Modeling and comparing energy consumption in basic metal industries by neural networks and ARIMA

This paper presents an artificial neural network approach for annual energy consumption in basic metal industries of Iran from 1987 to 2006. Manufacturing value added, gas price, electricity price, occupied person, and total investment are considered as variables for annual energy consumption. According to high fluctuations in this kind of industries, conventional methods do not seem to forecast energy consumption correctly and precisely. Artificial neural network based on a supervised multi-layer perceptron, multiple logarithmic regressions, and autoregressive integrated moving average models are utilized and compared each other for this sector of Iran´s industries. Neural networks model, which are presented in this study, have been trained with two different algorithms and one normalization method is used for pre- and post-process of data. Moreover, the logarithmic transformed data are used as inputs for the neural network models. By comparing results, the network model based on logarithmic data will reveal much more accuracy.