Annual energy consumption prediction using particle filters

This paper presents a framework for predicting the monthly-annual electric energy consumption (EC) using practical filters, sequential Monte Carlo methods. The particle filtering technique is utilized to describe and track the EC "signal" structure with respect to time. The state of the evolution of energy consumption is described as a nonlinear process of past states. Disturbance or noise associated with the energy consumption state of evolution is dealt with as a non-Gaussian process. The EC in Kuwait from 1992 to 2000 is used as training set to predict the monthly EC of the year 2001. The results show that the average percentage error between the actual and estimated EC for the year 2001 is 4.46%.