66 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 1, NO. 2, JULY 2010 Optimization of Wind Turbine Performance With Data-Driven Models Andrew Kusiak, Member, IEEE, Zijun Zhang, and Mingyang Li, Student Member, IEEE Abstract—This paper presents a multiobjective optimization model of wind turbine performance. Three different objectives, wind power output, vibration of drive train, and vibration of tower, are used to evaluate the wind turbine performance. Neural network models are developed to capture dynamic equations modeling wind turbine performance. Due to the complexity and nonlinearity of these models, an evolutionary strategy algorithm is used to solve the multiobjective optimization problem. Data sets at two different frequencies, 10 s and 1 min, are used in this study. Computational results with the two data sets are reported. Analysis of these results points to a reduction of wind turbine vibrations potentially larger than the gains reported in the paper. This is due to the fact that vibrations may occur at frequencies higher than ones reflected in the 10-s data collected according to the standard practice used in the wind industry. Index Terms—Blade pitch angle, data analysis, data mining, drive train acceleration, evolutionary strategy (ES) algorithm, multiobjective optimization, neural networks (NNs), power opti- mization, torque, tower acceleration, wind turbine vibrations. I. INTRODUCTION I NTEREST in renewable energy has increased in recent years due to environmental concerns and growing aware- ness of the limited supply of fossil fuels. The anticipated increase in the cost of electricity generated from fossil fuels due to carbon taxation has become a catalyst in the quest for clean energy. Wind energy has been most successfully commercialized among all forms of renewable energy. 1 Research in wind energy has significantly intensified in recent years. Areas with the most research progress include the design of wind turbines [1], [2], the design and reliability of wind farms [3]–[5], the control of wind turbines [6], [7], [22], [23], wind energy conversion [8], [9], the prediction of wind power [10], [11], and condi- tion monitoring of wind turbines [12], [13]. Optimization has been considered as one critical issue tightly involved in these wind energy research areas. Boukhezzar et al. [27] designed a nonlinear controller for optimizing the power of the DFIG generator [27]. Abdelli et al. [28] applied a multiobjective Manuscript received November 07, 2009; revised February 06, 2010; ac- cepted March 21, 2010. Date of publication April 12, 2010; date of current ver- sion June 23, 2010. This work was supported by funding from the Iowa Energy Center under Grant 07-01. The authors are with the Intelligent Systems Laboratory, The University of Iowa, Iowa City, IA 52242 USA (e-mail: andrew-kusiak@uiowa.edu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TSTE.2010.2046919 1 Available: http://en.wikipedia.org/wiki/Wind_power. genetic algorithm to optimize the efficiency of a small-scale turbine. The goal of this paper is to model and optimize wind turbine performance in three objectives, maximization of the power pro- duced by a wind turbine, and minimization of vibrations of the turbine’s drive train and tower. Numerous studies of wind power models have been reported in the literature [27], [29]. A passive control method using a tuned mass damper to mitigate vibrations of the blades and tower of a wind turbine was presented in [14]. The research re- ported in [15] discussed the estimation of aero-elastic damping of operational wind turbine modes based on experiments. The majority of the published research falls into parametric and physics-based models. This paper illustrates nonlinear and nonparametric models for optimization of wind power and vibration using a data-driven approach. Such an approach has been successfully applied to optimize power plants and industrial processes [32]. The sources of wind turbine vibrations [25] are diverse. The focus of this paper is on vibrations attributed to the control of wind turbines, e.g., control of the generator torque and blade pitch. Two parameters, drive train acceleration and tower ac- celeration, are selected to represent vibrations of the drive train and tower. Two data-driven models of wind turbine vibrations are developed, one to predict the drive train accelerations and the other to predict the tower accelerations. The power output is also modeled by a similar methodology. Neural network (NN) [16]–[18] is applied to extract these data-driven models from industrial (wind turbine) data. The three models are then integrated into a multiobjective optimization model [19]. As the models are nonparametric and nonlinear, obtaining analyt- ical form solutions is difficult, and therefore, an evolutionary strategy (ES) algorithm [20], [21], [26] is used to solve them. Different control preferences lead to numerous control strate- gies. The data used in this research was obtained from a large (150 MW) wind farm, and its sampling frequency is 0.1 Hz. Since the frequency of wind turbine vibrations is higher than 0.1 Hz, the information loss due to the low (0.1 Hz) frequency of available data has been reflected in the research results. To address the information loss, a 1-min (lower frequency) data set is derived from the 0.1-Hz (10-s) data set. Computational exper- iments with the two data sets, i.e., 10 s and 1 min, demonstrated a potential for further reduction of turbine vibrations. Due to the limited data frequency, this paper investigates the potential for vibration reduction by adjusting certain controllable param- eters, such as blade pitch angle and generator torque. Industrial implementation of the approach proposed in this paper calls for higher frequency data. 1949-3029/$26.00 © 2010 IEEE Authorized licensed use limited to: The University of Iowa. Downloaded on June 25,2010 at 17:49:01 UTC from IEEE Xplore. Restrictions apply.