Day-ahead electricity price forecasting using optimized multiple-regression of relevance vector machines

In deregulated, auction-based, electricity markets price forecasting is an essential participant tool for developing bidding strategies. In this paper, a day-ahead intelligent forecasting method for electricity prices is presented. The proposed approach is comprised of two steps. In the first step, a set of two relevance vector machines (RVM) is employed where each one provides next day predictions for the price evolution. In the second step, a multiple regression model comprised of the two relevance vector machines is built and the regression coefficients are computed using genetic based optimization. The performance of the proposed approach is tested on a set of electricity price hourly data from four different seasons and compared to those obtained by each of the relevance vector machines. The results clearly demonstrate, in terms of mean square error, the superiority of the proposed method over each individual RVM. (5 pages)