Forecasting day-ahead price spikes for the Ontario electricity market

Abstract

A novel methodology based on neural networks is presented to forecast day-ahead electricity spikes and prices. Day-ahead electricity prices are forecasted by the first neural network trained using a data set consisting of similar price days. Next, spike prices are identified from the forecasted prices using a spike classifier, and these spikes are re-forecasted by using neural networks trained over historical spike hours. Finally, a data re-constructor is used to achieve the overall day-ahead electricity spike and price forecasting. Numerical experiments are conducted using data from the wholesale electricity market of Ontario, Canada, and significant improvements are achieved in terms of forecasting accuracy.