Modelling the Belgian gas consumption using neural networks

摘要

In this paper an accurate neural network model is proposed for the gas consumption in Belgium. It is a static non-linear model, based on monthly data and contains the following inputs: temperature, difference between real and expected temperature, oil price, number of domestic clients and consumption by industry. Various interpretations are made on the identified models such as yearly error, normalized gas consumption, growth rate, uncertain linear model interpretation and sensitivity of the consumption with respect to the temperature. In contrast to traditional models, which depend only on the temperature, the present neural network models show excellent generalization ability, with small yearly errors on training and test set.