Modeling of Wind Speed Profile using Soft Computing Techniques

Modeling of Wind Speed Profile using Soft Computing Techniques

Pijush Samui (VIT University, India) and Yıldırım Dalkiliç (Erzincan University, Turkey)
DOI: 10.4018/978-1-4666-6631-3.ch010
OnDemand PDF Download:


This chapter examines the capability of three soft computing techniques (Genetic Programming [GP], Support Vector Machine [SVM], and Multivariate Adaptive Regression Spline [MARS]) for prediction of wind speed in Nigeria. Latitude, longitude, altitude, and the month of the year have been used as inputs of GP, RVM, and MARS models. The output of GP, SVM, and MARS is wind speed. GP, SVM, and MARS have been used as regression techniques. To develop GP, MARS, and SVM, the datasets have been divided into the following two groups: 1) Training Dataset – this is required to develop GP, MPMR, and RVM models. This study uses 18 stations' data as a training dataset. 2) Testing Dataset – this is required to verify the developed GP, MPMR, and RVM models. The remaining 10 stations data have been used as testing dataset. Radial basis function has been used as kernel functions for SVM. A detailed comparative study between the developed GP, SVM, and MARS models is performed in this chapter.
Chapter Preview


Renewable energy sources are increasingly utilised due to global political uncertainty and alarmingly increasing pollution levels in air, water and soil. Wind energy has become the focal point for energy developers due to the availability of megawatt size wind machines, accessible management facilities, ease and low cost of maintenance, government subsidies, tax benefits, etc. The power of wind is clean, inexhaustible, and a free source of energy. This source has served humankind for many centuries by propelling ships and driving wind turbines to grind grains and pump water. The cheap availability and plentiful supply of petroleum, the high cost and uncertainty of wind placed it at an economic disadvantage. However, after the 1973 oil embargo, it is realised that the world’s oil supplies would not last forever and other energy sources have to be developed.

For proper and efficient utilisation of wind power, the prediction of wind speed is very important. It is needed for site selection, performance prediction, planning of windmills and the selection of an optimal size of the wind machine for a particular site. A few no. of studies have been conducted at various locations in which the prediction of wind speed profile was done using different analytical tools such as: stochastic simulation (Lehner et al., 2012), Rayleigh and Weibull distribution functions (Vanderbei, 1995) Mesoscale Model (Ancona, 1999), Kalman filter(Ferris et al., 2000) and the use of Box-jenkins methodology(Lee et al., 2007). But, however for the efficient exploration of the wind energy potentials of a nation, a nationwide assessment is required and all the above developed models are location specific and hence are limited in terms of accuracy due to non-linear variability of wind characteristics in space and time. Also to improve the accuracy implication, more sampling points and smaller time frame (monthly) variability is required. Also for locations having few numbers of ground stations, a more sparse modelling and prediction tool is required for the development of the more precise wind speed profile.

This book chapter adopts Support Vector Machine (SVM), Genetic Programming (GP) and Multivariate Adaptive Regression Spline (MARS) for prediction wind speed (V) (m/sec) in Nigeria. This article employs the database collected from the work of Fadare (2010). SVM is developed by Vapnik (1995). It is constructed based on Statistical Learning Algorithm. Researchers successfully used SVM for solving different problems in engineering (Liu & Gan, 2008; Yazdi et al., 2009; Sonavane & Chakrabarti, 2010; Suetani et al., 2011; Kim & Ahn, 2012). GP is developed based on the concept of genetic algorithm (Koza, 1992). It has been successfully applied to a large number of difficult problems (Hernandez et al., 2004; Guven et al., 2009; Guven & Kisi, 2011). MARS has been developed by Friedman (1991). It does not assume any functional relationship between input and output variables. It has been successfully used to solve different problems (MacLean & Mix, 1991; Veaux et al., 1993; Ekman & Kubin, 1999; Prasad & Iverson, 2000; Jin et al., 2000; Ko & Osei-Bryson, 2004; Johannesonn & Sweeney, 2006; Sharda et al., 2008; Okine et al., 2003, 2009). A comparative study has been carried out between the developed SVM, GP and MARS models.

Complete Chapter List

Search this Book: