Data Driven Turbine Control in Tidal Power Generation: Development and comparison of data driven turbine control algorithms in a simulated underwater kite environment

Abstract

Minesto’s underwater kite extracts energy from ocean and tidal flows through a cross-flow motion and aims to convert as much energy as possible from the surrounding water-flow. To achieve this, designing a generator control signal that applies optimal torque on the turbine shaft is essential. This project aims to develop data-driven control systems for the kites generator that increases the power output relative to a baseline. ODE-based dynamical models of the kite and generator are derived to create a full simulation environment. The simulation is used to generate sensor data, as well as test and validate new control strategies. Two different approaches are investigated, a supervised predictive controller and a deep reinforcement learning controller where the control signal is generated by the learned policy. The supervised predictive controller approach involves creating a supervised dataset from the simulation and training a recurrent neural network to forecast future inflow water velocities. The knowledge of future inflow is then incorporated in the design of two new control methods, one that is built on the existing controller and another that is developed as a standalone control method. The deep reinforcement learning controller instead utilizes the simulation directly by iteratively stepping through the simulated environment and learning an optimal controller from the outcomes. This is achieved through the use of the state-of-the-art deep reinforcement learning algorithm Soft Actor-Critic. The solution also incorporates pre-training on generated data to validate a possible simulation-to-reality adaptation. Within the adopted simulation setting, the main finding is that predictive turbine control based on forecasted inflow can outperform a reactive baseline controller. The recurrent predictive controllers consistently improved generated power across unseen evaluation environments, indicating that short-term prediction and the periodicity of the kite motion are useful for control. The Soft Actor-Critic approach demonstrated learning capability but was more sensitive to reward design, partial observability, and tuning. Because the study relies on a simplified simulation model, the results should be interpreted as proof-of-concept and comparative evidence rather than as direct claims about real-system performance.