Application of the RAN algorithm to the problem of short term load forecasting

This paper shows the application of the resource allocation network (RAN) algorithm to the problem of electrical load forecasting in a Spanish utility company. The choice of the parameters of the algorithm is usually done manually. In this paper the possibility of automatic selection of parameters is investigated. These parameters are of paramount importance since they determine the final size of the network and its capacity to generalize to new situations. The number of training samples in this kind of problems is usually small. This fact has a strong influence in methods for obtaining neural models, but is rarely taken into account in the forecasting literature. The influence of the available training data is analyzed empirically.