An Inspection Route Planning Method for Offshore Wind Farm Operation and Maintenance Based on DDQN with a Variable Exploration Rate

To address the high safety risks and complex navigation requirements of offshore wind farm inspection, this paper proposes a DDQN-based path-planning method for unmanned surface vehicles (USVs). The method employs a Double Deep Q-Network (DDQN) with a reward-feedback-based adaptive exploration strategy to reduce performance variation across independent training runs. Offshore wind farm geometry, dynamic-obstacle information, and the International Regulations for Preventing Collisions at Sea (COLREGs) are incorporated into the state and reward design. Experiments were conducted using three independent training seeds, with 6000 training episodes for each seed. The Adaptive DDQN achieved a mean success rate of 83.67% with a standard deviation of 3.51 percentage points, compared with 86.67% and 13.80 percentage points for the Standard DDQN. The worst-run success rate increased from 71% to 80%, indicating lower cross-run variability despite a 3.0-percentage-point reduction in the mean success rate. Additional tests with two dynamic obstacles showed that the learned policy retained task-completion capability under increased environmental complexity. These results suggest that the proposed adaptive exploration strategy can improve the cross-run consistency of DDQN training for USV navigation in dynamic offshore wind farm environments.