| Abstract: | In this paper an improved neural network application for short-term load forecasting purposes is presented. To speed up the learning process on one side, and not to jeopardize the stability performance of the learning process on the other side, the adaptive approach to the learning-rate parameter has been employed. Also, instead of learning overall load characteristics, the preprocessing of input data has been designed with the idea to learn only load demand behavior that is important for a certain period. The proposed neural network has shown good performance, even in the case of the incomplete data temperature set and at high irregularities in weekly load data. |
