ENERGY CONSUMPTION PREDICTION USING SEQUENTIAL MONTE CARLO METHODS

O.A. Alsayegh and O.A. Almatar

Keywords

Energy consumption evolution, longterm prediction, nonGaussian,particle filters, stochastic behaviour analysis

Abstract

In this paper, a new time-dependent model is proposed for the prediction of annual monthly energy consumption (EC). The aim is to furnish information that will lead to better analyses of future load demand rather than to enhance the forecasting accuracy rate. Conventional forecasting methods predict a future EC value and neglect information about its stochastic behaviour. Forecasting the behaviour of the EC is the main goal of this work. Recursive Bayesian filters by Monte Carlo simulations are utilized for predicting and analyzing the annual monthly EC change. The EC change posterior density function is represented by a set of random samples with associated weights; estimates are then computed based on these samples and weights. As the number of samples increases, this Monte Carlo characterization becomes an equivalent representation to the usual functional description of the posterior probability density function, which reflects the behaviour of the stochastic nature of EC change. The proposed technique was used to predict and analyze the annual-monthly EC in Kuwait. The EC data in Kuwait for the years 1992–2000 were used as a training set to predict the EC for the year 2001. The prediction performance was tested against an artificial neural network (ANN) and ARIMAX models. Furthermore, statistical EC analysis was performed using posterior density functions resulting from the proposed technique.

Important Links:



Go Back