Prediction of power flow results in time-series-based planning with artificial neural networks and data pre-processing

Abstract

Time-series-based analysis of power systems requires long simulation times if the annual simulation of N-1 cases are to be analysed. Artificial neural networks can be trained to predict bus voltage magnitudes and line loadings to shorten these simulation times. In this study, the authors show how to reduce prediction errors by applying different data pre-processing methods including sampling methods, feature selection strategies, and scaling techniques. Results are shown for four realistic benchmark grids. The authors show that the maximum prediction error can be reduced by >30% when using pre-processing methods.