Training of an ANN Feed-Forward Regression Model to Predict Wind Farm Power Production for the Purpose of Active Wake Control

Abstract

Wind farm active wake control provides increased power production and reduced mechanical stress on assets by adjusting the yaw angle and power production of individual turbines. While the axial induction control limits a turbine’s power production to decrease the resulting turbulence intensity, the wake redirection control uses the yaw angle to steer wake formations away from subsequent turbines. When applying the wake redirection control, an appropriate choice for the yaw angles must be made that leads to an overall increase in the global power production of the wind farm, even when individual turbine performance decreases. In order to determine optimal yaw angle set points, the project SmartWind uses a regression model based on an artificial neural network to predict the power production for the current wind conditions as part of a control algorithm to optimize the operation of wind farms. In this paper, the training process of the regression model is investigated and the training performance is assessed using real measurements as training data for the network. The trained artificial neural network is able to estimate the wind farm power production for a set of proposed yaw angle set points with a sufficient accuracy of 2-3% of the rated power production. In conjunction with an algorithm determining reasonable yaw angle variations for the current wind conditions, it can thus be used in active wake control methods for wind farms.