ABSTRACT

Two modules constitute the neural system. They automatically perform both the learning and the association functions. A training set was built for each day for which the interpolation was requested. It was constructed with the load values of the 4 weeks immediately before the given day and the 4 weeks after the same day of the previous year. The values were updated to the present days with the yearly trend. To save time, the learning procedure was split into two steps:

1. A first learning of a preliminary training set with 300 -;- 500 cycles. This is for the set-up of the neural network.