Capacitor voltage transformer error online evaluation method

Abstract

The invention relates to the technical field of capacitor voltage transformer online monitoring, in particular to a capacitor voltage transformer error online evaluation method, which comprises the following steps of: 1, collecting a large amount of historical operation data with error tags of a CVT (capacitor voltage transformer); 2, selecting proper main features influencing errors; 3, preprocessing the historical data and the operation error, and constructing a data set; step 4, establishing an operation error evaluation model combining the variational auto-encoder and the convolutional neural network; 5, training the operation error evaluation model by using the training set and the test set; step 6, when the trained model is put into use, on-line monitoring data of the same type of capacitive voltage transformers needs to be subjected to same preprocessing and then is input into the trained operation error evaluation model to obtain an operation error; the operation error of the capacitor voltage transformer can be analyzed and evaluated on line; secondly, an evaluation model combining a variational auto-encoder and a convolutional neural network is built to improve the analysis accuracy;

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of capacitor voltage transformers, and in particular to an online error evaluation method for capacitor voltage transformers. Background Art

[0002] As a crucial component of electric energy metering, the accuracy and reliability of instrument transformers (CVTs) are directly linked to the fairness and justice of electric energy trade settlements. Capacitor voltage transformers (CVTs) utilize series capacitors to divide the voltage, which is then stepped down and isolated by electromagnetic transformers. As a voltage converter, CVTs can also couple carrier frequencies to transmission lines for long-distance communications, selective high-frequency line protection, and remote control.

[0003] In the actual operation of CVT, the mutual inductor error is affected by the working principle and harsh environment, which will cause the measurement deviation to exceed the limit during its working life.

[0004] For example, the invention patent CN115754875B discloses a voltage transformer error assessment method and device that uses periodic offline testing. However, due to the long calibration cycle (typically four years), errors can occur, impacting energy metering and the safe and stable operation of the power system. Offline testing requires a power outage, which is extremely difficult to achieve in high-voltage power grids. To accurately and timely detect errors in operating capacitor voltage transformers, an online capacitor voltage transformer error assessment method is urgently needed to address this issue. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a method for online error evaluation of capacitor voltage transformer by acquiring historical output data of capacitor voltage transformer, building an evaluation model combining variational autoencoder and convolutional neural network, preprocessing the feature data and error data of the collected data, training the network model to obtain an operation error evaluation model of capacitor voltage transformer, inputting the short-term operation data of the target capacitor voltage transformer into the evaluation model after the same preprocessing, obtaining the phase difference and ratio difference of the target transformer, and realizing online analysis and evaluation of the operation error of the capacitor voltage transformer; secondly, building an evaluation model combining variational autoencoder and convolutional neural network to improve the analysis accuracy.

[0006] A method for online error evaluation of a capacitor voltage transformer according to the present invention comprises the following steps:

[0007] Step 1: Collect a large amount of historical CVT operation data with error labels, including actual operation data and laboratory simulation data;

[0008] Step 2: Select appropriate main features that affect the error: voltage value and phase value;

[0009] Step 3: Preprocess historical data and operational errors and construct a data set;

[0010] Step 4: Establish a running error evaluation model combining variational autoencoder and convolutional neural network;

[0011] Step 5: Using the training set and the test set to train the operation error evaluation model to obtain a trained operation error evaluation model;

[0012] Step 6: When the trained model is put into use, the online monitoring data of the same type of capacitor voltage transformer needs to be preprocessed in the same way and then input into the trained operation error evaluation model to obtain the operation error.

[0013] Preferably, the preprocessing in step 3 includes missing value processing, duplicate data removal and normalization processing for the data including field measurements and laboratory simulations. In addition, all data are arranged in chronological order.

[0014] Preferably, the missing value processing: when the data has less than or equal to three consecutive missing values, the formula y is used i+1 =2y i +y i―1 deal with;

[0015] Where y i+1 is a missing value; i is the value immediately before the missing value; y i―1 is the value two moments before the missing value;

[0016] The repeated value processing: if there are multiple record data at a certain recording time, the data of the last record is retained;

[0017] Normalization: To further improve the training speed of the model, the data needs to be normalized before building the dataset. In particular, the running error is used as the label data and is not normalized.

[0018] The time-based sorting refers to the order in which all recorded data are sorted according to the time of recording.

[0019] Preferably, in step 4, an operation error evaluation model combining a variational autoencoder and a convolutional neural network is established. Specifically, the variational autoencoder is used to generate latent variables of high-quality historical data, and the convolutional neural network uses the latent variables to predict the operation error.

[0020] Preferably, when the variational autoencoder is trained, the input variables are fed into the encoder network, the dimensions are reduced, and the mean and standard deviation of each latent variable are output, and then the mean and standard deviation are used to calculate the value of z=e σ +μ constructs a latent vector, which is then fed into the decoder to reconstruct the input, where σ is the standard deviation and μ is the mean.

[0021] Preferably, in step five, the running error evaluation model is trained using the training set and the test set. Specifically, when training the running evaluation model, it is necessary to consider the reconstruction loss and KL divergence (Kullback-Leibler Divergence) of the variational autoencoder, as well as the error between the convolutional neural network output and the true running error, and give the three appropriate weights to construct the loss function in model training.

[0022] Preferably, the loss function of the variational autoencoder is designed to be two: KL divergence and reconstruction loss, and the reconstruction loss here uses mean square error.

[0023] Preferably, the mean square error:

[0024] Loss VAF =Loss KL +βLoss reconstruction

[0025]

[0026] Loss reconstruction =MSE(x,x′)

[0027] Where β is the coefficient.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. By acquiring historical output data from capacitor voltage transformers, an evaluation model combining a variational autoencoder and a convolutional neural network is constructed. The feature data and error data of the collected data are preprocessed, and the network model is trained to obtain an operational error evaluation model for the capacitor voltage transformer. The short-term operational data of the target capacitor voltage transformer is then preprocessed and input into the evaluation model to obtain the phase difference and ratio difference of the target transformer, thereby enabling online analysis and evaluation of the operational error of the capacitor voltage transformer.

[0030] 2. Secondly, build an evaluation model that combines variational autoencoders and convolutional neural networks to improve analysis accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the process structure of the present invention;

[0032] Figure 2 This is a structural diagram of the training process of the present invention combining a variational autoencoder and a convolutional neural network;

[0033] Figure 3 Schematic diagram of the structure of the variational autoencoder of the present invention;

[0034] Figure 4 Schematic diagram of the network structure of the present invention combining variational autoencoder and convolutional neural network;

[0035] Figure 5 It is a schematic diagram of the operating steps of the present invention. DETAILED DESCRIPTION

[0036] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0037] Example

[0038] like Figures 1 to 5 As shown, a method for online evaluation of capacitor voltage transformer error includes the following steps:

[0039] Step 1: Collect and save historical data and operating errors;

[0040] The voltage and phase data collected by the monitoring device are collected in a group of three transformers of phases A, B and C. The U a 、U b 、U c , Φ a , Φ b 、v c Sampling is performed, where the effective value of the voltage is counted in a period of one second, and the phase angle is output by taking the phase angle value at the first sampling moment in the cycle;

[0041] Step 2: Preprocess historical data and operational errors and construct a data set;

[0042] (1) Data preprocessing:

[0043] a1. Missing value processing:

[0044] When the data has three or fewer consecutive missing values, use the formula y i+1 =2y i +y i―1 deal with;

[0045] Where y i+1 is a missing value; iis the value immediately before the missing value; i―1 is the value two moments before the missing value;

[0046] a2. Duplicate value processing:

[0047] If there are multiple records at a certain recording time, the data of the last record will be retained;

[0048] a3. Normalization: To further improve the training speed of the model, the data needs to be normalized before building the dataset. In particular, the running error is used as the label data and is not normalized.

[0049] a4. Sort by time:

[0050] All recorded data should be sorted in the order of recording time;

[0051] (2) Constructing a dataset:

[0052] Considering the physical relationship between the three phases, when constructing the data set, the three phases A, B, and C CVT are grouped together, and their corresponding U a 、U b 、U c , Φ a , Φ b , Φ c is a record, 100 records are a sample, and the corresponding error is the label; so the sample shape in the dataset is [6, 100], which means that the sample has 6 variables and a length of 100, and the label shape is [6, 1], which means that the label is a set of CVT amplitude error and phase angle error.

[0053] Step 3: Establish a running error evaluation model combining variational autoencoder and convolutional neural network;

[0054] (1) Variational Autoencoder:

[0055] The variational autoencoder is an unsupervised neural network whose model is as follows Figure 3 As shown in Figure 1, it consists of an encoder network and a decoder network. Variational autoencoders are developed from autoencoders and solve the problem of non-regularized latent space in autoencoders.

[0056] When training a variational autoencoder, the input variables are fed into the encoder network, which reduces its dimension and outputs the mean and standard deviation of each latent variable. The mean and standard deviation are then used to calculate the value of z = e σ +μ constructs a latent vector, which is then fed into the decoder to reconstruct the input, where σ is the standard deviation and μ is the mean;

[0057] (2) Establish an operation error assessment model:

[0058] The CVT operation evaluation error model is as follows Figure 4 As shown. The model input channel (input_channels) is 6, and the output channel (output_channels) is 6;

[0059] To obtain feature information at different scales, a variational autoencoder introduces two convolutional networks (long-scale and short-scale) and concatenates their outputs. Convolutional kernels with kernel sizes of 15 and 2 are used for the long-scale and short-scale convolutional networks, respectively. Multi-scale convolution helps the model learn jointly at different scales. Multiple Conv1D layers are stacked in the encoder network to reduce the temporal dimension of the input, and a stride of 2 is used in each Conv1D to reduce the feature scale. The opposite operation is performed in the decoder network.

[0060] Step 4: Using the training set and the test set to train the operation error evaluation model to obtain a trained operation error evaluation model;

[0061] (1) Loss function design

[0062] The loss function of the variational autoencoder is designed to be two: KL divergence and reconstruction loss. The reconstruction loss here uses the mean square error, that is:

[0063] Loss VAE =Loss KL +βLoss reconstruction

[0064]

[0065] Loss reconstruction =MSE(x,x′)

[0066] Where β is the coefficient;

[0067] The loss function of the regression model is designed to be the mean square error:

[0068]

[0069] Where y i is the true value of the error; Output values for the error assessment model.

[0070] The loss function of the entire model is:

[0071] Loss=Loss reg +αLoss VAE

[0072] Where α is the coefficient

[0073] (2) Training model:

[0074] The dataset is divided into training set and test set in a ratio of 7:3. The batch size (batch_size) of the dataset is 8, and the nonlinear activation function is ReLU.

[0075] Step 5: When the trained model is put into use, the short-term operating data of the same type of capacitor voltage transformer needs to be preprocessed in the same way and then input into the trained operation error evaluation model to obtain the operation error.

[0076] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for online error evaluation of a capacitor voltage transformer, characterized in that: The following steps are involved: Step 1: Collect a large amount of historical CVT operation data with error labels, including actual operation data and laboratory simulation data; Step 2: Select appropriate main features that affect the error: voltage value and phase value; Step 3: Preprocess historical data and operational errors and construct a data set; Step 4: Establish a running error evaluation model combining variational autoencoder and convolutional neural network; Step 5: Using the training set and the test set to train the operation error evaluation model to obtain a trained operation error evaluation model; Step 6: When the trained model is put into use, the online monitoring data of the same type of capacitor voltage transformer needs to be preprocessed in the same way and then input into the trained operation error evaluation model to obtain the operation error.

2. The online error evaluation method for a capacitor voltage transformer according to claim 1, wherein: The preprocessing in step three includes missing value processing, duplicate data removal and normalization processing for the data including field measurements and laboratory simulations. In addition, all data must be arranged in chronological order.

3. A capacitor voltage transformer error online evaluation method according to claim 2, characterized in that: The missing value processing: When the data has less than or equal to three consecutive missing values, use the formula y i+1 =2y i +y i―1 deal with; Where y i+1 is a missing value; i is the value immediately before the missing value; i―1 is the value two moments before the missing value; The repeated value processing: if there are multiple record data at a certain recording time, the data of the last record is retained; Normalization: To further improve the training speed of the model, the data needs to be normalized before building the dataset. In particular, the running error is used as the label data and is not normalized. The time-based sorting refers to the order in which all recorded data are sorted according to the time of recording.

4. The online error evaluation method for a capacitor voltage transformer according to claim 1, wherein: In step 4, a running error evaluation model combining a variational autoencoder and a convolutional neural network is established. Specifically, the variational autoencoder is used to generate latent variables of high-quality historical data, and the convolutional neural network uses the latent variables to predict running errors.

5. The online error evaluation method for capacitor voltage transformer according to claim 4, characterized in that: When the variational autoencoder is trained, the input variables are fed into the encoder network, which reduces its dimension and outputs the mean and standard deviation of each latent variable. The mean and standard deviation are then used to calculate the mean and standard deviation of each latent variable through the formula: z = e σ +μ constructs a latent vector, which is then fed into the decoder to reconstruct the input, where σ is the standard deviation and μ is the mean.

6. The online error evaluation method for capacitor voltage transformer according to claim 1, characterized in that: In step five, the running error evaluation model is trained using the training set and the test set. Specifically, when training the running evaluation model, it is necessary to consider the reconstruction loss and KL divergence of the variational autoencoder, as well as the error between the convolutional neural network output and the true running error, and assign appropriate weights to the three to construct the loss function in model training.

7. The online error evaluation method for capacitor voltage transformer according to claim 6, characterized in that: The loss function of the variational autoencoder is designed to be two: KL divergence and reconstruction loss. The reconstruction loss here uses the mean square error.

8. The online error evaluation method for capacitor voltage transformer according to claim 7, characterized in that: The mean square error: Loss VAE =Loss KL +βLoss reconstruction Loss reconstruction =MSE(x,x′) Where β is the coefficient.

Images