Very short-term load forecasting using artificial neural networks

In a deregulated, competitive power market, utilities tend to maintain their generation reserve close to the minimum required by an independent system operator. This creates a need for an accurate instantaneous-load forecast for the next several dozen minutes. This paper presents a novel approach to very short-time load forecasting by the application of artificial neural networks to model load dynamics. The proposed algorithm is more robust as compared to the traditional approach when actual loads are forecasted and used as input variables. It provides more reliable forecasts, especially when the weather conditions are different from those represented in the training data. The proposed method has been successfully implemented and used for online load forecasting in a power utility in the United States. To assure robust performance and training times acceptable for online use, the forecasting system was implemented as a set of parsimoniously designed neural networks. Each network was assigned a task of forecasting load for a particular time lead and for a certain period of day with a unique pattern in load dynamics. Some details of this are presented in the paper.