Improving Modelling Accuracy of Aerodynamic Curve of a Wind Turbine Using Neural Networks

Abstract

This paper addresses improved modelling of one of important aerodynamic curves of a wind turbine by means of artificial neural networks (ANNs). Aerodynamic curves play an important role in designing controller in wind turbines. Inherent nonlinearity of these curves and dependence of their current values to the operating conditions, make the wind turbine controller design a challenging problem. Currently, there are two major approaches for modelling these curves: 1- lookup tables and 2-polynomial approximation. Lookup tables are discrete and hence not suitable for continuous controller design and polynomial approximations are not accurate enough. These drawbacks impose inaccuracy to the controller design. To overcome this weakness, ANN is utilized to identify the aerodynamic curves. Specially, rotor power coefficient (Cp) is the focus of this paper as this curve has a direct effect on the controller’s parameters both in below and above rated wind speed. As ANNs are universal approximators, they can model this curve with required accuracy. Using this approach in addition to identification of Cp and obtaining a high accuracy model for this curve, optimum critical parameters of this curve can be estimated. By employing these estimated values, a new controller gain is computed. This controller is used when the wind speed is below rated speed and the rotor speed should track a reference trajectory (named variable speed or region II). Simulation shows that with this new controller the overall power capture is improved at no cost.