Evaluating the spatial correlations of multi-area load forecasting errors

Abstract

The short-term load forecasting error highly affects the security and economic operation of power systems. The load in different areas are distinct in the composition of consumers, impact factors, and profiles, and are thus of different forecast ability. Understanding the correlations of load forecast error among different areas would provide significant insight on the ways of managing the forecast errors. This paper carries out empirical studies on the spatial correlations of multi-area short-term load forecasting errors in Guangdong Province of China. Firstly, Artificial Neural Network (ANN) algorithm is used to conduct the day ahead forecast for 21 cities. Secondly, spatial correlations between load forecasting errors are quantified by Pearson correlation and the relationship between Pearson correlation and spatial distance is studied. Finally, copula method is used to model the joint distribution of two cities' load forecasting errors. The study shows that the forecast errors of different cities have a strong correlation. The extent of correlation depends on the distance of two areas. The joint distribution of the forecast error between cities can be effectively modelled by Gaussian Copula.1