Forecasting day-ahead electricity prices using data mining and neural network techniques

Abstract

A feed forward neural network model combined with a data mining technique at the pre-processing stage is presented to forecast day-ahead hourly electricity prices for the wholesale electricity market in the province of Ontario, Canada. For each forecasting day, a set of 135 days is selected for the training of the neural network. Moreover, five similar prices days are identified for each hour from a set of 90 days corresponding to each training day. The average price of these five days at the particular hour is used as one of the inputs to the neural network to improve the forecasting accuracy. Forecasting experiments are carried out for nine days in 2012. Test results show that the proposed technique reduces the mean absolute percentage error significantly.