广东电力2023，Vol.36Issue(11)：114-121,8.DOI:10.3969/j.issn.1007-290X.2023.11.012
基于泛化强化学习的变电站巡检机器人路径规划研究
Study on Substation Inspection Robot Path Planning Based on Generalization Reinforcement Learning
摘要
Abstract
Aiming at the path planning problem for substation inspection robots,this paper propose a state generalization deep reinforcement learning(DRL)algorithm.On the basis of original algorithm,it introduce a regularization for generalization into the training process,which incorporates two auxiliary training pools to ensure the network receives more supervisory information in the early training stage,thus enhancing the model's generalization capacity in different scenarios of the same task.Compared to traditional training methods,the proposed method achieves state generalization in certain environments where traditional methods may fail,without sacrificing model performance.Experiments demonstrate that this proposed algorithm outperforms the deep quality-network(DQN)and double deep quality-network(DDQN)algorithms in path planning tasks with obstacles.关键词
强化学习/深度强化学习/路径规划/泛化性Key words
reinforcement learning/deep reinforcement learning/path planning/generalization分类
信息技术与安全科学引用本文复制引用
易仕琪,孔政敏,王帅,霍梓航..基于泛化强化学习的变电站巡检机器人路径规划研究[J].广东电力,2023,36(11):114-121,8.基金项目
国家科技创新2030课题项目(2021ZD0112702) （2021ZD0112702）
国家自然科学基金项目(62173256) （62173256）
广东省智能电网新技术企业重点实验室资助项目(2020B1212070025) （2020B1212070025）
