Forecasting Energy Demand in Iran: An Application of Evolutionary Algorithms-based Neural Networks Sadeghi Hosseina, Sohrabivafa Hosseinb, Hajian Mohammadhadic aAssistant Professor of Economics, Tarbiat Modares University, Tehran, Iran bMaster Student of Economics, Tarbiat Modares University, Tehran, Iran cPhD Candidate of Economics, Tarbiat Modares University, Tehran, Iran Online published on 6 August, 2014. Abstract Smart nonlinear techniques are of interest of researchers forecasting energy demand because it has a changeable nonlinear trend. The evolutionary algorithms and neural networks are of most applicable techniques in this respect, although they have some weaknesses; among which, their prerequisite to specific functional form and requiring excessive samples and pause in optimal local point in learning neural networks (BP* and LM†). The present research aims to provide a combined algorithm in order to utilize strengths of these techniques as well as reducing such weaknesses. For this, we investigate the efficiency of different techniques of forecasting energy demand during 1967–2001. Research findings indicate that employing the evolutionary algorithms for instruction of neural networks, with a limited data, obtains desirable results, and application of neural network based on PSO-GA‡ lead to more appropriate results. Comparison of results to other studies in this area implies that applied variables are able to explain the variations so that proposed algorithm has higher capability than of the other algorithms. Top Keywords Neural Networks, Genetic Algorithm, Particle Swarm Algorithm, Forecasting, Energy Demand. Top |