SHORT-TERM ELECTRICAL DEMAND FORECASTING IMPROVEMENT BY APPLYING GAUSSIAN FILTER TO THE TIME SERIES

Short-term electrical load forecasting especially for the next ۲۴ hours (next day) is one of the most interesting topics in the power distribution systems and smart grids. Principal component analysis (PCA), support vector regression (SVR) and artificial neural network (ANN) are among the widely used load forecasting methods. To improve the efficiency of these methods, it is proposed that a Gaussian filter is applied to the electrical load time series before forecasting implementation. This filtering removes the outlier points and redundant information contained in the electricity consumption curve and so, the forecasting error is significantly improved. The experimental results on an electricity consumption dataset collected by the residential customers in Ireland show that the mean absolute percentage error is reduced in average from approximately ۷.۵% to less than ۰.۳%.