A Data-Driven Pitch Angle and Torque Control Method for Enhanced Wind Farm Operation Performance and Efficiency

This paper proposes a Reversible Depth-wise Gated Network (RDG-Net) that controls pitch angle (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\beta $</tex-math></inline-formula>) and generator torque (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${{T}_g}$</tex-math></inline-formula>) for wind farm (WF). The proposed RDG-Net can accurately forecast the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\beta $</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${{T}_g}$</tex-math></inline-formula> outputs of wind turbines (WTs) for fatigue load suppression, power extraction and calculation efficiency improvement. The RDG-Net eliminates errors from linearization approximations in WT model, thus enhancing control performance. The method employs a reversible instance and depth-wise separable convolution (Revin-DSCNN) model for WTs. Then, a framework based on multi-head attention and gated graph recurrent neural networks (multi-GGRNNs) is adopted to model coupling relationships among WTs, such as the wake effect that significantly impacts performance. The proposed method replaces high-dimensional mathematical models and avoids computational complexity for joint modeling all WTs. Moreover, the RDG-Net models are implemented on distributed servers to gather real-time WF data for online training, thus enhancing adaptability and generalization ability. The effectiveness of the proposed RDG-Net is validated through simulations in MATLAB.