DeepLEAP: Deep-reinforcement-Learning-based control for Energy harvesting And Platform stabilization of wave energy converters integrated to floating offshore wind turbines

Project Description

Integrating wave energy converters (WECs) with floating offshore wind turbines (FOWTs), to form hybrid wind-wave systems, offers benefits like shared substructures and smoothed power output, providing a promising solution for offshore renewable energy development. The performance of these systems heavily depends on effective control of the integrated WECs, which can enhance energy harvesting and, simultaneously, platform stability, significantly reducing the levelized cost of energy (LCoE). However, current research has focused primarily on platform design, leaving the control system largely unexplored. Existing control approaches, such as damping and reactive controllers, lack flexibility, while model predictive control (MPC) is often overly simplified and suboptimal. This project aims to address the challenges of wind-wave system control posed by complex, changing dynamics and unpredictable sea conditions. This project will develop a novel control framework that employs deep reinforcement learning (DRL) to optimize control laws directly from high-fidelity system dynamics, maximizing performance without model simplification. To ensure safety, this project will further introduce a robust, shielding MPC layer after DRL to maintaining safety under varying wind-wave conditions. This DRL-MPC approach combines the control quality of DRL with the safety guarantees of MPC, while ensuring computational efficiency. The control system will be deployed on an embedded controller and validated through hardware-in-the-loop (HIL) testing. The work packages include FOWT-WEC modelling, DRL design, shielding MPC design, and HIL testing, and will leverage strong support from Prof. Ringwood’s group. The success of this project could revolutionize and set a new standard for the field of wind-wave system control.