Real-Time Vertical Path Planning Using Model Predictive Control for an Autonomous Marine Current Turbine.

This paper presents a predictive approach to address real-time vertical path planning for a marine current turbine (MCT) treated as an autonomous underwater vehicle (AUV), where the path control goal is to maximize the total harvested ocean current energy. The real-time path planning is formulated as a sequence of optimization problems over a prediction horizon with respect to the autonomous MCT model and underwater environment model. The ocean current velocity is modeled through a spatiotemporal neural network (STNN) trained using field-collected acoustic Doppler current profiler (ADCP) data. Model predictive control (MPC)-based approach is proposed to solve the optimizations, where the proposed approach takes advantage of fast discrete path planning (i.e., path planning in a gridded ocean environment) to seek the initial solution, as well as continuous path planning to improve the initial solution in a continuous ocean environment. Results demonstrate that the proposed reinforced continuous path planning algorithm can find a better solution (i.e., optimal path) than independent continuous path planning.