Long-term power consumption demand prediction: A comparison of energy associated and Bayesian modeling approach

This paper contributes with two different prediction approaches for long-term power consumption demand prediction using an artificial neural networks (ANN) short-term time series predictor filter. The techniques proposed here are non-linear stochastic models using the energy associated to series and Bayesian inference, implemented by ANN. The system has the advantage of requiring as input only the historical demand time series of power consumption and allows its extension to a forecast medium and long term 3-6-12-18 months forward. The paper predicts the power consumption in the area covered by the country during the period January 1980-November 2013 in Argentina. Thus, the next 18 forecasted values are presented by the evolution of total monthly power consumption demand of the National Interconnected System of Argentina. The computational results of the prediction comparison are evaluated against the classical non-linear ANN predictor on high roughness short term chaotic time series that shows a better performance of Bayesian approach in long-short-term forecasting.