基于路径搜索DQN的特殊车辆路线优化策略  

Special vehicle route optimization strategy based on route search DQN

作　　者：肖洪祥[1] 赵子寒 杨铁军 XIAO Hong-xiang;ZHAO Zi-han;YANG Tie-jun(College of Information Science and Engineering,Guilin University of Technology,Guilin 541006,China;College of Intelligent Medicine and Biotechnology,Guilin Medical University,Guilin 541199,China)

机构地区：[1]桂林理工大学信息科学与工程学院,广西桂林541006 [2]桂林医学院智能医学与生物技术学院,广西桂林541199

出　　处：《计算机工程与设计》2024年第10期3153-3160,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62166012);广西自然科学基金项目(2022GXNSFAA035644)。

摘　　要：为保障特殊车辆在复杂且易拥堵的城市交通环境下执行紧急任务的时效性与畅通性,提出一种基于路径搜索式深度Q网络(P-DQN)的特殊车辆路线优化策略。采用回溯法协助深度Q网络(DQN)解决路径搜索过程中的死路、回路问题,利用人工势场机制引导DQN搜索路径,避免路径结果过长。结合轮盘赌选择法与贪婪值自适应调整机制进一步提升DQN选取路段和建议行驶速度时的准确性。实验在InTAS数据集上对真实城市交通进行模拟,与RERoute、CH等SOTA方法相比,P-DQN获得的总价值提高约16%。To ensure the timeliness and smoothness of special vehicles performing emergency tasks in complex and congested urban traffic environment,a special vehicle route optimization strategy based on path search deep Q network(P-DQN)was proposed.The backtracking method was used to assist the deep Q network(DQN)to solve the problem of dead ends and loops in the path search process,and the artificial potential field mechanism was used to guide the DQN search path to avoid excessive path results.The roulette selection method and the greedy value adaptive adjustment mechanism were combined to further improve the accuracy of DQN when selecting road sections and suggesting driving speeds.Experiment simulated the real urban traffic on the InTAS dataset.The total value obtained using P-DQN is increased by about 16%compared with that using SOTA methods such as RERoute and CH.

关 键 词：特殊车辆 深度学习 路线优化 建议行驶速度 智能交通 强化学习 人工势场 轮盘赌选择法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]