A Decentralized Demanded Power Tracking and Voltage Control Method for Wind Farms Based on Data-Driven Sensitivities

Efficient power dispatch in wind farms (WFs) hinges on precise demanded power tracking. This study proposes a decentralized WF power tracking and voltage control method based on data-driven sensitivities (DDSs). This method relies only on local operational variables for model predictive control (MPC), achieving near-global optimal solutions. With a backpropagation algorithm, a new sensitivity calculation method is designed to yield DDSs by computing the gradients of a global mapping model (GMM). The voltage DDSs can be derived simply by calculating the gradient of the voltage GMM and can replace the voltage sensitivities in traditional MPC methods. The power DDSs establishes linear relationships between the power outputs of different wind turbines (WTs), simplifying the WF state-space equations to local prediction models for reducing the quadratic programming dimensions. The three control modes designed based on DDSs enable control without WF line parameters, reduce computational complexity, or combine both effects. The variable spacing constraint linearization method transforms nonlinear constraints into linear ones, addressing the nonlinear coupling between control variables. Testing on a WF with 32 WTs in MATLAB/Simulink demonstrates the effectiveness of the proposed method comparable to centralized control methods.