Reinforcement Learning-Based Wind Farm Control: Towards Large Farm Applications via Automatic Grouping and Transfer Learning

The high system complexity and strong wake effects bring significant challenges to wind farm operations. Conventional control methods may lead degraded power generation efficiency. A reinforcement learning (RL)-based approach is proposed in this paper handle these issues, which can increase the long-term farm-level subject while without requiring analytical models. method significantly distinct from existing RL-based approaches, whose computational complexities usually heavily with of total turbine numbers. In contrast, our greatly reduce training loads enhance efficiency via two novel designs: (1) automatic grouping (2) multi-agent-based transfer (MATL). Automatic Grouping divide a large into small groups by analyzing aerodynamic interactions between turbines utilizing some key principles graph theory. It enables separated conduction RL algorithms on groups, avoiding complex process costs applying entire farm. Based Grouping, MATL further allowing agents (i.e. turbines) inherit policies under potential group changes. Case studies dynamical simulator show that achieves clear increases than benchmark. also dramatically reduces compared typical methods, paving way for application general farms.