Data-Driven Control Strategy to Increase Active Power in Wake-Affected Wind Farms

The wake effect in wind farms (WFs) leads to wind energy loss of wind turbines (WTs). This paper proposes a data-driven control (DDC) strategy to optimize pitch angle and active power of WTs in wake-affected WFs. As the data resource, a model predictive control (MPC) WTs control method is proposed to increase the total active power generation of WF for generating the effective dataset, i.e., wind speed, direction variation, generator speed, pitch angle, generator torque, and active power. Then the DDC without computational complexity for joint modeling all WTs is proposed to accurately forecast the pitch angle and active power outputs of WTs by capturing the nonlinear aerodynamic interactions. It utilizes an attention-convolutional neural network (CNN) encoder and three-layer graph attention network (GAN)–based decoder as a data-driven framework. The attention-CNN encoder can capture both global and local features, enhancing the model’s expressive ability, robustness, and generalization capabilities. The three-layer GAN-based decoder captures the nonlinear aerodynamic interactions among WT parameters to improve the performance of pitch angle and active power outputs. The efficacy of DDC is substantiated through case studies involving a wake-affected WF layout in MATLAB.