Short-Term Load Forecasting Using System-Type Neural Network Architecture

Neural networks have been applied in various new ways to the problem of short-term load forecasting for power systems. Virtually all of these methods are based on using statistical patterns, which are perceived between the yearly load histories of the system to predict the forecasted year´s demand. The proposed method also uses a neural network approach, but differs from the others in how those patterns are perceived. Specifically, the proposed approach begins with the premise that the load demand for a given year can be given a structure which can then be related to the structure of the reference year, in such a way that a transformation can be found from the reference year´s structure to the forecasting year´s structure. The transformation depends upon how parameters, which influenced the load but can not be measured, move from the reference year to the forecasting year.