基于强化学习的沥青路面长期性能养护决策方法  

Long-term performance maintenance decisions for asphalt pavements based on reinforcement learning

作　　者：侯明业 王晓阳 徐青杰 杨博 王笑风 HOU Mingye;WANG Xiaoyang;XU Qingjie;YANG Bo;WANG Xiaofeng(Henan Communications Planning&Design Institute Co.,Ltd.,Zhengzhou 450000 China)

机构地区：[1]河南省交通规划设计研究院股份有限公司,河南郑州450000

出　　处：《山东科学》2023年第3期108-114,共7页Shandong Science

基　　金：河南省交通运输厅科技项目(2021T2;2021T8;2021G3)。

摘　　要：针对道路长期性能养护决策中庞大的数据分析问题,将深度确定性策略梯度(deep deterministic policy gradient,DDPG)强化学习模型引入到了养护决策分析中,将道路性能的提升及养护资金的有效利用作为机器学习的奖励目标,建立了一套科学有效的沥青路面长期性能养护决策方法,经过与DQN(deep Q-learning network)算法和Q-Learning算法进行对比,DDPG算法所需要的采样数据更少、收敛速度更快,表现更为优异,可有效提升道路服役性能的评估效率,对沥青路面多目标长期养护决策方案的制定起着重要的推动作用。To address the huge data analysis problem in the decision-making for long-term road performance maintenance,this paper introduces the deep deterministic policy gradient(DDPG)reinforcement learning model in the maintenance decision analysis.A set of scientific and effective decision-making methods for long-term performance maintenance of asphalt pavements has been established through machine learning.These methods can improve road performance and make effective use of maintenance funds.Compared with the deep Q-learning network and Q-Learning algorithms,the DDPG algorithm requires less sampling data,converges faster,performs better,and can effectively improve the evaluation efficiency of the road service performance.Therefore,the proposed model plays an important role in the development of multi-objective maintenance decision-making for asphalt pavements.

关 键 词：交通工程 沥青路面 养护决策 强化学习 深度确定性策略梯度模型 

分 类 号：U411[交通运输工程—道路与铁道工程]