Development of future energy scenarios with intelligent algorithms: Case of hydro in Turkey Didem Cinar a, * , Gulgun Kayakutlu a , Tugrul Daim b a Istanbul Technical University, Department of Industrial Engineering, Istanbul, Turkey b Portland State University, Portland, OR, USA article info Article history: Received 25 May 2009 Received in revised form 2 October 2009 Accepted 21 December 2009 Available online 20 January 2010 Keywords: Renewable energy Neural networks Back-propagation Genetic algorithms abstract Energy production is considered as one of the key indicators for economic development. It is vital to improve the renewable energy production for global sustainability, while leveraging the national resources. This study is contributing to the demonstration of using genetic algorithms (GA) in the development of future energy scenarios as well as to the strategic energy studies in Turkey. The fore- casting model developed in this study uses forward feeding back-propagation (BP) method improved by GA. The proposed model is applied in the Turkish case. The test errors are shown to emphasize the positive difference between the proposed model and the classical BP model. The results highlight that there is strong evidence indicating that the government should reconsider their current energy strategies. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction Energy sustainability, stability and variety are considered to be vital for the economic development. Since energy is an inevitable input for all industries, the sustainable supply of energy resources becomes an essential part of the national economical strategies. Availability of energy resources at a reasonable cost and energy utilization without causing negative social effects are essential. It is well known that the supply of fossil energy is finite whereas renewable energy sources including hydropower are generally considered to be available and sustainable over the relatively long term timelines [1]. Hence, it is important to balance the use of fossil and green energy production, which can be accomplished by determining the available domestic resources and making changes in the energy policies accordingly [2]. Turkey is analyzed as a case study because it is a representative of many countries that are highly dependent on fossil energy, and will be more, unless more attention is paid to the renewable energy resources [3]. Share of fossil energy importation in 1970 was 24.59% and grew three times to become 73.19% in 2006 [4]. It is evident that renewable energy must be used more widely to be less dependent on foreign resources [5]. Hydro energy is taken in this study as a representative of renewable energy resources in Turkey after analyzing local energy forecasts. The choice is based on two major reasons: i) Hydroelectric energy is one of the most abundant resources of Turkey; and ii) deregulation to decentralize energy investments is only completed for hydroelectric energy – studies for the other renewable energies are on-going. Forecasting the production of hydropower until 2012 will provide input to strategies that can improve both national energy policies and individual energy investments. Many of the forecasting studies in the past used various forms of econometric methods. These techniques need some assumptions and cannot solve complex nonlinear patterns. Because of the drawbacks of statistical techniques, recent studies have started using artificial intelligence methods, like artificial neural networks (ANN), genetic algorithms (GA) and ant colonies considering the contingency [6–8]. ANN is widely accepted as an alternative way to handle complex problems. Control, robotics, pattern recognition, forecasting, medicine, power systems, manufacturing, optimiza- tion, signal processing and social/psychological sciences are various application areas that used ANN as in renewable energy problems in diverse ways [9]. In this study, a hybrid forecasting model is developed to integrate ANN techniques and GA. The forward feeding back-propagation (BP) method is improved by calculating the design parameters through the use of specific crossover and mutation operations. Improvements with GA are observed in literature to define the model or initial weights as proposed by Wang and Huang [10]. This study contributes by completing the * Corresponding author. Istanbul Technical University, Department of Industrial Engineering 34367 Maçka, Istanbul, Turkey. Tel.: þ90 212 293 13 00; fax: þ90 212 240 72 60. E-mail address: cinard@itu.edu.tr (D. Cinar). Contents lists available at ScienceDirect Energy journal homepage: www.elsevier.com/locate/energy 0360-5442/$ – see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.energy.2009.12.025 Energy 35 (2010) 1724–1729