基于噪声DQN的智能船舶全局路径规划方法  

The intelligent ship global path planning method based on Noise⁃DQN

作　　者：詹天碧 冯辉[1,2] 徐海祥[1,2] 汪咏 ZHAN Tianbi;FENG Hui;XU Haixiang;WANG Yong(School of Naval Architecture,Ocean and Energy Power Engineering,Wuhan University of Technology,Wuhan 430063,China;Key Laboratory of High Performance Ship Technology,Ministry of Education,Wuhan University of Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学船海与能源动力工程学院,武汉430063 [2]武汉理工大学高性能船舶技术教育部重点实验室,武汉430063

出　　处：《大连海事大学学报》2025年第1期43-53,共11页Journal of Dalian Maritime University

基　　金：国家自然科学基金资助项目(52371374;51979210)。

摘　　要：为解决基于深度Q网络(DQN)算法进行智能船舶全局路径规划时存在的规划路径距离障碍物过近、拐点过多、大拐角及算法收敛速度较慢等问题,提出基于噪声DQN(Noise-DQN)的全局路径规划方法。首先,为使智能船舶与障碍物保持安全距离,减少路径拐点和大拐角,在传统奖励函数的基础上增加了额外的航向奖励函数、时间奖励函数、拐点奖励函数和安全奖励函数。然后,针对复杂航行场景中算法收敛速度慢的问题,在DQN神经网络的输出层引入参数噪声,提高了DQN网络收敛速度。最后,针对大连和舟山实际海域环境开展仿真研究。结果表明,本文提出的Noise-DQN算法相比传统DQN算法,收敛速度得到了显著提升;规划的全局路径在安全性和经济性方面得到了大幅度提高,更加符合船舶的实际航行需求。研究成果可为智能船舶全局路径规划提供一定的参考。In order to address the challenges encountered in global path planning via deep Q⁃network (DQN) algorithm,such as paths being planned too close to obstacles, excessiveturning points, large turning angles, and slow convergencespeed, a global path planning method using Noise⁃DQN wasproposed for an intelligent ship. Firstly, to maintain safe dis⁃tances between intelligent ships and obstacles, while reducingpath turning points and angles, additional reward functions in⁃cluding heading, time, turning point, and safety constraintswere incorporated. Then, to address slow convergence speedin complex navigation scenarios, parameter noises were intro⁃duced into the output layer of the DQN neural network, there⁃by increasing the convergence speed of the DQN network. Fi⁃nally, simulation studies were conducted in real⁃world marineenvironments of Dalian and Zhoushan waters. The results showthat the proposed Noise⁃DQN algorithm has significantly im⁃proved convergence speed compared to the traditional one.The resulting global path has been significantly improved interms of safety and economy, and is more compatible with thepractical navigation needs of ships. This research can providea reference for global path planning of intelligent ships.

关 键 词：智能船舶 路径规划 深度强化学习 深度Q网络(DQN)算法 高斯噪声 

分 类 号：U675.79[交通运输工程—船舶及航道工程]