Deep learning-aided model predictive control of wind farms for AGC considering the dynamic wake effect

To provide automatic generation control (AGC) service, wind farms (WFs) are required to their operation dynamically track the time-varying power reference. Wake effects impose significant aerodynamic interactions among turbines, which remarkably influence WF dynamic production. The nonlinear and high-dimensional nature of wake model, however, brings extremely high computation complexity obscure design controllers. This paper overcomes difficulty brought by model proposing a novel control-oriented reduced order deep-learning-aided predictive (MPC) method. Leveraging recent advances in computational fluid dynamics (CFD) high-fidelity data that simulates flows, two deep neural network (DNN) architectures specially designed learn reduced-order (ROM) can capture dominant flow dynamics. Then, MPC framework is constructed explicitly incorporates obtained ROM coordinate different turbines while considering interactions. proposed method evaluated widely-accepted simulator whose accuracy has been validated realistic measurement data. A 9-turbine case larger 25-turbine studied. By reducing states many orders magnitude, burden greatly. Besides, through method, range AGC signals be tracked extended compared with existing greedy controller.