Remote monitoring and control method for data of power system

Abstract

The invention discloses an electric power system data remote monitoring and control method, and relates to the technical field of electric power system monitoring, and the method comprises the steps: collecting real-time data in the operation process of an electric power system, carrying out the data encryption and compression, and transmitting the data; the data processing center receives the data, and decompresses, decrypts and cleans the data; performing feature learning on the current and voltage data through a convolutional neural network model, and extracting key feature data reflecting the operation state of the power system; clustering the equipment temperature and mechanical vibration data through a clustering algorithm, and identifying abnormal feature data of the equipment operation state; carrying out fault diagnosis, and predicting operation data of the power system in a future period of time by using LSTM (Long Short Term Memory); and when abnormal data is detected or a potential fault is predicted, early warning information is sent out, and the fault position and type are identified. According to the method, the operation state of the power system can be comprehensively and accurately monitored, and the operation safety and reliability of the power system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and particularly relates to a method for remote monitoring and control of power system data. Background Art

[0002] With the continuous expansion of the scale and the increasing complexity of the structure of the power system, ensuring the safe and stable operation of the power system has become a crucial task. During the operation of the power system, data such as current, voltage, equipment temperature, and mechanical vibration contain rich operation status information. Effective collection, processing, and analysis of these data can timely detect potential faults and prevent accidents. At the same time, effective processing of the data during the operation of power equipment can also increase the safety of the circuit operation.

[0003] On the one hand, in the face of the increasingly complex network attack environment, traditional single encryption methods are difficult to ensure the security and integrity of power equipment operation data. Power equipment operation data contains key parameters. Once leaked or tampered with, it may lead to power system failures and even major safety accidents. At the same time, traditional data compression technologies cannot effectively process the massive power equipment operation data, which not only increases the pressure of data transmission but also reduces the efficiency of data storage and processing.

[0004] On the other hand, most traditional power system data monitoring methods rely on the analysis of a single data type or use simple threshold judgment methods. For example, only by monitoring whether the amplitudes of current and voltage exceed the set thresholds to judge whether there are faults in the system. This method cannot fully explore the potential information in the data and is prone to missed or misjudged faults. Moreover, for the temperature and vibration data of equipment, traditional methods often lack effective analysis means and are difficult to detect abnormal states of equipment in advance. In addition, in the process of data processing, traditional methods do not handle noise and outliers finely, affecting the accuracy and reliability of data analysis.

[0005] Therefore, the remote monitoring and control of power system data need to be further studied to improve the safety and reliability of power equipment operation. Summary of the Invention

[0006] The problem to be solved by the present invention is: to provide a method for remote monitoring and control of power system data, which can comprehensively and accurately monitor the operation state of the power system by real-time collecting multi-source data, combining data processing and intelligent algorithms, timely detect faults and make predictions, and improve the safety and reliability of power system operation.

[0007] To solve the above technical problems, the technical solution provided by the present invention is: a method for remote monitoring and control of power system data, including the following steps: S1. Real-time collect the real-time current, voltage, equipment temperature, and mechanical vibration data during the operation of the power system, encrypt the preliminarily processed data through a hybrid encryption algorithm, compress the encrypted data through a compression algorithm based on deep learning, and transmit it to the data processing center; S2. The data processing center decompresses and decrypts the received data, and cleans the data to remove noise, outliers, and duplicate data in the data; S3. Build a convolutional neural network model to perform feature learning on the current and voltage waveform data, and extract key feature data reflecting the operating state of the power system; through a clustering algorithm, perform clustering analysis on the equipment temperature and mechanical vibration data to identify abnormal feature data of the equipment operating state; S4. Adopt the SVM (Support Vector Machine) algorithm to perform fault diagnosis on the current operating state of the power system according to the corresponding relationship between the feature data and the known fault types, and use the LSTM (long short-term memory) to predict the operating data of the power system for a period of time in the future; S5. Set different levels of warning thresholds. When the detected data exceeds the corresponding warning threshold or a potential fault is predicted, send a warning message according to the severity and impact range of the fault, and mark the fault location and type on the monitoring interface.

[0008] Preferably, in step S1, real-time data is collected by deploying different types of sensors at each key node of the power system, and the sensors include: current sensors, voltage sensors, temperature sensors, and vibration sensors.

[0009] Preferably, the preliminarily processed data is encrypted through a hybrid encryption algorithm. First, the data is preliminarily encrypted using a symmetric encryption algorithm, and then the symmetric encryption key is encrypted using an asymmetric encryption algorithm. The specific method is as follows: S101. Key generation: Symmetric encryption key generation: Randomly generate a symmetric encryption key with a length of bits ; Asymmetric encryption key pair generation: Use the RSA algorithm to generate a public key and a private key.

[0010] S102. Symmetric encryption of data: The collected and preliminarily processed data is , and the AES algorithm is used to encrypt the data ; The symmetric encryption function is expressed as , then the encrypted data is: .

[0011] S103. Asymmetric Encryption of Symmetric Encryption Key: Use the public key of the asymmetric encryption algorithm to encrypt the symmetric encryption key ; The asymmetric encryption function is expressed as , then the encrypted symmetric encryption key is: .

[0012] S104. Final Encrypted Data: The final encrypted data for transmission consists of the encrypted data and the encrypted symmetric encryption key , and is expressed as: .

[0013] Preferably, use a compression algorithm based on deep learning to compress the encrypted data and remove redundant information in the data. The specific method is as follows: S111. Preprocess the encrypted data and convert it into a format suitable for input to the deep learning model, that is, for the encrypted byte sequence, convert it into a numerical vector .

[0014] S112. Build a deep learning model: Use an autoencoder as the deep learning model for data compression; the autoencoder consists of an encoder and a decoder; The encoder maps the input data to a low-dimensional latent space to achieve data compression; the encoder consists of multiple neural network layers, with a total of layers, calculate the output of the th layer , and output the vector in the latent space: ; The decoder reconstructs the vector in the latent space into an output similar to the input data ; The decoder also consists of multiple neural network layers, with a total of layers, calculate the output of the th layer and output the reconstructed data: . .

[0015] S113. Model Training: Use the operating data of several power equipment for training. The training objective is to minimize the reconstruction error. Use MSE as the loss function and use the stochastic gradient descent method to update the weights and biases of the encoder and decoder to minimize the loss function. .

[0016] S114. Data Compression: After the model training is completed, use the trained encoder to compress the encrypted data to obtain the vector in the latent space .

[0017] Preferably, in step S2, the data processing center decompresses and decrypts the received data. The specific method is as follows: S201. Data Decompression: At the receiving end, use the trained decoder to decompress the compressed data to obtain the reconstructed data , and perform inverse preprocessing on the reconstructed data to obtain the decrypted data.

[0018] S202. Decryption of the Symmetric Encryption Key: The receiving party uses the private key of the asymmetric encryption algorithm to decrypt the encrypted symmetric encryption key ; The asymmetric decryption function is expressed as , then the decrypted symmetric encryption key is: .

[0019] S203. Decryption of Data: The receiving party uses the decrypted symmetric encryption key to decrypt the encrypted data ; The symmetric decryption function is expressed as , then the decrypted data is: .

[0020] Preferably, in step S3, extract the key features reflecting the operating state of the power system. The specific method is as follows: S301. Intercept the continuously collected current and voltage data into fixed-time window time series segments with a length of T, convert them into two-dimensional matrices, and perform normalization processing.

[0021] S302. Build a convolutional neural network model: The input layer includes: the first convolutional layer and the second convolutional layer. The activation functions of both convolutional layers are selected as ReLU; the first convolutional layer uses a small convolutional kernel with a stride of 1, and the second convolutional layer uses a large convolutional kernel; Max pooling is used to reduce the dimension of the output of the convolutional layer. The pooling window size is 2×1, and the stride is 2; The pooled feature vectors are flattened into one-dimensional vectors and input into the fully connected layer.

[0022] S303. Feature learning: The convolutional kernel is slid on the time series segment, and local patterns are extracted through the weight sharing mechanism. Among them, small convolutional kernels are used to capture high-frequency noise and transient disturbances, and large convolutional kernels are used to capture low-frequency trends and steady-state features; The shallow convolutional layer is used to learn basic features, including the waveform slope and harmonic amplitude within a single period. The convolutional layer is used to learn high-order features through cross-layer connections, including the distortion mode of the multi-period waveform and the phase difference between current and voltage. Branches with different convolutional kernel sizes are built in parallel to extract features at different time scales.

[0023] S304. Feature screening and fusion: The importance scores of features are calculated through L1 regularization, and the features with high contribution to fault diagnosis or clustering analysis are screened out. The feature maps of current and voltage are fused by channel concatenation to generate composite features containing the coupling relationship of electrical quantities, and the key feature vectors reflecting the operating state of the power system are obtained; The key feature vectors reflect the waveform distortion degree, transient disturbance type, steady-state operating parameters, and the time series characteristics of abnormal patterns.

[0024] Preferably, in step S301, the continuously collected current and voltage waveform data are converted into a two-dimensional matrix and normalized. The specific method is as follows: S3011. The original time-domain current signal , voltage signal are band-pass filtered: ; ; Among them, is the impulse response of the band-pass filter.

[0025] S3012. The filtered current signal , voltage signal are segmented by a sliding window: The continuous real-time current and voltage data after filtering are intercepted into fixed-time window time series segments with a length of T at intervals of one second. Each segment corresponds to a two-channel input sample ; The window length is N, the stride is S, and the current window and voltage window signals of the kth window are: ; 。

[0026] S3013. Extract the time-frequency features of the current window signal and the voltage window signal: ; ; Among them, is the window function, is the time index, is the frequency index, is the exponential form of the Fourier transform, representing the value of a complex sinusoidal signal with frequency f at discrete time n.

[0027] S3014. Calculate the spectral amplitude matrix: ; 。

[0028] S3015. Normalize the spectral amplitude matrix.

[0029] S3016. Construct a multi-channel matrix: Combine the two-dimensional matrices of current and voltage into a two-channel matrix 。

[0030] Preferably, in step S3, identify the abnormal modes of the device operation, and the specific method is as follows: Normalize the temperature data and vibration data to eliminate the influence of dimensions; Combine the temperature data and vibration data of each device into a two-dimensional feature vector: ; Among them, is the device 's temperature value, is the device 's vibration value; Set the number of clusters , and randomly select initial cluster centers: 。

[0031] Calculate the Euclidean distance from each data point to each cluster center and assign it to the nearest cluster: ; Among them, is the cluster label to which the data point belongs, is the th cluster center in the According to the data points within the current cluster, recalculate the cluster center: ; Among them, is the set of data points in the th cluster, is the number of data points in the cluster; When the cluster centers no longer change significantly, that is , When it is a preset threshold or the maximum number of iterations is reached, stop the iteration.

[0032] After the iteration is completed, label the cluster tags according to business experience or historical data: If the average temperature of a cluster is significantly higher than the normal threshold or the vibration amplitude exceeds the equipment operation standard, it is determined as an abnormal mode cluster, and the other cluster is a normal mode cluster.

[0033] For the newly collected feature vectors, calculate the cluster to which it belongs: ; If the value is the abnormal cluster label, it is determined that the equipment operation state is abnormal, and record the abnormal feature data collected when the equipment operation state is abnormal.

[0034] S317. Integrate the key feature data and the abnormal feature data into unified feature data to obtain a unified feature vector: Based on the key feature data and the abnormal feature data scale difference, perform secondary normalization on the key feature data and perform feature splicing to obtain the integrated feature: ; Among them, represents the dimension of the Euclidean space, is the numerical encoding of the abnormal label. When it is normal, , when it is abnormal, .

[0035] Preferably, in step S4, perform a fault diagnosis on the current operating state of the power system. The specific method is as follows: S401. Divide the feature data obtained in step S3 into a training set and a test set according to a ratio of 7:3; S402. Construct a non-linear SVM model using the radial basis kernel function, set the value ranges of the hyperparameters C and γ, use 5-fold cross-validation, and use the test set accuracy as the evaluation index to determine the optimal hyperparameters C and γ through grid search; S403. Input the training set into the SVM model with the determined hyperparameters for training. After the training is completed, input the test set into the trained SVM model to obtain the predicted fault type labels for each test sample; S404. For each test sample, compare the predicted fault type label with the actual fault type label. If they are the same, the sample is diagnosed correctly; if they are different, the sample is diagnosed incorrectly. Count the number of samples diagnosed correctly in the test set and calculate the test set accuracy: S405. Input the feature data that has been collected in real time and processed into the trained SVM model. The model outputs the corresponding fault type label. If the output fault type label is one of the known fault types, it is determined that the current power system has this type of fault. If the output fault type label is the normal operation label, it is determined that the current operation of the power system is normal.

[0036] Preferably, in step S4, use LSTM to predict the operation data of the power system for a period of time in the future. The specific method is as follows: S411. Organize the historical operation data of the power system into a one-dimensional time series, and then use the sliding window method to convert the series into input-output pairs according to the set window size and prediction step length. At the same time, normalize the data. S412. Build an LSTM network. The dimension of the input layer is set according to the data characteristics. The hidden layer consists of multiple LSTM units, and the output layer is a fully connected layer. The number of neurons corresponds to the prediction step length. Capture the long-term dependence relationship of the data through the forget gate, input gate, output gate, and cell state update mechanism. S413. Use the mean square error as the loss function and Adam as the optimizer to train the model with the training data, and adjust the model parameters through the backpropagation algorithm. S414. Input the data collected in real time into the trained model for prediction, and inverse normalize the prediction result to restore it to the original data scale.

[0037] Preferably, in step S5, the data processing center sends warning information to relevant personnel through one or more of SMS, email, and acoustic-optical alarms.

[0038] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects: 1. By collecting multi-source data such as current, voltage, equipment temperature, and mechanical vibration, and performing comprehensive processing and analysis, the present invention can comprehensively and accurately reflect the operation state of the power system. Compared with the traditional analysis method of single data type, the accuracy and reliability of monitoring are greatly improved.

[0039] 2. The method of the present invention uses convolutional neural network, clustering algorithm, SVM algorithm, and LSTM to deeply mine and analyze the data. The convolutional neural network can effectively extract the key features of the current and voltage waveforms. The clustering algorithm can accurately identify the abnormal patterns of the equipment operation state. The SVM algorithm realizes accurate fault diagnosis, and LSTM can reliably predict the future operation data of the power system, improving the accuracy of fault diagnosis and prediction.

[0040] 3. By setting the warning threshold, the present invention can timely send warning messages before or when a fault occurs, and clearly mark the fault location and type on the monitoring interface, facilitating the maintenance personnel to quickly locate the fault and take corresponding measures, reducing the impact of the fault on the operation of the power system, and improving the safety and stability of the power system.

[0041] 4. The method of the present invention can automatically complete tasks such as data collection, processing, analysis, diagnosis, and prediction, reducing manual intervention, improving the maintenance efficiency, and reducing the maintenance cost. Description of the Drawings

[0042] Figure 1 is the flow chart of the method for remotely monitoring and controlling power system data of the present invention; Figure 2 is the block diagram of the system for remotely monitoring and controlling power system data of the present invention. Detailed Embodiments

[0043] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the application will be further elaborated in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments made by other researchers in the field based on this embodiment fall within the protection scope of the present invention. At the same time, for the step numbers in the embodiments of the present invention, they are only set for the convenience of elaboration and explanation, and no limitation is made on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0044] In an embodiment of the present invention, a method for remotely monitoring and controlling power system data, as Figure 1 shown, includes the following steps:

[0045] Step 1: Real-time collection and transmission of multi-source data

[0046] First, deploy current sensors, voltage sensors, temperature sensors, and vibration sensors at key nodes of the power system (such as transformers, circuit breakers, and transmission line interfaces).

[0047] Current monitoring: Use a non-intrusive current sensor based on the Rogowski coil principle to achieve non-contact measurement of alternating current and avoid damaging the original circuit structure.

[0048] Voltage monitoring: A voltage sensor based on capacitive voltage division technology is used to obtain high-precision voltage signals in real time, which is applicable to both high-voltage and low-voltage scenarios.

[0049] Temperature monitoring: Select an A-level accuracy Pt100 thermal resistance temperature sensor with a temperature measurement range of -200°C to 850°C to meet the temperature rise monitoring requirements of power equipment.

[0050] Vibration monitoring: Equipped with a three-axis accelerometer vibration sensor, the sampling frequency can reach 10 kHz to capture the time-domain and frequency-domain characteristics of the mechanical vibration of the device.

[0051] Then, each sensor collects data in real-time at a synchronous sampling frequency of 1000 Hz, encrypts the collected and preliminarily processed data using a hybrid encryption algorithm, and compresses the encrypted data using a deep learning-based compression algorithm.

[0052] Specifically, in this embodiment, the hybrid encryption algorithm first uses a symmetric encryption algorithm to perform preliminary encryption on the data, and then uses an asymmetric encryption algorithm to encrypt the symmetric encryption key.

[0053] The method for encrypting the collected and preliminarily processed data is as follows: S101. Key generation: Symmetric encryption key generation: Randomly generate a symmetric encryption key with a length of bits ; Asymmetric encryption key pair generation: Use the RSA algorithm to generate a public key and a private key.

[0054] S102. Symmetric encryption of data: The collected and preliminarily processed data is , and the data is encrypted using the AES algorithm; the symmetric encryption function is expressed as , then the encrypted data is: .

[0055] S103. Asymmetric encryption of the symmetric encryption key: Use the public key of the asymmetric encryption algorithm to encrypt the symmetric encryption key ; the asymmetric encryption function is expressed as , then the encrypted symmetric encryption key is: .

[0056] S104. Final encrypted data: The final encrypted data for transmission is composed of the encrypted data and the encrypted symmetric encryption key , and is expressed as: .

[0057] Specifically, in this embodiment, the encrypted data is compressed using a deep learning-based compression algorithm to remove redundant information from the data. The specific method is as follows: S111. Preprocess the encrypted data and convert it into a format suitable for input to the deep learning model, that is, for the encrypted byte sequence, convert it into a numerical vector ; wherein, is the preprocessing function that converts the byte sequence into a numerical vector.

[0058] S112. Build a deep learning model: Use an autoencoder as the deep learning model for data compression. The autoencoder consists of two parts: an encoder and a decoder; The encoder maps the input data to a low-dimensional latent space to achieve data compression; the encoder consists of multiple neural network layers, with a total of layers. The output of the layer is calculated as follows: ; wherein, is the weight matrix of the layer, is the bias vector of the layer, is the activation function; Finally, the output of the encoder is a vector in the latent space: .

[0059] The decoder reconstructs the vector in the latent space into an output similar to the input data ; the decoder also consists of multiple neural network layers, with a total of layers. The output of the layer is calculated as follows: ; wherein, is the weight matrix of the layer, is the bias vector of the layer, is the activation function; Finally, the output of the decoder is the reconstructed data: .

[0060] S113. Model training: Train using the operating data of several power devices. The training objective is to minimize the reconstruction error, and use MSE as the loss function: ; wherein, is the number of training data, is the th input data, is the th reconstructed data; Use the stochastic gradient descent method to update the weights and biases of the encoder and decoder to minimize the loss function .

[0061] S114. Data compression: After the model training is completed, use the trained encoder to compress the encrypted data to obtain the vector in the latent space: ; wherein, is the trained encoder.

[0062] Then, transmit the data to the data processing center through the 5G communication satellite network or the LoRaWAN wireless communication protocol to ensure low power consumption and stability of data transmission.

[0063] Specifically, in this embodiment, when the 5G signal is good in the area where the power device is located, preferentially transmit the compressed and encrypted data to the data processing center through the 5G communication network; if the 5G signal is poor or interrupted, automatically switch to satellite communication to ensure the continuity of data transmission; during the data transmission process, adopt an adaptive transmission rate adjustment strategy to dynamically adjust the data transmission rate according to the network bandwidth and signal quality to ensure stable data transmission.

[0064] Step 2. Data cleaning and preprocessing

[0065] The data processing center performs decryption and decompression operations on the received data, and at the same time cleans the data to remove noise, outliers and duplicate data in the data.

[0066] Among them, noise filtering can adopt the adaptive Wiener filtering algorithm, outlier detection and elimination can use the method based on the interquartile range (IQR), and the removal of duplicate data can be achieved through the hash value comparison algorithm. Calculate the hash value of each data record. If the hash value is repeated, keep the earliest collected record to ensure data uniqueness.

[0067] Specifically, in this embodiment, the data is decompressed and decrypted to restore the original data. The specific method is as follows: S201. Data decompression: At the receiving end, the trained decoder is used to decompress the compressed data to obtain the reconstructed data: ; where is the trained decoder; The reconstructed data is subjected to inverse preprocessing to obtain the decrypted data: ; where is the inverse preprocessing function, which is used to convert the numerical vector into a byte sequence.

[0068] S202. Decryption of the symmetric encryption key: The receiving party uses the private key of the asymmetric encryption algorithm to decrypt the encrypted symmetric encryption key ; the asymmetric decryption function is denoted as , then the decrypted symmetric encryption key is: .

[0069] S203. Decryption of the data: The receiving party uses the decrypted symmetric encryption key to decrypt the encrypted data ; the symmetric decryption function is denoted as , then the decrypted data is: .

[0070] Step 3. Feature extraction and pattern recognition First, extract the current and voltage waveform features: The continuously collected current and voltage waveform data are intercepted into fixed-time window time series segments of length T at a time interval of one second, and each segment corresponds to a two-channel input, forming a two-channel input sample , and each time series segment is normalized.

[0071] Current and voltage waveform feature extraction: The original time-domain current signal , voltage signal are subjected to band-pass filtering: ; ; where is the impulse response of the band-pass filter; The filtered current signal and voltage signal are segmented by a sliding window: The window length is N, the step size is S, and the current window signal of the kth window is: ; similarly, the voltage window signal of the kth window is obtained .

[0072] Extract the time-frequency features of the current window signal and the voltage window signal: ; ; Where: is the window function (such as the Hanning window), is the time index, is the frequency index.

[0073] Calculate the spectral amplitude matrix: ; .

[0074] Matrix normalization: ; .

[0075] Construct a multi-channel matrix: Combine the two-dimensional matrices of current and voltage into a two-channel matrix: .

[0076] Then, construct a Convolutional Neural Network (CNN) model, whose input layer dimension is ; The multi-layer convolutional layer includes the first convolutional layer with a 3×1 small convolutional kernel and a stride of 1, and the second convolutional layer with a 7×1 large convolutional kernel. The activation function of both convolutional layers is selected as ReLU; Max pooling is used to reduce the dimension of the output of the convolutional layer, the pooling window size is 2×1, and the stride is 2; The pooled feature vectors are flattened into one-dimensional vectors, and the features are further fused through the fully connected layer; The extracted features are screened through L1 regularization, and the feature maps of current and voltage are fused by means of channel splicing.

[0077] Next, use the convolutional kernel to slide on the time series segment, and extract local patterns with the help of the weight sharing mechanism; Among them, the small convolutional kernel captures high-frequency noise and transient disturbances; The large convolutional kernel captures low-frequency trends and steady-state features; Shallow convolutional layer: Learn basic features, including the waveform slope and harmonic amplitude within a single period; Deep convolutional layer: Through cross-layer connections, learn high-order features, including the distortion pattern of multi-period waveforms and the phase difference between current and voltage; Parallelly build branches with different convolutional kernel sizes to extract features of different time scales respectively, and then fuse them by splicing; Calculate the feature importance score using L1 regularization, and select the features with relatively high contribution to subsequent fault diagnosis or clustering analysis.

[0078] Specifically, for the dual-channel data of current and voltage, after the convolutional layer, the feature maps of the two are fused through channel splicing to generate a composite feature containing the coupling relationship of electrical quantities; finally, a key feature vector that can reflect the operating state of the power system is obtained, and this key feature vector reflects the waveform distortion degree, transient disturbance type, steady-state operating parameters, and the time-series characteristics of abnormal patterns.

[0079] Next, perform clustering analysis on the equipment temperature and vibration data: normalize the temperature data and vibration data to eliminate the influence of dimensions: ; Among them, is the original data, is the data after normalization; Combine the temperature data and vibration data of each device into a two-dimensional feature vector: ; Among them, is the temperature value of device , is the vibration value of device .

[0080] In this embodiment, set the number of clusters , and randomly select initial cluster centers: ; Calculate the Euclidean distance from each data point to each cluster center and assign it to the nearest cluster: ; Among them, is the cluster label to which the data point belongs, is the th iteration of the th cluster center; According to the data points within the current cluster, recalculate the cluster center: ; Among them, is the set of data points in the th cluster, is the number of data points in this cluster; When the cluster center no longer changes significantly, that is, , is the preset threshold or when the maximum number of iterations is reached, stop the iteration.

[0081] Finally, after the iteration is completed, label the cluster labels according to business experience or historical data: if the average temperature of a certain cluster is significantly higher than the normal threshold or the vibration amplitude exceeds the equipment operation standard, it is determined as an abnormal mode cluster; the other cluster is a normal mode cluster; For the newly collected feature vectors, calculate the clusters they belong to: ; if is equal to the abnormal cluster label, it is determined that the device is operating abnormally, and the abnormal feature data collected when the device is operating abnormally is recorded.

[0082] Fuse the key feature data and the abnormal feature data into unified feature data, that is, a unified feature vector: (1)For the case where there is a significant scale difference between the key feature data and the abnormal feature data, perform secondary normalization on the key feature data; (2)Perform feature splicing to obtain the integrated feature: ; where is the numerical encoding of the abnormal label. When normal, , when abnormal, .

[0083] Step 4, Fault Diagnosis and Prediction

[0084] First, perform fault diagnosis. The method is as follows: Divide the feature data obtained in Step 3 into a training set and a test set according to a ratio of 7:3, construct a non-linear SVM model using a radial basis kernel function, and set the hyperparameters C and γ.

[0085] In this embodiment, the value range of C is [0.1, 100], and the value range of γ is [0.001, 10]. Use 5-fold cross-validation, use the test set accuracy as the evaluation index, and determine the optimal hyperparameters C and γ through grid search.

[0086] Input the training set into the SVM model with the determined hyperparameters for training. After training is completed, input the test set into the trained SVM model to obtain the predicted fault type label for each test sample.

[0087] For each test sample, compare the predicted fault type label with the actual fault type label. If they are the same, the sample is diagnosed correctly; if they are different, the sample is diagnosed incorrectly. Count the number of samples diagnosed correctly in the test set and calculate the test set accuracy: Test set accuracy = (Number of samples diagnosed correctly / Total number of test set samples) × 100%.

[0088] Input the real-time collected and processed feature data into the trained SVM model. The model outputs the corresponding fault type label. If the output fault type label is one of the known fault types, it is determined that the power system currently has this type of fault; if the output fault type label is the normal operation label, it is determined that the power system is currently operating normally.

[0089] Then, perform running data prediction as follows: First, organize the historical operation data of the power system into a one-dimensional time series, and then use the sliding window method to convert the series into input-output pairs according to the set window size and prediction step length; at the same time, normalize the data.

[0090] Build an LSTM network. The dimension of the input layer is set according to the data characteristics. The hidden layer consists of multiple LSTM units, and the output layer is a fully connected layer. The number of neurons corresponds to the prediction step length; capture the long-term dependence relationship of the data through the forget gate, input gate, output gate, and cell state update mechanism.

[0091] Use the mean square error as the loss function and Adam as the optimizer to train the model with the training data, and adjust the model parameters through the backpropagation algorithm; input the real-time collected data into the trained model for prediction, and inverse-normalize the prediction result to restore it to the original data scale.

[0092] Step 5, Early warning processing

[0093] Set different levels of early warning thresholds. When the detected data exceeds the corresponding threshold or a potential fault is predicted, according to the severity and impact range of the fault, send early warning information to relevant personnel through one or more of SMS, email, and audible and visual alarms, and mark the fault location and type with different colors and icons on the monitoring interface so that the staff can take measures in time.

[0094] For example, set the following early warning levels: Level 1 early warning (severe fault): The trigger condition is that the SVM diagnoses a short circuit / overload fault, or the temperature > 100°C and the vibration > 8 m / s². Synchronously notify through audible and visual alarms (red flashing + buzzer), SMS (response within 10 seconds), and email; Level 2 early warning (abnormal state): The trigger condition is that the cluster analysis determines an abnormal mode cluster, or the LSTM prediction error > 5%. Notify through a yellow mark on the monitoring interface + email.

[0095] Furthermore, based on the method of the present invention, remote monitoring and control of the 10 kV transmission line data of a certain substation are carried out.

[0096] I. Data acquisition Deploy the following sensors at the transmission line nodes (sampling frequency 1000 Hz, continuously collect 1 second of data): Current sensor: Rogowski coil, measure the current waveform data (partial sampling points): (The effective value is about 50 A); Voltage sensor: capacitive voltage division, measure the voltage waveform data: (Effective value is approximately 10 kV); Temperature sensor: Pt100, measured temperature value: 35 °C; Vibration sensor: triaxial accelerometer, average vibration amplitude: 2.5 m / s²; Data transmission: Transmit 1000 groups of current / voltage data, temperature, and vibration values within 1 second to the data processing center through the LoRaWAN protocol.

[0097] II. Data Cleaning and Preprocessing

[0098] 1. Current / Voltage Waveform Cleaning Noise filtering: Apply adaptive Wiener filtering to the current waveform to remove high-frequency noise (an abnormal point is corrected from 52.1 A to 51.0 A).

[0099] Outlier detection: Use the interquartile range (IQR) method. Let the lower quartile of the current data be and the upper quartile be Then the interquartile range is: ; Calculate the of the current data.

[0100] The outlier judgment threshold range is: .

[0101] Remove outliers outside this range (if any).

[0102] Duplicate data removal: Through hash value comparison, there are no duplicate records.

[0103] 2. Temperature / Vibration Data Normalization

[0104] Let the original data be and the normalized data be Then the normalization formula is: ; Original temperature value: 35 °C, historical temperature range is , after normalization: ; Original vibration value: 2.5 m / s 2 , historical vibration range is , after normalization: ; Combine into a two-dimensional feature vector: .

[0105] III. Feature Extraction and Pattern Recognition

[0106] 1. Current / Voltage Waveform Feature Extraction (CNN Model)

[0107] Time window: 1 second of data is captured as a time series segment, and the segment length is , dual channel input (current + voltage), forming .

[0108] Normalization: Normalize the current / voltage value to interval.

[0109] Convolutional layer calculation: The first convolution layer uses a 3×1 convolution kernel with a step size of 1 and an output channel number of , the output dimension is: ; Used to extract high-frequency transient features (such as harmonics).

[0110] The second convolution layer uses a 7×1 convolution kernel with a step size of 1 and an output channel number of , the output dimension is: ; Used to extract low-frequency steady-state features (such as fundamental wave trends).

[0111] Pooling layer: max pooling (window 2×1, step size 2), dimension after dimensionality reduction: .

[0112] Feature fusion: After flattening into a one-dimensional vector, a key feature vector (including 50-dimensional features such as waveform distortion rate and phase difference) is generated through a fully connected layer.

[0113] 2. Temperature / vibration cluster analysis

[0114] Historical data training to obtain cluster centers: Normal cluster center: (corresponding to temperature 30°C, vibration 2m / s 2 ); Abnormal cluster center: (corresponding to temperature 50°C, vibration 4m / s 2 ); Calculate the Euclidean distance for the current data, assuming the data point is , the cluster center is , the Euclidean distance formula is: ; in is the feature vector dimension, where .

[0115] Calculate the current data point arrive The Euclidean distance of: ; Calculate the Euclidean distance to : ; Allocate to the nearest cluster: normal cluster (label 1), determine that the device status is normal.

[0116] IV. Fault Diagnosis and Prediction

[0117] 1. Fault Diagnosis (SVM Model)

[0118] Parameters of the trained SVM model: Kernel function: Radial Basis Kernel (RBF), optimal hyperparameters: C = 10, γ = 0.1; Input features: Current / voltage CNN feature vector (50 dimensions).

[0119] Input the current feature vector into the SVM model, and the output label is normal operation (no fault type match), determine that the system is running normally.

[0120] 2. Prediction of Operating Data (LSTM Model)

[0121] Historical data: Sequence of effective current values in the past 1 hour (unit: A): (A total of 60 points, 1 point per minute);

[0122] Sliding window: Set the window size w = 5 and the prediction step p = 1 to generate input-output pairs (e.g., input [48, 50, 52, 49, 51] output 50).

[0123] Normalization: Normalize the data to [0, 1].

[0124] LSTM model prediction: Input the current 5-point data , and predict that the current value in the next minute is 50.5 A (after inverse normalization).

[0125] Prediction error: Compare with the actual value (50.8 A), and the error calculation formula is: The calculated error is: , and no warning is triggered.

[0126] V. Warning Handling

[0127] Current status: Temperature 35°C < 100°C, vibration 2.5 m / s 2 <8 m / s 2 , diagnosed as normal by SVM, and the LSTM prediction error 0.59% < 5%.

[0128] No warning is triggered, and the monitoring interface shows a green indicator (normal operation).

[0129] An embodiment of the present invention further provides a remote monitoring and control system for power system data, which is used to implement the above method, such as Figure 2 shown, including: Data acquisition module: It includes current sensors, voltage sensors, temperature sensors and vibration sensors deployed at each key node of the power system, which are used to collect multi-source data during the operation of the power system in real time, and transmit the data to the data processing module through a wireless communication module; Data processing module: Receive the data transmitted by the data acquisition module, clean, extract features and analyze the data, including convolutional neural network feature extraction of current and voltage waveforms, and clustering analysis of equipment temperature and vibration data, and provide processed data for subsequent fault diagnosis and prediction; Fault diagnosis and prediction module: Adopt the SVM algorithm to diagnose the faults of the current operating state of the power system, and use LSTM to predict the future operating data of the power system; Early warning module: Used to set different levels of early warning thresholds. When data anomalies are detected or potential faults are predicted, warning information is sent to relevant personnel through various methods, and the fault location and type are visually displayed on the monitoring interface.

[0130] The above describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A method for remote monitoring and control of power system data, characterized in that It includes the following steps: S1. Real-time collect the real-time current, voltage, equipment temperature, and mechanical vibration data during the operation of the power system, encrypt the preliminarily processed data through a hybrid encryption algorithm, compress the encrypted data through a deep learning-based compression algorithm, and transmit it to the data processing center; S2. The data processing center decompresses and decrypts the received data, and cleans the data to remove noise, outliers, and duplicate data in the data; S3. Construct a convolutional neural network model to perform feature learning on the current and voltage waveform data, and extract key feature data reflecting the operating state of the power system; through a clustering algorithm, perform clustering analysis on the equipment temperature and mechanical vibration data to identify abnormal feature data of the equipment operating state; S4. Adopt the SVM algorithm to perform fault diagnosis on the current operating state of the power system according to the corresponding relationship between the feature data and the known fault types, and use LSTM to predict the operating data of the power system for a period of time in the future; S5. Set different levels of warning thresholds. When the detected data exceeds the corresponding warning threshold or a potential fault is predicted, send a warning message according to the severity and impact range of the fault, and mark the fault location and type on the monitoring interface.

2. The remote monitoring and control method for power system data according to claim 1, wherein In step S1, the preliminarily processed data is encrypted through a hybrid encryption algorithm, and the specific method is as follows: S101. Key generation: Symmetric encryption key generation: Randomly generate a symmetric encryption key with a length of bits, denoted as: ; where represents generating a -bit random number; ​ Asymmetric encryption key pair generation: Use the RSA algorithm to generate the public key and private key of the asymmetric encryption algorithm; S102. Symmetric encryption of data: The collected and preliminarily processed data is , and the AES algorithm is used to encrypt the data . The symmetric encryption function is expressed as , then the encrypted data is: ; S103. Asymmetric encryption of the symmetric encryption key: The public key of the asymmetric encryption algorithm is used to encrypt the symmetric encryption key; the asymmetric encryption function is denoted as , then the encrypted symmetric encryption key is: ; S104. Final encrypted data: The encrypted data finally used for transmission consisting of the encrypted data and the encrypted symmetric encryption key is represented as: .

3. The remote monitoring and control method for power system data according to claim 2, characterized in that, In step S1, the encrypted data is compressed through a deep learning-based compression algorithm, and the specific method is as follows: S111. Convert the encrypted data into a numerical vector suitable for input to a deep learning model : ； Among them, is a preprocessing function used to convert a byte sequence into a numerical vector; S112. Construct a deep learning model: Use an autoencoder as the deep learning model for data compression, including an encoder and a decoder; The encoder maps the input data to a vector in a low-dimensional latent space , achieving data compression; the decoder reconstructs the vector in the latent space into an output similar to the input data ; S113. Model training: Use the operating data of several power equipment for training, with the goal of minimizing the reconstruction error, and use MSE as the loss function: ； Among them, is the number of training data, is the th input data, is the th reconstructed data; Update the weights and biases of the encoder and decoder using stochastic gradient descent to minimize the loss function ; S114. Data Compression: Compress the encrypted data using the trained encoder to obtain a vector in the latent space : ； Among them, is a trained encoder.

4. The remote monitoring and control method for power system data according to claim 3, characterized in that In step S112, both the encoder and the decoder are composed of multiple neural network layers; The encoder has a total of layers. The output of the layer is calculated as follows: ； The encoder outputs a vector in the latent space: ; Among them, is the weight matrix of the layer, is the bias vector of the layer, is the activation function; The decoder has a total of layers. The calculation method for the output of the layer is as follows: ； Decoder output is the reconstructed data: ; Among them, is the weight matrix of the th layer, and is the bias vector of the 5. The remote monitoring and control method for power system data according to claim 3, characterized in that, In step S2, the data processing center decompresses and decrypts the received data, and the specific method is as follows: S201. Data decompression: At the receiving end, the trained decoder is used to decompress the compressed data to obtain the reconstructed data: ； Among them, is the trained decoder; The reconstructed data is preprocessed inversely to obtain the decrypted data : ； Among them, is an inverse preprocessing function used to convert a numerical vector into a byte sequence; S202. Decryption of the symmetric encryption key: The recipient uses the private key of the asymmetric encryption algorithm to decrypt the encrypted symmetric encryption key ; The asymmetric decryption function is expressed as , then the decrypted symmetric encryption key is: ; S203. Decryption of data: The recipient uses the decrypted symmetric encryption key to decrypt the encrypted data ; if the symmetric decryption function is denoted as , then the decrypted data is: .

6. The remote monitoring and control method for power system data according to claim 1, characterized in that In step S3, the key feature data reflecting the operating state of the power system is extracted, and the specific method is as follows: S301. Intercept the continuously collected current and voltage data into fixed-time window time series segments of length T, convert them into two-dimensional matrices, and perform normalization processing; S302. Construct a convolutional neural network model: The input layer includes: the first convolutional layer and the second convolutional layer. The activation functions of both convolutional layers are selected as ReLU; the first convolutional layer uses a small convolutional kernel with a stride of 1, and the second convolutional layer uses a large convolutional kernel; Use max pooling to reduce the dimension of the output of the convolutional layer. The pooling window size is 2×1, and the stride is 2; Flatten the pooled feature vector into a one-dimensional vector and input it into the fully connected layer; S303. Feature learning: The convolutional kernel slides on the time series segments, and local patterns are extracted through the weight sharing mechanism. Among them, small convolutional kernels are used to capture high-frequency noise and transient disturbances, and large convolutional kernels are used to capture low-frequency trends and steady-state features. The shallow convolutional layer is used to learn basic features, including the waveform slope and harmonic amplitude within a single period. The convolutional layer learns high-order features through cross-layer connections, including the distortion mode of multi-period waveforms and the phase difference between current and voltage. Branches with different convolutional kernel sizes are built in parallel to extract features at different time scales. S304. Feature screening and fusion: Calculate the feature importance scores through L1 regularization, screen out the features with high contribution to fault diagnosis or clustering analysis, and fuse the feature maps of current and voltage by means of channel splicing to generate composite features containing the coupling relationship of electrical quantities, and obtain the key feature vectors reflecting the operating state of the power system.

7. The remote monitoring and control method for power system data according to claim 6, characterized in that, In step S301, the continuously collected current and voltage waveform data are converted into a two-dimensional matrix and normalized. The specific method is as follows: S3011. Band-pass filter the original time-domain current signal and voltage signal : ； ； wherein, is the impulse response of the band-pass filter; S3012. Perform sliding window segmentation on the filtered current signal and voltage signal : The filtered continuous real-time current and voltage data are intercepted at intervals of one second to form a time-series segment of a fixed time window with a length of T. Each segment corresponds to a dual-channel input sample ; The window length is N, the step size is S, and the current window and voltage window signals of the k-th window are: ； ； S3013. Extract time-frequency features from the current window signal and voltage window signal: ； ； Among them, is the window function, is the time index, is the frequency index, is the exponential form of Fourier transform; S3014. Calculate the spectral amplitude matrix: ； ； S3015. Normalize the spectral amplitude matrix: ； ； S3016. Construct a multi-channel matrix: Combine the two-dimensional matrices of current and voltage into a two-channel matrix: 。 8. The remote monitoring and control method for power system data according to claim 7, characterized in that, In step S3, identify the abnormal feature data of the device operating state. The specific method is as follows: S311. Normalize the temperature data and vibration data to eliminate the influence of dimensions; S312. Combine the temperature data and vibration data of each device into a two-dimensional feature vector: ； Among them, is the temperature value of the device , is the vibration value of the device . S313. Set the number of clusters , randomly select initial cluster centers: ; Calculate the Euclidean distance from each data point to each cluster center and assign it to the cluster with the closest distance: ； Among them, is the data point belonging to the cluster label, is the th cluster center of the iteration; Recalculate the cluster center according to the data points within the current cluster: ； Among them, is the set of data points of the th cluster, and is the number of data points in this cluster; S314. When the change in the cluster center is less than the preset threshold, , stop the iteration when it reaches the preset threshold or the maximum number of iterations; S315. Label the cluster labels according to business experience or historical data: If the average temperature of the cluster is higher than the normal threshold or the vibration amplitude exceeds the device operation standard, it is determined as an abnormal mode cluster; the other cluster is a normal mode cluster; S316. For the newly collected feature vectors , calculate the cluster to which they belong : ； Among them, is the th clustering center. If the value is an abnormal cluster label, it is determined that the device is operating abnormally, and the abnormal feature data collected when the device is operating abnormally is recorded. S317. Fuse the key feature data and abnormal feature data into unified feature data to obtain a unified feature vector: Based on key feature data and abnormal feature data due to scale differences, the key feature data is secondarily normalized, and feature splicing is performed to obtain integrated features: ； Among them, represents the dimension of the Euclidean space, is the numerical encoding of the anomaly label.

9. The remote monitoring and control method for power system data according to claim 8, characterized in that, In step S4, conduct fault diagnosis on the current operating state of the power system. The specific method is as follows: S401. Divide the feature data obtained in step S3 into a training set and a test set according to a ratio of 7:3; S402. Construct a non-linear SVM model using the radial basis kernel function, set the value range of hyperparameters, use 5-fold cross-validation, and use the test set accuracy as the evaluation index to determine the optimal hyperparameters through grid search; S403. Input the training set into the SVM model with determined hyperparameters for training. After training, input the test set into the trained SVM model to obtain the predicted fault type labels for each test sample; S404. For each test sample, compare the predicted fault type label with the actual fault type label. If they are the same, the sample diagnosis is correct; if they are different, the sample diagnosis is incorrect; Count the number of samples with correct diagnosis in the test set and calculate the test set accuracy: S405. Input the processed feature data into the trained SVM model and output the corresponding fault type labels.

10. The remote monitoring and control method for power system data according to claim 1, characterized in that, In step S4, use LSTM to predict the operation data of the power system for a period of time in the future. The specific method is as follows: S411. Take the historical operation data of the power system as a one-dimensional time series. Through the sliding window method, according to the set window size and prediction step length, convert the series into input-output pairs and normalize the data. S412. Build an LSTM network. The dimension of the input layer is set according to the data characteristics. The hidden layer consists of multiple LSTM units, and the output layer is a fully connected layer. The number of neurons corresponds to the prediction step length. Capture the long-term dependence relationship of the data through the forget gate, input gate, output gate, and cell state update mechanism. S413. Use the mean square error as the loss function and Adam as the optimizer to train the model with the training data, and adjust the model parameters through the backpropagation algorithm. S414. Input the real-time collected data into the trained model for prediction, and inverse normalize the prediction result to restore it to the original data scale.

Images