Abstract—This article purposes the model of mid-term energy consumption load forecasting (MTLF) by using artificial neural network based on the two and the three years ahead. This load demand forecasting is a useful tool for a unit commitment and a fuel reserve planning in the power system. Both two and three years ahead forecasting uses two patterns for comparing the accuracy in this research. The results show the two years ahead of load forecasting, model no.2 can be reduced error which Mean Absolute Percentage Error (MAPE) is 4.35%. In three years ahead, the load forecasting model no.1 and no.2, MAPE are almost equal which are equal 4.65% and 4.70%, respectively. Finally, the experimental results show two and three years ahead for the load forecasting by using differential models and a varies number of a neuron in the hidden layer for finding the minimum MAPE of each model.

Index Terms—Energy consumption; Forecasting; Neural network; Yearly ahead.

Pituk Bunnoon is with Rajamangala University of Technology Srivijaya,1 Radchadumniennok Road, Boyang, Songkhla, Thailand (Email:pituk.b@rmutsv.ac.th).

Cite: Pituk Bunnoon, "Mid-Term Load Forecasting Based on Neural Network Algorithm: a Comparison of Models," International Journal of Computer and Electrical Engineering vol. 3, no. 4, pp. 600-605, 2011.