Forecasting Natural Gas Consumption Using Pso Optimized Least Squares Support Vector Machines

This paper proposes an effective model based on the least squares support vector machines (LSSVM) and the particle swarm optimization (PSO), termed PSO-LSSVM, for prediction of natural gas consumption, as an important energy resource. The salient feature of mapping nonlinear data into high dimension feature space, distinguishes LS-SVM as a powerful approach for forecasting and estimation. Optimization of the model’s parameters by a fast and efficient PSO algorithm results in an optimized model which is employed for prediction of annual natural gas consumption in Iran and Unites States. Promising results were obtained for prediction of Iranian gas consumption from 1998 to 2006 and U.S. gas consumption from 2001 to 2005. Besides, comparison to an optimized multi-layer preceptron (MLP) network, using error indices of MAPE and NMSE demonstrated the superior performance of the proposed PSO-LSSVM approach.