文献类型：期刊文献

中文题名：基于深度强化学习的船舶路径规划方法研究

英文题名：Research on ship path planning method based on deep reinforcement learning

作者：杨长兵[1,2];张海华[1];刘焕牢[2]

机构：[1]南方海洋科学与工程广东省实验室(湛江),广东湛江524000;[2]广东海洋大学机械工程学院,广东湛江524000

年份：2024

卷号：48

期号：10

起止页码：128

中文期刊名：信息技术

外文期刊名：Information Technology

收录：CSTPCD

基金：广东省科研事业单位重点领域研发计划(2020B1111-500001)。

语种：中文

中文关键词：船舶路径规划;深度强化学习;改进人工势场;奖励函数;路径平滑

外文关键词：ship path planning;deep reinforcement learning;improved artificial potential field;reward function;path smoothing

中文摘要：针对现有路径规划算法在面对复杂环境时需要大量先验信息,并存在计算量大、转折过多、搜索精准度差等问题,使用深度强化学习算法可以弥补上述缺陷,但是存在算法本身收敛慢等问题。针对此问题,提出使用改进人工势场法(APF)对深度强化学习算法的奖励函数进行优化处理,并通过贝塞尔曲线对路径进行平滑处理,最终输出相对平滑的船舶航行路径。在相同环境下,将改进算法模型与现有方法的路径规划效果进行比较分析,结果表明,DQN-APF算法在生成的路径长度、平滑度、规划完成时间等船舶路径综合规划参数能力上得到了提升。

外文摘要：The existing path planning algorithm needs a large amount of prior information in the face of complex environment,and has problems such as large amount of computation,excessive transition and poor search accuracy.The use of deep reinforcement learning algorithm can make up the above defects,but the algorithm itself convergence is slow and other problems.To solve this problem,an improved Artificial Potential Field method(APF)is proposed to optimize the reward function of the deep reinforcement learning algorithm,and the path is smoothed by Bessel curve.Finally,a relatively smooth sailing path is output.Under the same environment,the effect of the improved algorithm model and the existing method of path planning is compared and analyzed.The results show that the DQN-APF algorithm has improved the ability of the generated path length,smoothness,planning completion time and other ship path comprehensive planning parameters.