Peak load forecasting using analyzable structured neural network

This paper presents daily peak load forecasting using an analyzable structured neural network. Recently, a number of neural network approaches for peak load forecasting have been proposed. The purpose of these studies is to construct a nonlinear model for accurate forecasting using multilayer neural networks. However, conventional multilayer neural networks are said to be a black box. Namely, it is difficult to explain reasons of forecasting results. The proposed neural network has hidden units with connecting weights between only one group of related input units. This feature allows to analyze independent relations between input and output units. The effectiveness of the proposed method is shown by a comparison with actual and extracted correlation from the trained neural network