Improving Floating Offshore Wind Farm Flow Control With Scalable Model-Based Deep Reinforcement Learning

This paper proposes a model-based deep reinforcement learning (DRL) framework to maximize the total power output and minimize the fatigue load of a floating offshore wind farm subject to wake effect. Recognizing the extensive interactions required for the DRL training, we first develop an open-source physics-based model that describes the time-averaged dynamics of the floating wind farm. This model is designed with sufficient fidelity to support the wind farm control and high computational efficiency to facilitate DRL training. Subsequently, a model-based DRL approach is proposed, featuring simultaneous learning of system dynamics and optimal control policies. This dual learning process enhances the scalability of the DRL agent, making the framework suitable for large-scale floating wind farms. Finally, the effectiveness of the proposed scheme is validated by case studies with a dynamic floating wind farm simulator FAST.Farm. Note to Practitioners—This paper was motivated by the problem of improving the energy production and reducing fatigue load for floating offshore wind farms affected by wake effects, but it also applies to other types of wind farms. Traditional approaches to wind farm optimization often struggle with the computational complexity of modeling turbine interactions and wake dynamics. To address this, this paper proposes a new approach for optimal wind farm flow control using model-based DRL, which simultaneously learns system dynamics and optimizes control strategies. This integrated approach enhances the training stability and scalability of the DRL agent, making it suitable for large-scale wind farms. Simulation results on a dynamic wind farm simulator FAST.Farm suggest that this approach is feasible but it has not yet been incorporated into a real wind farm energy management system. Future work will explore its integration with real-time monitoring systems.