Extreme Learning Machine Based Control of Grid Side Inverter for Wind Turbines

Abstract

The use of controller topology called back-to-back is becoming more widespread in full rated control of wind turbines. In back-to-back converter topology, to control the grid side inverter, it is necessary to control the dq currents and dc bus voltage using the vector control method. In order to perform the vector control method, it is important to know the LCL filter parameters used at the inverter output and to select the PI controller parameters in accordance with the obtained transfer function. In the classical design of the controller, the optimal modulus PI controller method is preferred because it facilitates the design process. In this study, as a new method, a controller structure called extreme learning machine based on single hidden layer feed forward artificial neural network is proposed to control the grid side converter. Since the proposed controller structure is analytically trained, it provides a faster solution than the iterative solutions of classical artificial neural networks. Various simulation results are presented on a wind turbine model in which permanent magnet synchronous generator is used to convert mechanical energy from the wind into electric energy. The modulation of the inverter used for energy conversion is performed by the sinusoidal pulse width modulation technique. The simulation results indicated that the extreme learning machine based controller provided successful results.