Abstract
Hydrokinetic turbines (HKTs) are a promising renewable energy source due to the consistency and high energy density in river and tidal resources. One of the primary barriers to the widespread adoption of HKT technologies is a high levelized cost of energy (LCOE). Considering the marine operating environment, the operation and maintenance costs are substantial. The power electronic converter, a key element in the electrical energy conversion system, is a common point of failure in direct-drive turbine applications—leading to increased maintenance efforts. This work presents a reinforcement learning (RL) method built within a quadratic feedback torque control framework to balance energy generation with power electronic device lifetime. The effectiveness of the RL-based control scheme is compared against a static baseline controller through two year-long tidal case studies. The results showed that the proposed method reduced cumulative damage on the device by upwards of 75% but reduced energy generation by up to 25.2%. Using a custom real-time cost estimation function that considers the sale of energy and an estimate of the costs associated with operating a device at a given temperature, it was found that the RL method can increase net income by up to 45.4% depending on the energy market conditions.