Dynamic Energy-Efficient Path Planning for Unmanned Surface Vehicles Based on SAC-BSTFN
To address the limited endurance of Unmanned Surface Vehicles (USVs) in time-varying sea conditions, this study investigates global energy-efficient path planning that balances obstacle avoidance and energy efficiency. First, based on ship seakeeping theory, an energy consumption model incorporating wave height, wave period, speed, and wave-encounter angle is constructed to represent the impact of dynamic sea states on resistance. The model is assessed through formula–program consistency verification, a multi-factor input ablation on a physics-constrained benchmark, and external trend validation against published towing-tank added-resistance data. Second, the planning problem is modeled as a Partially Observable Markov Decision Process (POMDP). Built upon the Soft Actor-Critic (SAC) algorithm, a Bimodal Spatio-Temporal Feature Fusion Network (BSTFN) is proposed to achieve deep fusion of spatial perception information and historical temporal sea state sequences for decision-making. Furthermore, a composite reward function is designed, integrating energy consumption penalties, heading guidance, and smoothness constraints. Simulation results demonstrate that the proposed method effectively utilizes favorable encounter angles to avoid high sea state regions. While maintaining high task success rates, it significantly reduces average energy consumption, effectively enhancing the endurance and robustness of USVs in complex dynamic environments.