A New Forward-Backward Linear Algorithm for Electrical Load Demand Prediction
One hour-ahead load forecasts are critically important for the reliable and cost-effective operation of modern power systems, especially, in the deregulated economy, where effective on-spot price fixing is always a major concern. In this paper, a time series modeling for these very short term forecasts is proposed by using a new auto-regressive algorithm, specifically designed for appropriate handling of large data records. The model divides the bulky data record into short segments and searches for the AR coefficients that simultaneously model the data with the least mean squared error. This approach can accurately forecast the one hour-ahead and one day-ahead loads of the weekdays. The proposed method can provide more accurate results than the conventional techniques, such as, standard AR-based algorithms, Burg and the seasonal Box-Jenkins ARIMA (SARIMA) and many of the intelligent hybrid approaches, where artificial neural networks are combined with evolutionary search methods. Obtained results from extensive testing of the proposed model confirm the validity of the developed approach. Three years load demand data of New South Wales (NSW) Australian power grid from 2011 to 2013 are used in the experimentation.