Fault Diagnosis Method for High Voltage Power Metering System Based on the PCA-ELM Algorithm

The current fault diagnosis of high-voltage power metering systems lacks the processing of information dimensions, which leads to the low correctness of fault diagnosis results. To this end, a fault diagnosis method based on the PCA-ELM algorithm for high-voltage power metering systems is proposed. Eliminate data information noise. Extract fault information features. Based on PCA to reduce the information dimension, the ELM algorithm is used to analyze the fault information characteristics. The fault information characteristics correspond to the fault types to complete the fault diagnosis of the high-voltage power metering system. Comparative experiments are designed for verification. The measurement results show that the accuracy rate of fault diagnosis for this method is 99.89%, and the average accuracy rate of fault type diagnosis is 98.87%, which substantially improves the correct rate of diagnosis results.


Introduction
A high-voltage power metering system is a very important part of basic power facilities.The main working equipment of high-voltage power metering systems are voltage transformers, voltage transformers, and energy meters.In the process of operation of these devices, if a failure occurs, it will have a more serious impact on the whole system.In high-voltage power metering systems, load currents are transmitted, and as demand changes, so does the size of the load, which leads to faults such as short circuits in high-voltage power metering systems, which need to be diagnosed promptly and accurately [1].PCA can process a large amount of data information and reduce the dimensionality of the information feature space.PCA reduces the loss of the information contained in the original indicators and enables a comprehensive analysis of the collected data, reducing the workload of data analysis and increasing the accuracy of data analysis.ELM is a feedforward neural network, which reduces the training error through the setting of weights and the formulation of hidden layers.ELM has been widely used in many fields [2].Combining PCA with ELM to form an extreme learning machine neural network based on principal component analysis can reduce the workload of data analysis by reducing unnecessary analysis indexes in the case of a large amount of data information, while ELM is used to analyze the extracted data features and complete the analysis of data information features in the process of multiple training iterations.In the high-voltage power metering system, the algorithm of PCA-ELM can accurately analyze the large amount of data information generated in the operation of the system, extract the fault characteristics in a large amount of data information, and complete a higher quality fault diagnosis of the high-voltage power metering system [3].

Eliminate Data Information Noise
High-voltage power metering systems generate a lot of data information during operation.This data information is not directly usable because of the presence of a lot of interfering information, i.e., noise.Noise is a useless signal generated by an independent random process that has nothing to do with the content of the information.The noise will be eliminated and the real data information can be obtained.We design an active filter circuit in a high-voltage power system.The active filter circuit can suppress data information outside the specified frequency range by setting the frequency at which the signal is allowed to pass [4].We set the cutoff frequency f and determine the proportionality between the damping R and the cutoff frequency as shown in the following table : Table 1 Proportional relationship between cutoff frequency and damping F(Hz) R (μF) 1-10 20-1 10-100 1-0.1 100-1000 0.1-0.01 The active filter circuit is designed according to the proportional relationship required in the noise elimination process.The active filter circuit designed in this paper is shown in the following figure: In Equation (1), b0, b1, and b2 are the frequencies set to allow passage, λ is the damping factor in this circuit, ω0 is the intrinsic frequency, and P is the frequency of the data information in the highvoltage power metering system.In this function expression, when the values of b0 and b1 are both 0, the data information allowed to pass is low-frequency information.

Extract Fault Information Features
The implied layer is the data information processing module.Let the real data information be Xi, i=1,2...n, and the desired output fault signal be Yi∈{-1,1}, where -1 is characteristic of normal operation data information and 1 is characteristic of fault data information.Then the classification hyperplane equation is: In Equation ( 2), (•) denotes the vector dot product, and η is the parameter of the constraint condition.The constraint condition is the parameter that separates the normal data information from the faulty data information.To make the extraction of fault information features more adequate, the classification interval is designed to satisfy the following constraints.
After the classification of the real data information is completed in the implicit layer, the extracted fault data information is passed to the undertaking layer.The bearer layer is responsible for filtering and classifying the features of the fault data information [5][6][7].As shown in the above figure, because of the different magnitudes of the fault information features, the influence of the magnitudes needs to be eliminated before the dimensionality reduction process.Normalize the principal component features of fault information.Then the individual principal components of the fault data information are arranged in the order from largest to smallest according to the correlation between the fault information features.The variance share S of each principal component is calculated as:

Analyzing Features Based on PCA-ELM Algorithm
In Equation ( 4), a1, a2, ..., ai are the eigenvectors corresponding to the fault information eigenvalues, and ZX1, ZX2, ..., and ZXi are the raw variable values after normalization.Among all the results, the one with the largest variance is the first principal component.ELM will continue to analyze and process the fault information features after the dimensionality reduction process, and the process of analyzing and processing the fault information features after the PCA process is shown in the following figure.The number of rows is the number of input fault information features.Columns are the number of nodes.The optimal number of nodes is obtained, and the number of nodes in the implied layer should meet the contribution rate of more than 99.5% to the ELM output matrix.In the ELM algorithm, there is no error node in the output layer, so the output weight can be solved to minimize the error function to complete the analysis of fault information characteristics.The error function F of the ELM algorithm is calculated as: In Equation ( 5), G is the output matrix of the hidden layer, T is the fault information feature after PCA processing, and ‖ is the Frobenius parametric number of the matrix elements.Through iterative calculation, the matrix weights of the ELM algorithm are continuously reduced to avoid redundancy and improve the stability and generalization ability of the ELM algorithm.Finally, according to the calculated matrix weights, the fault information features after PCA processing are analyzed and the analysis results are output to complete the analysis of fault information features of high-voltage power metering system based on the PCA-ELM algorithm.

Diagnose Faults in the High-Voltage Power Metering System
The fault type of each fault message is diagnosed based on the results of the fault message characteristics analyzed by the PCA-ELM algorithm.The fault diagnosis is divided into its types according to the fault location and the cause of the fault, as shown in the following table: According to the correspondence in the above table, the fault information characteristics processed by the PCA-ELM algorithm correspond with the fault location and fault phenomenon to obtain the detailed fault information of the high-voltage power metering system and complete the diagnosis of system faults.

Experimental preparation
To verify the feasibility of the fault diagnosis method of a high-voltage power metering system based on the PCA-ELM algorithm proposed in this paper, a fault diagnosis comparison experiment is designed.All the data information generated by the operation of a high-voltage power metering system in a city is collected, which contains 105, 326 items.Among them, it contains 1630 IA fault information, 3604 Ia fault information, 586 Ib fault information, 423 Ua fault information, 899 Ub fault information, 2547 ua fault information, 1368 ub fault information, a total of 11,057 pieces of all fault information, and 94, 269 pieces of normal operation data information.The above data information is aggregated into an experimental dataset, and 70% of the data information of each category in this dataset is extracted for algorithm training.The remaining 30% of the data information in this dataset is used for the data of the fault diagnosis experiment.Finally, there are 28280 normal data, 489 IA fault messages, 1081 Ia fault messages, 176 Ib fault messages, 127 Ua fault messages, 270 Ub fault messages, 764 ua fault messages, and 410 ub fault messages in the diagnosis experiment, totaling 31597.
To verify the feasibility of the proposed method, comparison object 1 is the fault diagnosis method of a high-voltage power metering system based on the Elman network.Comparison object 2 is a fault diagnosis method of a high-voltage power metering system based on momentum BP neural network.In this fault diagnosis test experiment, the fault sign parameter of the Elman network is set to 8, the input vector dimension to 8, the output vector dimension is set to 4, and the number of hidden layers is 17.The algorithm is trained by the gradient descent learning method.The learning rate of the momentum BP neural network learning algorithm is set to 0.1.logsig function is used as the transfer function of BP neurons.The momentum factor is set to 0.8.The training times of all three fault diagnosis methods for high-voltage power metering systems are set to 1000 times.

Analysis of experimental results
The 31597 data information in the experimental dataset is applied to three fault diagnosis methods respectively to test the correct rate of fault diagnosis of different diagnosis methods under the same training times.The fault diagnosis method of a high-voltage power metering system based on the PCA-ELM algorithm proposed in this paper is named Method 1.The Elman network-based fault diagnosis method for high-voltage power metering systems is named Method 2. The momentum BP neural network-based fault diagnosis method for high-voltage power metering systems is named Method 3. The experimental results are shown in the following table: According to the above table, the correct rate of determining fault data information for three types of high-voltage power metering systems can be derived, and the number of information determined as fault information is compared with the number of real fault information to derive the correct rate results shown in the following table: We can see from Table 4 that method 1 proposed in this paper has a high correct rate for fault determination of data information of high-voltage power metering systems, which is 3.95% and 4.58% higher than the other two fault diagnosis methods, respectively.
According to the experimental results in Table 3, the correct diagnosis rate of the three diagnostic methods for each fault type was calculated and the results are shown in the following table: We can see from Table 5 that method 1 proposed in this paper has a high correct diagnosis rate for various fault types, which is above 97%, and the average correct diagnosis rate is 98.87%.In the other two diagnosis methods, the highest correct rates are 77.91% and 90.24%, respectively.The highest correct rates are 22.09% and 9.76% lower than those of the fault diagnosis methods proposed in this paper, with a large gap.In addition, the average correct rates of the other two diagnostic methods were 63.98% and 54.67%, respectively.The average correct rates are 34.89% and 44.20% lower than those of the diagnosis methods proposed in this paper, which has a large gap.Thus, the fault diagnosis method of the high-voltage power metering system based on the PCA-ELM algorithm proposed in this paper can significantly improve the correct diagnosis rate of high-voltage power metering system faults, which has high practical significance.

Conclusion
In this paper, a fault diagnosis method based on the PCA-ELM algorithm for high-voltage power metering systems is proposed for the problem that the correct rate of diagnosis results is not high in fault diagnosis of high-voltage power metering systems.The fault information of high-voltage power metering systems is diagnosed by PCA descending fault information features and ELM algorithm analyzing fault information features.The diagnosis method effectively avoids the influence of highdimensional information on the diagnosis results and greatly improves the accuracy of fault diagnosis, which has a greater application value in the operation and maintenance of high-voltage power metering systems.

Figure 1
Figure 1 Active Filter Circuit Diagram We create an active filter circuit as shown above.Establish the transfer function of the data signal according to this circuit diagram.The transfer function expression is K(P), and its calculated expression is: 2 0 1 2 2 2 0 b 2 P b P b K P P P λ ω 0 Write the PCA-ELM-based algorithm to analyze the features of fault information.PCA is used to analyze the principal components of the fault data information.The operation flow is shown in the following figure.

Figure 2 PCA
Figure 2 PCA Analysis of Fault Information Principal Component Process

Figure 3
Figure 3 ELM Algorithm Process

Table 4
Accuracy of Fault Diagnosis by Different Fault Diagnosis Methods

Table 5
Diagnostic accuracy of different fault types using different methods