Analysis of Features Used in Short-Term Electricity Price Forecasting for Deregulated Markets

Abstract

With the liberalization of the Turkish electricity market, accurately forecasting short-term electricity prices became an important issue for the market players. Although majority of price estimation studies use historical prices, it is known that factors like demand, load, fuel prices and weather conditions affect price forecasting. In this study, we examine the impact of calendar data, historical prices and loads, weather conditions and currencies on short-term electricity price forecasting for Turkish market. We test the combinations of feature subsets on the feed forward neural network forecast model. Moreover, we observe the effect of training set size on forecast. Our results indicate that the best feature subset combination is calendar data, historical prices and load prediction.