Selection of input variables for electricity price forecast

Abstract

The selection of input variables is vital for performance of electricity price forecast model when artificial neural network (ANN) is used to forecast electricity price. However, the selection of input variables mainly is based on experience and heuristics in current studies. This paper aims to compare three categories methods, (1) the relevance between input variables and electricity price, (2) the relationship between different input variables and (3) the structure of ANN, on selecting input variables in terms of electricity price. In order to avoid influence of structure of ANN on performance of selecting input variables, the back-propagation (BP) network and Elman network that represent different structure of ANN respectively are used to forecast electricity price. The real electricity market data from Nordic and California electricity market is used to examine the performance of electricity price forecast model, the result shows that the electricity price forecast model formed using input variables selected by category, the relevance between input variables and electricity price, has the best performance of electricity price forecast.