基于分层强化学习的矿区无人驾驶车辆路径规划算法  

Path Planning Algorithm of Unmanned Vehicle in Mining Area Based on Hierarchical Reinforcement Learning

作　　者：魏晓娟 李纪云[1] 巩闯 WEI Xiaojuan;LI Jiyun;GONG Chuang(College of Modern Information Technology,Henan Polytechnic,Zhengzhou 450046,China;School of Electronic Engineering,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]河南职业技术学院现代信息技术学院,河南郑州450046 [2]华北水利水电大学电子工程学院,河南郑州450046

出　　处：《金属矿山》2024年第11期192-198,共7页Metal Mine

基　　金：2022年度河南省重点研发与推广专项(科技攻关)项目(编号:222102240049);2022年度河南职业技术学院校级科研项目(编号:2022ZK35)。

摘　　要：在复杂而危险的矿区环境中,矿区无人驾驶车辆的路径规划涉及如何使车辆智能地选择最佳路径,以实现安全和高效运行。然而,传统的路径规划算法难以有效应对矿区内多变的路况和环境。提出了一种基于分层强化学习的矿区无人驾驶车辆路径规划算法,该算法通过分层强化学习技术训练图指针网络,求解矿区无人驾驶车辆路径规划问题。为将矿区无人驾驶车辆节点的向量映射成低维稠密向量,首先对图嵌入层的上下文向量进行均值化处理,用于保持网络的全局属性。再将交叉熵损失函数的范式加入分层强化学习的基准函数中,用于衡量2个不同驾驶车辆间的差异分布程度。试验结果表明:该算法在复杂的矿区环境下能够实现高效、安全、智能的路径选择,且模型收敛速度、时间花费上的优化效果优于传统算法和专业求解器,并具有良好的适应性和泛化能力。研究结果对于提高矿区无人驾驶的自主性、效率和安全性具有重要意义。In complex and dangerous mining environments,path planning for unmanned vehicles in mining areas involves how to enable the vehicle to intelligently choose the best path to achieve safety and efficient operation.However,the traditional path planning algorithm is difficult to effectively deal with the changing road conditions and environment in the mining area.A path planning algorithm for unmanned vehicles in mining area based on hierarchical reinforcement learning is proposed.The algorithm trains graph pointer network by hierarchical reinforcement learning technology to solve the path planning problem of unmanned vehicles in mining area.In order to map the vector of the unmanned vehicle nodes in mining area into a low-dimensional dense vector,firstly,the context vector of the graph embedding layer is normalized to maintain the global properties of the network.Then,the cross entropy loss function is added to the benchmark function of hierarchical reinforcement learning to measure the difference distribution degree between two different driving vehicles.The experimental results show that this algorithm can realize efficient,safe and intelligent path selection in complex mining environment,and the optimization effect of model convergence speed and time cost exceeds the traditional algorithm and professional solver,and has good adaptability and generalization ability.The study results are of great significance for improving the autonomy,efficiency and safety of autonomous driving in mining areas.

关 键 词：矿区无人驾驶车辆 分层强化学习 路径规划 图指针网络 

分 类 号：TD67[矿业工程—矿山机电]