Capacitor Voltage Transformer Error Prediction Method Based on Fusion Convolution Module

A Capacitor Voltage Transformer (CVT) error prediction method based on fusion convolution module is proposed in this paper. Firstly, the CVT error sequence and its influencing factor sequence are decomposed by Seasonal-Trend Decomposition Using LOESS (STL) to construct a threedimensional input matrix. Then, the matrix is input into continuous feature extraction module. Each feature extraction module contains a fusion convolution module and a residual network. Inside the fusion convolution module, the data is subjected to three-dimensional convolution and deep separable convolution respectively, and the data is extracted and fused from the whole and the local respectively. Finally, the extracted features are reduced in dimension and input into the Long Short-Term Memory neural network (LSTM) for prediction output. The experimental results show that the prediction accuracy of the proposed method is better than the classical prediction model in deep learning.