An Application of Multistep Long-Short Term Memory Algorithm to Predict the Performance of Capacitor Voltage Transformer

This paper introduces how capacitor voltage transformers are monitored and concludes four key parameters of a capacitor voltage transformer, which are, ratio difference f
U
, phase error δ
U
, dielectric loss tanδ, and capacitance C. It then proposes a method with the Hampel filter to clean out the outliers of obtained parameters. Cleaned data are further stationarized and scaled to fit the requirements of a deep learning algorithm based on multistep Long-Short Term Memory neutral network, which analyzes the processed data to evaluate and predict the short-term operating performance of a capacitor voltage transformer.