Maximum power point tracking (MPPT) system of small wind power generator using RBFNN approach

Abstract

A novel approach of combination of radial basis function neural network (RBFNN) and particle swarm optimization (PSO) is proposed to achieve the maximum power point tracking (MPPT) in this study. The measured data of the small wind generator (250 W), including wind speed, generator speed and output power of wind power generator, are applied to estimate the wind speed and output power by the proposed wind speed ANN_wind and power estimation ANN_Pe-PSO modules, respectively. Using the predicted results by the two modules of Matlab/Simulink, the MPPT point can be obtained by manipulating the generator speeds. The experimental results show that the proposed RBFNN-based approach can increase the maximum output power of the wind power generator even if the wind speed and load varies.

Introduction

Small wind power generators are widely used in metropolitan areas nowadays, and from the official reports of Taiwan Power Company, the installation capacities of wind generators in Taiwan have grown total 278MVA in past years (Bureau of Taiwan Energy, 2010). Thus, how to offer more power to electrical utilities by promoting the efficiency of wind power is the major concern to all researchers. The metropolitan wind energy always varies instantaneously and the maximum power output cannot be easily obtained. Many literatures relating to the maximum power point tracking (MPPT) have been proposed (Li et al., 2005, Veerachary et al., 2003). These studies revealed that wind energy systems are nonlinear and difficult to be controlled. Neural networks (NNs) have been used for modeling of the complex systems because the models structured by NNs can descript the relationship between input and output of a complex nonlinear system without any detailed analytical model of the system (Hsu, 2010, Yilmaz and Oezer, 2009). In these literatures, a number of approaches based on the NN structure have emerged as a tool for the difficult control problem of the unknown nonlinear systems. However, most of NN based approaches have to use all the training data as the neurons of hidden layer, and the higher complexity consequently consumes much computation time. And it is inappropriate to small wind generators in metropolitan areas whose wind energy sources always varied. To solve the problem, an appropriate NN for MPPT should have the characteristics of lower complexity and computation time, and the instantaneous wind energy might be easily and efficiently captured. We therefore propose a novel approach which combines the advantage of the radial basis function neural network (RBFNN) and particle swarm optimization (PSO) to estimate the appropriate speed of a wind turbine (Acharjee and Goswami, 2009, Balasubramanian et al., 2009, Chang et al., 2009, Chen and Yan, 2008, Cheng et al., 2009, Dhanalakshmi et al., 2009, El-Zonkoly et al., 2009, Kuo et al., 2009, Lei et al., 2009, Quah, 2008, Subashini et al., 2009, Sun, 2009, Tang et al., 2009, Tseng et al., 2009, Wu, 2010, Wu et al., 2009, Zahara and Kao, 2009). In this study, we used an experimental test based on a wind turbine testing platform to verify the superiority of the proposed approach in MPPT when the wind speed and load impedance vary simultaneously.