As the global shipping industry continues to evolve at a rapid pace, the path planning of intelligent ships in intricate, obstacle-laden, and ocean current-impacted environments has emerged as a focal point of intense research. Amidst the ocean currents that can significantly alter ship speed, ensuring the safe navigation of intelligent ships poses a formidable challenge. To tackle this challenge head-on, this paper constructs a path planning framework based on the Double Deep Q-Network(DDQN). A dynamic composite reward mechanism is seamlessly integrated into this framework, which takes into account both the artificial potential field method and the influences of ocean currents. Additionally, an Adaptive \(\varepsilon \) -Greedy Strategy is introduced to strike an optimal balance between exploration and exploitation. Moreover, a quasi-uniform B-spline function is employed for path optimization, aiming to enhance the navigation stability of intelligent ships. Comprehensive simulation results vividly demonstrate that the Improved DDQN (IDDQN) algorithm outshines benchmark methods, including DDQN, Dueling DQN, A*, APF, and RRT, in several key metrics. The number of turns in the IDDQN implementation was significantly reduced (3.38 ± 1.41), showing a decrease of 44.9-63.0% compared to the baseline algorithms. In terms of the downstream proportion metric, which reflects the alignment with environmental structure, IDDQN achieved 0.930 ± 0.048, outperforming both Dueling DQN and DDQN. Additionally, IDDQN demonstrated exceptional robustness across different environmental structures, with the path length remaining highly stable (coefficient of variation = 4.03%), which was lower than all compared methods. In conclusion, the IDDQN algorithm offers a highly effective solution for the safe path planning of intelligent ships navigating through complex marine environments influenced by ocean currents.
