Robust Wind Farm Power Optimization Using Model Adaptation via Neural Network Learning

In a wind farm, the wake interactions between turbines can significantly reduce the power output of the wind farm. Cooperative wind farm control considering the interactions can improve the power output of the overall wind farm. However, in general, only a simplified wind farm power generation model can be available due to the highly complex nature of the interactions. This leads to model mismatch and limits the performance of model-based methods. This paper proposes an adaptive calibration model based power optimization method for cooperative wind farm control to handle the model mismatch. The proposed method mainly consists of two iterative phases, i.e., adaptation and optimization. In the adaptation phase, the model mismatch is approximated using neural network with great nonlinear fitting capability, and a calibrated model is then obtained. This phase fully exploits simplified yet computationally tractable power generation model and readily available power generation data from the wind farm. To improve the efficiency of the adaptation phase, a sample data utilization method is developed via mining knowledge from the historical measurements of real wind farm. In the optimization phase, the control actions of all wind turbines are determined using calibrated model, achieving robust power optimization for wind farm. The simulation results demonstrate the effectiveness of the proposed method in compensating for the effect of model mismatch on the wind farm power generation performance.