Evaluation of time series methods for distribution system state forecasting.

Abstract

High proliferation of DERs in the distribution system necessitates forecasting ability for state estimator. A dynamic state estimation is a fundamental tool for realizing distribution system automation. Selection of appropriate methodological framework is an essential step before implementing dynamic state estimation. We applied Vector Error Correction Model (VECM), Random Forest and Long Short Term Memory Network (LSTM) time series models to dynamic power system state estimation problem and found out that the LSTM model outperforms the other two in prediction accuracy.