A method and device for predicting error of capacitor voltage transformer

Abstract

The present invention belongs to the technical field of online monitoring of electric power metering and discloses a method and device for predicting capacitor voltage transformer errors. The method comprises: inputting capacitor voltage transformer error data before the current moment and capacitor voltage transformer environmental parameter data at the current moment into a pre-trained prediction model to obtain the capacitor voltage transformer error at the current moment; wherein the prediction model is obtained by inputting historical capacitor voltage transformer error data and capacitor voltage transformer environmental parameter data into an RBF neural network training. Beneficial effects: The method can predict capacitor voltage transformer errors, promptly detect when the capacitor voltage transformer metering error exceeds a limit, and help maintenance personnel understand the deterioration trend of the metering error.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of electric power metering, and in particular to a method and device for predicting errors of a capacitor voltage transformer. Background Art

[0002] The accuracy and reliability of transformer metering performance are directly related to the fairness and justice of electricity trade settlements. In actual operation, capacitance voltage transformer errors, influenced by factors such as data acquisition principles and harsh environments, can lead to measurement deviations exceeding limits over their operating life. Therefore, accurate and rapid diagnosis of measurement errors exceeding tolerance is crucial. Furthermore, timely prediction of the deterioration trend of these errors is essential to facilitate the scheduling of inspections and maintenance by relevant operation and maintenance personnel. Failure to promptly detect transformer degradation will impact grid operations. Summary of the Invention

[0003] The purpose of the present invention is to provide a capacitor voltage transformer error prediction method and device, which can predict the error of the capacitor voltage transformer, promptly detect the measurement error of the capacitor voltage transformer exceeding the limit and help maintenance personnel understand the deterioration trend of the measurement error.

[0004] In order to achieve the above object, the present invention provides a capacitor voltage transformer error prediction method, comprising:

[0005] The capacitor voltage transformer error data before the current moment and the capacitor voltage transformer environmental parameter data at the current moment are input into a pre-trained prediction model to obtain the capacitor voltage transformer error at the current moment; wherein, the prediction model is obtained by inputting the capacitor voltage transformer error history data and the capacitor voltage transformer environmental parameter data into the RBF neural network training.

[0006] Furthermore, the method for constructing the prediction model includes:

[0007] Determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer;

[0008] A capacitor voltage transformer error model is constructed according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer;

[0009] An autoregressive RBF neural network is constructed based on the capacitor voltage transformer error model. The RBF network is clustered and optimized using the ant colony algorithm to determine the basis function center and radius of the RBF network. The weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network are obtained using the particle swarm algorithm.

[0010] The historical error data of capacitor voltage transformer and the environmental parameter data of capacitor voltage transformer are input into the RBF neural network to obtain a trained prediction model.

[0011] Furthermore, the capacitor voltage transformer error model is specifically:

[0012]

[0013] Where δ is the ratio difference, γ is the angle difference, x1 is the leakage current caused by the CVT interphase electric field, and x2 is the secondary load of the CVT.

[0014] Furthermore, the autoregressive RBF neural network includes an input layer, a hidden layer and an output layer;

[0015] The input data of the input layer is: x = [x1, x2, Y(k-1)]; where x1 is the leakage current caused by the CVT interphase electric field, x2 is the secondary load of the CVT, and Y(k-1) is the error output at the past moment;

[0016] The output of the hidden layer is expressed as:

[0017]

[0018] Where H is the hidden layer output, X is the input vector (x1, x2, Y(k-1)), b is the width of the Gaussian basis function, b>0; C j is the center vector of the implicit function neuron;

[0019] The output of the output layer is expressed as:

[0020] Y(k)=ω1H1+ω2H2+ω3H3;

[0021] Where: ω is the weight of the output layer, and Y(k) is the predicted output of the autoregressive RBF neural network.

[0022] Furthermore, the error history data of the capacitor voltage transformer input into the RBF neural network is preprocessed before being input into the RBF neural network;

[0023] The capacitor voltage transformer error history data is preprocessed to obtain second capacitor voltage transformer error history data, wherein the ratio of normal capacitor voltage transformer samples to faulty capacitor voltage transformer samples in the second capacitor voltage transformer error history data is a first ratio.

[0024] Furthermore, the preprocessing of the capacitor voltage transformer error historical data to obtain the second capacitor voltage transformer error historical data is specifically:

[0025] The historical error data of capacitor voltage transformer is reconstructed by resampling method to increase the proportion of capacitor voltage transformer fault samples;

[0026] Data enhancement is performed on the reconstructed capacitor voltage transformer error historical data to obtain second capacitor voltage transformer error historical data.

[0027] Furthermore, the resampling method is specifically as follows:

[0028] For each sample X in the capacitor voltage transformer fault sample set, calculate its distance to all samples in the capacitor voltage transformer fault sample set using the Euclidean distance as the standard to obtain its k nearest neighbors;

[0029] According to the sample imbalance ratio, a sampling ratio is set to determine the sampling rate N. For each sample X, several samples are randomly selected from its k nearest neighbors, and the selected neighbors are recorded as X n ;

[0030] For each randomly selected neighbor X n , we get the sample X by the first formula new , the first formula is specifically:

[0031] X new =X+rand(0,1)*|XX N |.

[0032] The present invention also discloses a capacitor voltage transformer error prediction device, comprising: a data input module, a prediction model and an output module;

[0033] The data input module is used to input capacitor voltage transformer error data before the current moment and capacitor voltage transformer environmental parameter data at the current moment;

[0034] The prediction model is obtained by inputting the historical data of capacitor voltage transformer error and the environmental parameter data of capacitor voltage transformer into RBF neural network training;

[0035] The output module is used to output the prediction error.

[0036] Furthermore, the method for constructing the prediction model includes:

[0037] Determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer;

[0038] A capacitor voltage transformer error model is constructed according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer;

[0039] An autoregressive RBF neural network is constructed based on the capacitor voltage transformer error model. The RBF network is clustered and optimized using the ant colony algorithm to determine the basis function center and radius of the RBF network. The weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network are obtained using the particle swarm algorithm.

[0040] The historical error data of capacitor voltage transformer and the environmental parameter data of capacitor voltage transformer are input into the RBF neural network to obtain a trained prediction model.

[0041] Furthermore, the capacitor voltage transformer error model is specifically:

[0042]

[0043] Where δ is the ratio difference, γ is the angle difference, x1 is the leakage current caused by the CVT interphase electric field, and x2 is the secondary load of the CVT.

[0044] Compared with the prior art, the capacitor voltage transformer error prediction method and device according to the embodiment of the present invention have the following beneficial effects:

[0045] (1) The error of the capacitor voltage transformer can be predicted, the measurement error of the capacitor voltage transformer can be discovered in time, and the maintenance personnel can understand the deterioration trend of the measurement error.

[0046] (2) The present invention uses an autoregressive RBF neural network to predict CVT errors, and uses resampling and data enhancement to preprocess the historical CVT error data, thereby improving the balance of the input data and thus improving the accuracy of the model prediction;

[0047] (3) The ant colony algorithm is used to determine the center and radius of the RBF basis function, which simplifies the network structure, improves the convergence speed, and reduces the algorithm running time;

[0048] (4) The particle swarm algorithm is used to dynamically update the weights of the RBF output layer, which improves the convergence speed of the algorithm and reduces the algorithm running time. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of a method for predicting errors of a capacitor voltage transformer according to the present invention;

[0050] Figure 2 The diagram is a topological structure diagram of an RBF neural network in a method for predicting errors of a capacitor voltage transformer according to the present invention. DETAILED DESCRIPTION

[0051] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0052] Example 1:

[0053] A capacitive voltage transformer (CVT) divides the voltage of a series capacitor, then steps it down and isolates it through an electromagnetic transformer. As a voltage conversion instrument, a CVT can also couple carrier frequencies to transmission lines for long-distance communications, selective high-frequency line protection, and remote control. Compared to conventional electromagnetic voltage transformers, CVTs offer advantages such as high impulse insulation strength, simple manufacturing, compact size, and light weight, as well as numerous economic and safety advantages.

[0054] In order to predict the error of the capacitor transformer, it is necessary not only to avoid the inaccuracy of the information source of the secondary information system, but also to reduce the loss of electricity metering and ensure the normal operation of the measurement, control and protection devices. Therefore, how to predict the trend of CVT error changes to provide early warning of CVT risks is a technical challenge.

[0055] like Figure 1 As shown, a method for predicting an error of a capacitor voltage transformer includes:

[0056] The capacitor voltage transformer error data before the current moment and the capacitor voltage transformer environmental parameter data at the current moment are input into a pre-trained prediction model to obtain the capacitor voltage transformer error at the current moment; wherein, the prediction model is obtained by inputting the capacitor voltage transformer error history data and the capacitor voltage transformer environmental parameter data into the RBF neural network training.

[0057] Through the pre-trained prediction model and the capacitor voltage transformer error data before the current moment and the capacitor voltage transformer environmental parameter data at the current moment, the capacitor voltage transformer error at the current moment can be accurately and reliably predicted, the measurement error of the capacitor voltage transformer exceeding the limit can be discovered in time, and maintenance personnel can be helped to understand the degradation trend of the measurement error.

[0058] In this embodiment, the method for constructing the prediction model includes:

[0059] Step S1, determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer;

[0060] Step S2, constructing a capacitor voltage transformer error model according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer;

[0061] Step S3, constructing an autoregressive RBF neural network based on the capacitor voltage transformer error model, clustering and optimizing the RBF network using an ant colony algorithm, determining the basis function center and radius of the RBF network, and obtaining the weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network using a particle swarm algorithm;

[0062] Step S4: inputting the capacitor voltage transformer error history data and the capacitor voltage transformer environmental parameter data into the RBF neural network to obtain a trained prediction model.

[0063] In the present invention, different parameters in the environmental parameters of the capacitor voltage transformer include: primary voltage frequency, ambient temperature and humidity, accumulated contamination, phase-to-phase leakage current of the capacitor voltage transformer, and secondary load of the capacitor voltage transformer. However, different parameters have different effects on the error of the capacitor voltage sensor.

[0064] Table 1 Key parameters affecting CVT metering performance

[0065]

[0066] Referring to Table 1, among the known factors affecting CVT metering performance, the voltage frequency of the power system is generally very stable, with fluctuations far less than ±0.5 Hz. Therefore, the primary voltage frequency is not a major parameter affecting CVT metering performance. Ambient temperature and humidity typically directly affect the capacitance and dielectric loss factor of the CVT's voltage divider capacitor. Existing research shows that within the common temperature range of -25°C to 45°C, the impact of environmental factors on CVT metering performance is far below the error variation limit of a Class 0.2 CVT. As primary equipment operating outdoors for long periods of time, CVTs inevitably accumulate contamination on their outer surfaces. However, existing research shows that surface contamination only significantly affects CVT metering performance in high-humidity, heavily polluted areas. If the installation height of the CVT during online operation is different from that during offline calibration, there will be some difference in the CVT's ground leakage current, affecting the CVT's ratio difference characteristics. However, the impact of installation height on metering performance is relatively low and is not a major parameter affecting CVT metering performance. CVTs installed at substations are typically three-phase, side-by-side. Interphase leakage current is unavoidable between the three phases. During online operation, when all three phases are simultaneously energized, this interphase leakage current significantly affects the ratio and phase difference characteristics of the CVT. The CVT can be considered an equivalent voltage source with relatively low internal resistance, so changes in load size significantly affect the CVT's metering performance. For the simulated CVT, a secondary load change exceeding 30VA results in a ratio change exceeding 0.1% and a phase difference change exceeding 2'.

[0067] In summary, the key parameters affecting the CVT metering performance are the leakage current caused by the CVT interphase electric field and the secondary load of the CVT. These two factors should be used to establish an error model.

[0068] In this embodiment, the capacitor voltage transformer error model is specifically:

[0069]

[0070] Where δ is the ratio difference, γ is the angle difference, x1 is the leakage current caused by the CVT interphase electric field, and x2 is the secondary load of the CVT.

[0071] In this embodiment, assuming that there are n sets of data samples of CVT error and environmental parameters, the CVT error sequence can be expressed as:

[0072]

[0073] Where: δ i is the difference data of group i; γ i is the ith group of angular difference data.

[0074] In this embodiment, assuming that there are n sets of data samples of CVT error and environmental parameters, the CVT error sequence can be expressed as:

[0075]

[0076] in:

[0077]

[0078] B1 and B2 are matrices to be determined.

[0079] During field operation, the error of CVT is the result of the interaction of various environmental parameters. There is no clear functional relationship between the error and the environmental parameters, and it is difficult to obtain the analytical values ​​of B1 and B2. Therefore, it is necessary to establish a neural network for prediction.

[0080] Reference Figure 2 In this embodiment, the autoregressive RBF neural network includes an input layer, a hidden layer and an output layer;

[0081] The input data of the input layer is: x = [x1, x2, Y(k-1)]; where x1 is the leakage current caused by the CVT interphase electric field, x2 is the secondary load of the CVT, and Y(k-1) is the error output at the past moment;

[0082] The output of the hidden layer is expressed as:

[0083]

[0084] Where H is the hidden layer output, X is the input vector (x1, x2, Y(k-1)), b is the width of the Gaussian basis function, b>0; C j is the center vector of the implicit function neuron;

[0085] The output of the output layer is expressed as:

[0086] Y(k)=ω1H1+ω2H2+ω3H3;

[0087] Where: ω is the weight of the output layer, and Y(k) is the predicted output of the autoregressive RBF neural network.

[0088] In this embodiment, the RBF network is clustered and optimized by the ant colony algorithm to determine the basis function center and radius of the RBF network, specifically:

[0089] When the clustering probability between two capacitor voltage transformer error samples is greater than a preset reference probability, the two capacitor voltage transformer error samples are recorded as one class, the cluster centers of the error samples belonging to the same class are calculated, and it is judged whether the overall error of the capacitor voltage transformer meets the requirements. If the requirements are met, the algorithm is stopped;

[0090] If the requirements are not met, update the pheromone of different paths, reclassify the error samples, and determine whether the overall error of the capacitive voltage transformer meets the requirements.

[0091] In this embodiment, the ant colony algorithm is used to optimize the clustering of the RBF network, and the center and radius of the basis function of the RBF network are determined, which can simplify the network structure and improve the convergence speed.

[0092] In this embodiment, the specific optimization steps are as follows:

[0093] Initialize the parameters of the CVT error samples. Let the number of CVT error samples be n, m be the sample attribute, the sample attributes include normal samples and fault samples, r be the clustering radius, and P0 be the reference probability.

[0094] Calculate the distance between two samples;

[0095] d ij =||(x i -x j )||, i = 1, 2,..., n;

[0096] Initialize the pheromone on each path;

[0097]

[0098] Calculate the clustering probability between two error samples of the transformer;

[0099]

[0100] When P ij ≥ P0, x i and x<​​​​​​​​​​​​​​​​​​​​​​​

[0107]

[0108] Where ρ is the volatility coefficient and Q is the pheromone mass, and these two values ​​are constants.

[0109] Repeat the steps from calculating the clustering probability between two error samples of CVT to updating the pheromone until convergence and the basis function center is obtained.

[0110] In this embodiment, the interphase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer have different effects on the training of the RBF neural network model. The optimal weight parameters should be obtained by adjusting the weights of the two factors so that the performance of the trained neural network model is optimal. The particle swarm algorithm will be used to start from the random solution and iteratively find the optimal solution to obtain the weights of the two factors in the CVT error state evaluation. The specific method is as follows: the weights corresponding to various factors are ω n (n=1, 2), that is, the weight corresponding to the phase-to-phase leakage current of the capacitor voltage transformer is w1, and the weight corresponding to the secondary load of the capacitor voltage transformer is w2, w1+w2=1. Let w1 be the value w to be optimized in the particle swarm algorithm. The particle updates its own speed and position through the two extreme value positions. The update formula is:

[0111]

[0112] in, v i k is the speed of particle i in space; w is the network output weight; k is the number of iterations; c1c2 is the learning factor used to balance the relative importance between pbest and gbest, and r1r2 takes random values ​​between 0 and 1.

[0113] The linear decreasing weight method is used to dynamically update the w value. The weight update formula is:

[0114]

[0115] Among them, k represents the current number of iterations, k max is the maximum number, w max is the maximum weight, usually 0.9, w min The minimum weight is usually 0.1. The weight w1 corresponding to the phase-to-phase leakage current of the capacitor voltage transformer is the w obtained by the weight update formula. The final weight w corresponding to the phase-to-phase leakage current of the capacitor voltage transformer is obtained by the above method. 1， The weight w corresponding to the secondary load of the capacitor voltage transformer 2， As the output weight of the CVT error state prediction model, it participates in the training of the RBF neural network model to obtain a model with better performance.

[0116] In actual working conditions, there are abundant normal CVT samples and few faulty samples. Class imbalance will lead to problems such as poor model diagnostic ability and generalization ability. The resampling method is used to change the distribution of CVT error history data, and then balance the ratio of normal CVT samples to faulty CVT samples, so that the CVT error history data set reaches a balanced state and the model diagnostic ability is improved.

[0117] In this embodiment, the error history data of the capacitor voltage transformer input into the RBF neural network is preprocessed before being input into the RBF neural network;

[0118] The capacitor voltage transformer error history data is preprocessed to obtain second capacitor voltage transformer error history data, wherein the ratio of normal capacitor voltage transformer samples to faulty capacitor voltage transformer samples in the second capacitor voltage transformer error history data is a first ratio, preferably 1:1.

[0119] In this embodiment, the preprocessing of the capacitor voltage transformer error historical data to obtain the second capacitor voltage transformer error historical data is specifically:

[0120] The historical error data of capacitor voltage transformer is reconstructed by resampling method to increase the proportion of capacitor voltage transformer fault samples;

[0121] Data enhancement is performed on the reconstructed capacitor voltage transformer error historical data to obtain second capacitor voltage transformer error historical data.

[0122] During the preprocessing process, the CVT error history dataset is subjected to two data enhancement processes. The first time, when the CVT error history dataset is reconstructed, data enhancement is performed on the faulty CVT samples to balance the unbalanced distribution ratio of various types of samples in the CVT error dataset; the second time, a secondary enhancement is performed on the reconstructed CVT error dataset to expand the sample size and improve the feature extraction capability from the data level.

[0123] Furthermore, data enhancement can be achieved by flipping, cropping, scaling, translating, radial transformation, adding noise, etc.

[0124] Furthermore, SMOTE (Synthetic Minority Oversampling Technique) can be used for resampling. It is an improvement on the random oversampling algorithm. Since random oversampling adopts a strategy of simply duplicating samples to increase minority class samples, it is easy to cause the model to overfit, that is, the information learned by the model is too specific and not general enough. The basic idea of ​​the SMOTE algorithm is to analyze the minority class samples and artificially synthesize new samples based on them and add them to the dataset.

[0125] In this embodiment, the resampling method is specifically:

[0126] For each sample X in the capacitor voltage transformer fault sample set, calculate its distance to all samples in the capacitor voltage transformer fault sample set using the Euclidean distance as the standard to obtain its k nearest neighbors;

[0127] According to the sample imbalance ratio, a sampling ratio is set to determine the sampling rate N. For each sample X, several samples are randomly selected from its k nearest neighbors, and the selected neighbors are recorded as X n ;

[0128] For each randomly selected neighbor X n , we get the sample X by the first formula new , the first formula is specifically:

[0129] X new =X+rand(0,1)*|XX N |.

[0130] Example 2:

[0131] The present invention also discloses a capacitor voltage transformer error prediction device, comprising: a data input module, a prediction model and an output module;

[0132] The data input module is used to input capacitor voltage transformer error data before the current moment and capacitor voltage transformer environmental parameter data at the current moment;

[0133] The prediction model is obtained by inputting the historical data of capacitor voltage transformer error and the environmental parameter data of capacitor voltage transformer into RBF neural network training;

[0134] The output module is used to output the prediction error.

[0135] In this embodiment, the method for constructing the prediction model includes:

[0136] Determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer;

[0137] A capacitor voltage transformer error model is constructed according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer;

[0138] An autoregressive RBF neural network is constructed based on the capacitor voltage transformer error model. The RBF network is clustered and optimized using the ant colony algorithm to determine the basis function center and radius of the RBF network. The weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network are obtained using the particle swarm algorithm.

[0139] The historical error data of capacitor voltage transformer and the environmental parameter data of capacitor voltage transformer are input into the RBF neural network to obtain a trained prediction model.

[0140] The error prediction device of Example 2 can be implemented according to the error prediction method of Example 1, so the error prediction device will not be repeatedly described and limited in Example 2.

[0141] In summary, compared with the prior art, the method and device for predicting errors of a capacitor voltage transformer according to the embodiment of the present invention have the following beneficial effects:

[0142] (1) The error of the capacitor voltage transformer can be predicted, the measurement error of the capacitor voltage transformer can be discovered in time, and the maintenance personnel can understand the deterioration trend of the measurement error.

[0143] (2) The present invention uses an autoregressive RBF neural network to predict CVT errors, and uses resampling and data enhancement to preprocess the historical CVT error data, thereby improving the balance of the input data and thus improving the accuracy of the model prediction;

[0144] (3) The ant colony algorithm is used to determine the center and radius of the RBF basis function, which simplifies the network structure, improves the convergence speed, and reduces the algorithm running time;

[0145] (4) The particle swarm algorithm is used to dynamically update the weights of the RBF output layer, which improves the convergence speed of the algorithm and reduces the algorithm running time.

[0146] 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 substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A capacitor voltage transformer error prediction method, characterized in that: include: Inputting capacitor voltage transformer error data before the current moment and capacitor voltage transformer environmental parameter data at the current moment into a pre-trained prediction model to obtain the capacitor voltage transformer error at the current moment; wherein the prediction model is obtained by inputting the historical capacitor voltage transformer error data and the capacitor voltage transformer environmental parameter data into an RBF neural network training; The method for constructing the prediction model includes: Determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer; A capacitor voltage transformer error model is constructed according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer; An autoregressive RBF neural network is constructed based on the capacitor voltage transformer error model. The RBF network is clustered and optimized using the ant colony algorithm to determine the basis function center and radius of the RBF network. The weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network are obtained using the particle swarm algorithm. The historical error data of capacitor voltage transformer and the environmental parameter data of capacitor voltage transformer are input into the RBF neural network to obtain a trained prediction model.

2. A capacitor voltage transformer error prediction method according to claim 1, characterized in that: The capacitor voltage transformer error model is specifically: ； Where, is the ratio difference, is the angular difference; is the leakage current caused by the CVT interphase electric field, It is the secondary load of CVT.

3. A capacitor voltage transformer error prediction method according to claim 1, characterized in that: The autoregressive RBF neural network includes an input layer, a hidden layer and an output layer; The input data of the input layer is: ;in, is the leakage current caused by the CVT interphase electric field, is the secondary load of CVT, Y(k-1) is the error of output at the past moment; The output of the hidden layer is expressed as: ； Among them, H is the hidden layer output, X is the input vector , b is the width of the Gaussian basis function, b>0; is the center vector of the implicit function neuron; The output of the output layer is expressed as: ； in: is the weight of the output layer, and Y(k) is the predicted output of the autoregressive RBF neural network.

4. A capacitor voltage transformer error prediction method according to claim 1, characterized in that: The error history data of the capacitor voltage transformer input into the RBF neural network is preprocessed before being input into the RBF neural network; The capacitor voltage transformer error history data is preprocessed to obtain second capacitor voltage transformer error history data, wherein the ratio of normal capacitor voltage transformer samples to faulty capacitor voltage transformer samples in the second capacitor voltage transformer error history data is a first ratio.

5. A capacitor voltage transformer error prediction method according to claim 4, characterized in that: The preprocessing of the capacitor voltage transformer error historical data to obtain the second capacitor voltage transformer error historical data is specifically: The historical error data of capacitor voltage transformer is reconstructed by resampling method to increase the proportion of capacitor voltage transformer fault samples; Data enhancement is performed on the reconstructed capacitor voltage transformer error historical data to obtain second capacitor voltage transformer error historical data.

6. A capacitor voltage transformer error prediction method according to claim 5, characterized in that: The resampling method is specifically: For each sample in the capacitor voltage transformer fault sample , using Euclidean distance as the standard to calculate the distance from it to all samples in the capacitor voltage transformer fault sample set, and obtain its k nearest neighbors; According to the sample imbalance ratio, a sampling ratio is set to determine the sampling magnification N. For each sample , randomly select several samples from its k nearest neighbors and record the selected neighbors as ; For each randomly selected neighbor , the sample is obtained by the first formula , the first formula is specifically: 。 7. A capacitor voltage transformer error prediction device, characterized in that: include: Data input module, prediction model and output module; The data input module is used to input capacitor voltage transformer error data before the current moment and capacitor voltage transformer environmental parameter data at the current moment; The prediction model is obtained by inputting the historical data of capacitor voltage transformer error and the environmental parameter data of capacitor voltage transformer into RBF neural network training; The output module is used to output the prediction error; The construction of the prediction model includes: Determining influencing parameters of a capacitor voltage transformer error model according to the influence of different parameters in the capacitor voltage transformer environmental parameters on the capacitor voltage transformer error, wherein the influencing parameters are: a phase-to-phase leakage current of the capacitor voltage transformer and a secondary load of the capacitor voltage transformer; A capacitor voltage transformer error model is constructed according to the phase-to-phase leakage current of the capacitor voltage transformer and the secondary load of the capacitor voltage transformer; An autoregressive RBF neural network is constructed based on the capacitor voltage transformer error model. The RBF network is clustered and optimized using the ant colony algorithm to determine the basis function center and radius of the RBF network. The weights of the capacitor voltage transformer phase-to-phase leakage current and the capacitor voltage transformer secondary load in the RBF neural network are obtained using the particle swarm algorithm. The historical error data of capacitor voltage transformer and the environmental parameter data of capacitor voltage transformer are input into the RBF neural network to obtain a trained prediction model.

8. The capacitor voltage transformer error prediction device according to claim 7, characterized in that: The capacitor voltage transformer error model is specifically: ； Where, is the ratio difference, is the angular difference; is the leakage current caused by the CVT interphase electric field, It is the secondary load of CVT.

Images