基于深度强化学习技术的变电站巡检机器人的路径规划研究  

Research on Path Planning of Substation Inspection Robot Based on Deep Enhanced Learning Technology

作　　者：张晶[1] 蔡志全 韩永成 高丹[1] ZHANG Jing;CAI Zhiquan;HAN Yongcheng;GAO Dan(Tangshan Polytechnic College,Tangshan,Hebei 063202,China)

机构地区：[1]唐山工业职业技术学院,河北唐山063202

出　　处：《冶金动力》2021年第3期4-7,14,共5页Metallurgical Power

基　　金：2020年唐山市应用基础研究计划项目(课题编号:20130229b);2019年唐山市科学技术研究与发展计划〈第七批〉项目(课题编号:19130225g)

摘　　要：由于机器人的路径规划能力强弱决定着变电站巡检机器人的作业质量,针对当前的变电站巡检机器人路径规划误差大、效率低下的问题,设计了改进型卷积神经网络GNN与Q网络相结合的高效规划路径巡检机器人。解决巡检机器人在复杂环境下克服运动路径规划中遇到的障碍,提高变电站巡检机器人的工作效率。使用深度强化学习技术以提高机器人在巡检环境中的探索能力和自学习能力,降低巡检机器人路径规划时间,路径规划精准度达到98.3%。为变电站巡检机器人智能化巡检奠定了基础,为变电站无人化的实现提供智能解决方案。Because the ability of robot path planning determines the operation quality of substation inspection robot,aiming at the problems of big error and low efficiency of current substation inspection robot path planning,an improved convolution neural network GNN combined with Q network is designed to effectively plan the path inspection robot.The aim is to overcome the obstacles in the path planning of inspection robot in complex environment,improve the efficiency of inspection robot at substation.Deep enhanced learning technology is used to improve the robot’s exploration and self-learning ability in the inspection environment and reduce the path planning time of inspection robot.The path planning accuracy has reached 98.3%.It lays a foundation for the intelligent inspection of substation inspection robot,and provides an intelligent solution for the realization of unmanned substation.

关 键 词：巡检机器人 路径规划 深度强化学习 无人化 

分 类 号：TP23[自动化与计算机技术—检测技术与自动化装置]