B-spline neural network for real and reactive power control of a wind turbine

The reliability of a microgrid lies highly into quality and response of control schemes. This article presents a primary control to manipulate real and reactive power flow between a wind turbine and a maingrid, based on an artificial neural network using B-spline functions trained with the least mean square error algorithm. The system was modeled by differential equations to represent the mechanical turbine, induction machine, transformer, maingrid, and excitation system, which is based on a voltage source inverter. Besides, a wind pattern obtained from a Mexican meteorological station at Benito Juárez, Oaxaca, was used on the analysis. The performance of the control was verified using constant references and considering step changes on the values that produce a sudden decrease in the mechanical torque reference by 20% and simultaneously, an increase in the reactive power reference by 20%. A maximal deviation of 6% during 1.4 s and 1.67 s for the constant and step change references values was obtained. The stable state error was below 5% in the mechanical torque for both experiments. Regarding the reactive power, errors below 2% and 6% were, respectively, obtained for the analyzed cases. In both cases, the voltage magnitude had a maximal deviation of 1% with respect to the reference. The proposed strategy successfully controlled the power flow considering the 5% criterion, overcoming the nonlinearity of the wind turbine power coefficient and having a small voltage deviation.