Optimal Path Planning Based on ID3QN for Vessel Navigation in Port Areas

Effective path planning is crucial for ensuring navigational safety and optimizing resource application during vessel operations in port areas. Despite advances in DRL methods, exploration inefficiency persists as a fundamental challenge. To address it, we propose ID3QN: an improved Double Dueling Deep Q-Network (D3QN) algorithm. This framework integrates an Artificial Potential Field (APF) method to address reward sparsity and employs an adaptive εε-greedy strategy to improve exploration efficiency. The APF-based reward function incorporates four refined components: distance reward, collision avoidance penalty, exclusion penalty, and maritime domain reward. Concurrently, the adaptive strategy dynamically adjusts exploration rates based on real-time vessel feedback. Experimental validation within a 650×650-pixel simulation environment using 200 randomized navigational charts demonstrates ID3QN's superiority over baseline models (DQN with ε=0.1, D3QN, and DQN with APF) across three metrics: training efficiency, navigational safety (98.5% success rate), and path optimality (average length: 508.34 pixels). These results confirm ID3QN's effectiveness in optimizing intelligent vessel trajectory planning for port navigation.