Estimation of Useful Variables in Wind Turbines and Farms Using Neural Networks and Extended Kalman Filter

Having access to certain variables, which are normally not measured, could be beneficial to maintenance, condition monitoring, and control of wind turbines and farms. However, incorporating additional sensors to measure such variables could increase the overall cost significantly. It is therefore proposed in this feasibility study that only one turbine in a wind cluster of several turbines be equipped with a sensor and each of the remaining turbines with an estimator, instead of equipping each turbine with an expensive sensor. Each nonlinear estimator is constructed based on neural network (NN), and the only turbine equipped with a sensor is used to train the NN-based estimator, applied to each of the remaining turbines. On the other hand, it is much more complicated and expensive to estimate the wind speed that the turbine experiences using the same approach as it would require at least one expensive light detection and ranging system as a sensor (and for training the NN-based estimator) within a cluster. However, estimating the wind speed could improve maintenance, condition monitoring, and control of wind turbines and farms in many ways. In the second part of the paper, an extended Kalman filter (EKF), instead of NN is therefore constructed to estimate the wind speed. Although the EKF could be used for different purposes, it is exploited here to improve the NN-based estimator from the first part of this paper. The simulation results are presented demonstrating that 1) the NN-based estimator could estimate important variables successfully, 2) the EKF could estimate the wind speed, and 3) the NN-based estimator could be refined or further improved by exploiting the wind speed estimated by the EKF.