Abstract
As a sustainable and reliable renewable energy source, tidal-wave energy has become an increasing subject of interest for generation in coastal areas to meet the growing demand for this type of predictable and environmentally friendly power. Nevertheless, highly dynamic and stochastic marine conditions such as variations of tidal velocity, turbulence intensity, hydrodynamic loading represent challenges for maintaining stable operation of grid-connected tidal energy systems. Such uncertainties can greatly hamper power extraction efficiency, structural safety and the stability of the grid. This paper presents a new reinforcement learning–based multi-objective optimization and adaptive control framework for grid-connected tidal energy conversion systems to maximize the energy capture while guaranteeing the system reliability and robustness against uncertainties of operation environments. The proposed framework combines deep reinforcement learning (RL) - agent with a multi-objective optimization mechanism to facilitate smart, real-time decision-making. The RL agent learns the optimal control policies and keeps interacting with tidal system and grid environment by adjusting operating parameters of turbine such as blade pitch angle, generator torque and power converter settings. The objective function consists of a weighted linear combination of several objectives such as maximizing power, minimizing loads and fatigue on structures, mechanical stress minimization, and grid integration stability requirements. An evolutionary multi-objective optimization layer is then built on top, to drive control parameters dynamically by resolving Pareto-optimal trade-offs among conflicting objectives. A grid-connected tidal energy system with high fidelity, is modeled including hydrodynamic turbine models, drivetrain dynamics, power electronic converters and the grid interface components. Realistic tidal flow profiles with added turbulence, random perturbations and environmental variability are used to obtain representations of stochastic marine conditions. The method is tested under various operating scenario, such as different tidal speed, impulsive flow variation and grid disturbances. Simulation results confirm the effectiveness of the presented RL-based multi-objective control framework compared to traditional control approaches, including fixed-pitch control and model predictive control (MPC). The system has higher power capture efficiency, less load fluctuation, and greater structural durability. In particular, the power extraction increases by up to 20%, while structural load changes are reduced by more than 30%, which prolongs system life. In addition, the suggested controller is capable of keeping stable operation of the grid with enhanced power quality and smaller variations in stochastic states. The findings demonstrate the success of combining reinforcement learning with multi-objective optimization in advanced control approaches for tidal energy. This tool provides a scalable and adaptive framework for next-generation tidal energy systems to support reliable, efficient coastal power generation. This work furthers intelligent renewable energy systems at sea and assists the transition to a sustainable marine energy infrastructure.