一种基于物理蒸馏强化学习的船舶预设时间跟踪控制方法
By employing physical distillation reinforcement learning, combined with teacher and student network architectures and an execution-evaluation network, the problem of precise tracking and energy consumption optimization in unknown dynamic environments during ship control was solved, achieving efficient and economical navigation within a preset time.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN MARITIME UNIVERSITY
- Filing Date
- 2026-05-15
- Publication Date
- 2026-07-17
AI Technical Summary
Existing ship control technologies lack the ability to effectively integrate prior physical knowledge into deep reinforcement learning, making it difficult to achieve accurate tracking and optimal control under strict time constraints. Furthermore, traditional methods face risks of energy consumption spikes and model divergence in unknown dynamic environments.
A physical distillation-based reinforcement learning approach is adopted. The ship's energy dissipation rules are extracted through a teacher-student network architecture. An adaptive virtual control law is constructed by combining backstepping and a saturated preset time function. An execution-evaluation network is introduced to perform optimal feedback control, thereby achieving path tracking within a preset time.
In an unknown dynamic environment, the ship path tracking error was strictly converged, reducing energy consumption and optimizing control costs, thereby improving the system's environmental adaptability and navigation performance.
Smart Images
Figure CN122411029A_ABST