and the encryption key are stored using the distributed storage technology and the backup strategy; wherein the encryption formula is: C = E k (M)C is the encrypted ciphertext;

[0155] The transmission encryption formula of the Transport Layer Security Protocol (TLS1.3) is used to encrypt data transmission, and the identity and data transmission of both parties in data communication are verified using a certificate-based two-way authentication mechanism; the transmission encryption formula is: C t It is the data encrypted during transmission. is the transmission encryption function, M t is the transmitted data, k t is the transmission key;

[0156] Use access control lists (ACLs) and role-based access control (RBAC) to manage permissions for system users and devices, and use fine-grained access control (FGAC) to manage access permissions for data;

[0157] Establish a logging system to record all data access, transmission and operation behaviors, formulate data protection strategy, access control strategy and emergency response strategy for the logging system, and regularly analyze and audit the logging system; wherein, the emergency response strategy includes data leakage emergency plan, system intrusion detection and emergency response process; the design process of the emergency response process is detection → analysis → isolation → repair → recovery → summary, and the operation guide and responsible person information are stored in each step.

[0158] Embodiment 2:

[0159] Please refer to Figure 2 , is a system structure diagram of a switch state monitoring system provided by an embodiment of the present invention, the system comprising: a data acquisition module, a short-term prediction module, a long-term prediction module and a fault probability analysis module;

[0160] The data acquisition module is used to obtain the knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data;

[0161] The short-term prediction module is used to input the knife switch state data at the current moment into the trained short-term prediction model, so that the short-term prediction model generates the knife switch short-term prediction state data according to the knife switch state data at the current moment;

[0162] The long-term prediction module is used to input the short-term prediction state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term prediction state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and the diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term prediction state data of the knife switch;

[0163] The fault probability analysis module is used to input the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculate the predicted probability of the knife switch failure, and obtain the knife switch state result based on the predicted probability and the preset fault probability threshold.

[0164] In this embodiment, after the fault probability analysis module, it also includes: a fault type classification module;

[0165] The fault type classification module is used to obtain the knife switch status result. When the knife switch status result is a fault, the long-term predicted state data of the knife switch is input into the trained fault classification model based on the support vector machine, so that the fault classification model based on the support vector machine performs fault classification and identification according to the long-term predicted state data of the knife switch to obtain the fault type of the long-term predicted state of the knife switch.

[0166] In this embodiment, the short-term prediction module includes a first preprocessing fusion unit and a switch short-term state prediction unit;

[0167] The first preprocessing and fusing unit is used to preprocess and fuse the image data, vibration data, temperature data and sound data of the knife switch state data at the current moment to obtain the knife switch fusion state data at the current moment;

[0168] The knife switch short-term state prediction unit is used to obtain the knife switch fusion state data at the current moment through the input layer of the short-term prediction model, use the convolution layer to extract features of the knife switch fusion state data at the current moment, use the attention mechanism to enhance the features extracted by the convolution layer to obtain enhanced knife switch fusion state features, and perform short-term prediction on the enhanced knife switch fusion state features through the fully connected layer to obtain the knife switch short-term prediction state data.

[0169] In this embodiment, the long-term prediction module includes a second preprocessing fusion unit and a knife switch long-term state prediction unit;

[0170] The second preprocessing and fusion unit is used to preprocess and fuse the knife switch state data at each historical moment to obtain the knife switch fusion state data at each historical moment;

[0171] The knife switch long-term state prediction unit is used to extract features of the knife switch short-term prediction state data and the knife switch fusion state data at each historical moment through the temporal convolution layer and diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the knife switch long-term prediction state data.

[0172] Based on the above method embodiment, another embodiment is provided;

[0173] Another embodiment of the present invention provides a knife switch status monitoring terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements a knife switch status monitoring method described in any one of the above method embodiments of the present invention.

[0174] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the switch status monitoring terminal device.

[0175] The switch state monitoring terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The switch state monitoring terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art may understand that, for example, the switch state monitoring terminal device may also include an input / output device, a network access device, a bus, etc.

[0176] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the knife switch state monitoring terminal device, and uses various interfaces and lines to connect the various parts of the entire knife switch state monitoring terminal device.

[0177] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the knife switch status monitoring terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0178] Based on the above method embodiment, another embodiment is provided;

[0179] Another embodiment of the present invention provides a storage medium, comprising a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a knife switch status monitoring method described in any one of the above method embodiments of the present invention.

[0180] Wherein, the above-mentioned storage medium is a computer-readable storage medium. If the module / unit integrated with the switch state monitoring system / terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0181] It should be noted that the above-mentioned terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the above-mentioned terminal device is merely an example and does not constitute a limitation on the terminal device. It may include more or fewer components, or a combination of certain components, or different components.

[0182] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring knife switch status, characterized in that: include: Obtaining knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data; Inputting the current switch state data into the trained short-term prediction model, so that the short-term prediction model generates the switch short-term prediction state data according to the current switch state data; Input the short-term prediction state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term prediction state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term prediction state data of the knife switch; The long-term predicted status data of the knife switch is input into the trained probability-based Bayesian anomaly detection model to calculate the predicted probability of knife switch failure. The knife switch status result is obtained based on the predicted probability and the preset failure probability threshold.

2. A method for monitoring knife switch status according to claim 1, characterized in that: The step of inputting the current switch state data into the trained short-term prediction model so that the short-term prediction model generates the switch short-term prediction state data according to the current switch state data includes: Preprocessing and fusing the image data, vibration data, temperature data and sound data of the knife switch state data at the current moment to obtain the knife switch fusion state data at the current moment; The current switch fusion state data is obtained through the input layer of the short-term prediction model, and the convolution layer is used to extract the features of the current switch fusion state data. The features extracted by the convolution layer are enhanced using the attention mechanism to obtain enhanced switch fusion state features. The enhanced knife switch fusion state features are predicted in the short term through the fully connected layer to obtain the knife switch short-term prediction state data.

3. A method for monitoring knife switch status according to claim 2, characterized in that: The preprocessing and fusion processing of the image data, vibration data, temperature data and sound data of the knife switch state data at the current moment to obtain the knife switch fusion state data at the current moment includes: Using linear interpolation method, the image data, vibration data, temperature data and sound data with different time stamps are aligned to the same time axis to obtain the image data, vibration data, temperature data and sound data on the same time axis; The vibration data on the same time axis are denoised using a low-pass filter to obtain denoised vibration data; The temperature data and the sound data on the same time axis are smoothed and denoised using a median filter to obtain denoised temperature data and sound data; The image data, denoised vibration data, denoised temperature data and sound data are calibrated and feature vectors are extracted, and the extracted feature vectors are weighted fused to obtain the switch fusion state data at the current moment.

4. A method for monitoring knife switch status according to claim 1, characterized in that: The model training of the TCN-diffusion model includes: Obtain the switch status data at each historical moment, and perform preprocessing and fusion processing on the switch status data at each historical moment to obtain the switch fusion status data at each historical moment; The feature data of the knife switch fusion state at each historical moment are extracted through the time series convolution layer to obtain the feature data of the knife switch fusion state at each historical moment; Through the positive diffusion process in the diffusion layer of the TCN-diffusion model, positive diffusion noise is added to the knife switch fusion state feature data at each historical moment to obtain the positive diffusion knife switch state data; Select the state data of the positive diffusion knife switch at a certain moment, set a fixed time length, and make a long-term prediction of the state data of the positive diffusion knife switch at the selected moment for a fixed time length through the reverse diffusion process in the diffusion layer of the TCN-diffusion model to obtain the long-term prediction state data of the knife switch for a fixed time length; According to the long-term predicted state data of the knife switch of a fixed time length and the real knife switch state data, the loss value is calculated using the mean square error loss function; wherein the real knife switch state data is the knife switch fusion state data at the historical moment corresponding to the fixed time length; According to the loss value, the Adam optimization algorithm is used to optimize the model parameters of the TCN-diffusion model; Repeat the model training until the loss value converges to obtain the trained TCN-diffusion model.

5. A knife switch state detection method according to claim 1, characterized in that: The model training of the probability-based Bayesian anomaly detection model includes: Obtain historical fault data and historical normal data of the knife switch, and package the historical fault data and historical normal data of the knife switch and divide them into a training set and a verification set; Perform data preprocessing and feature extraction on the historical fault data and historical normal data of the training set to obtain the historical fault feature data and historical normal feature data of the switch status; Calculate the fault feature probability and prior probability of the switch state based on the historical fault data, historical normal data, historical fault feature data and historical normal feature data of the switch state; The fault feature probability, normal feature probability and fault prior probability of the switch state are used as model parameters of the Bayesian anomaly detection model to build a Bayesian anomaly detection model; The accuracy of the Bayesian anomaly detection model is evaluated through the validation set to obtain the accuracy evaluation result; According to the accuracy evaluation results, the model parameters of the Bayesian anomaly detection model are adjusted and the accuracy is re-evaluated until the accuracy reaches the preset accuracy, thereby obtaining a trained Bayesian anomaly detection model.

6. A knife switch state detection method according to claim 1, characterized in that: After inputting the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculating the predicted probability of the knife switch failure, and determining the knife switch state result based on the predicted probability and the preset failure probability threshold, the method further includes: Obtain the knife switch status result. When the knife switch status result is a fault, input the knife switch long-term prediction status data into the trained fault classification model based on support vector machine, so that the fault classification model based on support vector machine performs fault classification and identification according to the knife switch long-term prediction status data to obtain the fault type of the knife switch long-term prediction status.

7. A switch status monitoring system, characterized in that: include: Data acquisition module, short-term prediction module, long-term prediction module and fault probability analysis module; The data acquisition module is used to obtain the knife switch status data at the current moment and at each historical moment; wherein the knife switch status data includes image data, vibration data, temperature data and sound data; The short-term prediction module is used to input the knife switch state data at the current moment into the trained short-term prediction model, so that the short-term prediction model generates the knife switch short-term prediction state data according to the knife switch state data at the current moment; The long-term prediction module is used to input the short-term prediction state data of the knife switch and the knife switch state data at each historical moment into the trained TCN-diffusion model, extract features of the short-term prediction state data of the knife switch and the knife switch state data at each historical moment through the time series convolution layer and the diffusion model layer of the TCN-diffusion model, and perform long-term prediction based on the extracted features to obtain the long-term prediction state data of the knife switch; The fault probability analysis module is used to input the long-term predicted state data of the knife switch into the probability-based Bayesian anomaly detection model, calculate the predicted probability of the knife switch failure, and obtain the knife switch state result based on the predicted probability and the preset fault probability threshold.

8. A knife switch status monitoring system according to claim 7, characterized in that: After the fault probability analysis module, it also includes: a fault type classification module; The fault type classification module is used to obtain the knife switch status result. When the knife switch status result is a fault, the long-term predicted state data of the knife switch is input into the trained fault classification model based on the support vector machine, so that the fault classification model based on the support vector machine performs fault classification and identification according to the long-term predicted state data of the knife switch to obtain the fault type of the long-term predicted state of the knife switch.

9. A knife switch status monitoring terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements a knife switch status monitoring method as described in any one of claims 1 to 6 when executing the computer program.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a knife switch status monitoring method as described in any one of claims 1 to 6.

Images