基于DDQN强化学习的沥青路面养护决策  

Fine Maintenance Decision of Asphalt Pavement based on DDQN Reinforcement Learning

作　　者：石文康 徐勋倩 康峰沂 顾钰雯 GANHOUEGNON Eric Patrick SHI Wenkang;XU Xunqian;KANG Fengyi;GU Yuwen;GANHOUEGNON Eric Patrick(School of Transportation and Civil Engineering,Nantong University,Nantong 226019,China;Nantong Highway Development Center,Nantong 226019,China)

机构地区：[1]南通大学交通与土木工程学院,江苏南通226019 [2]南通市公路事业发展中心,江苏南通226019

出　　处：《粉煤灰综合利用》2024年第4期147-153,共7页Fly Ash Comprehensive Utilization

基　　金：国家重点研发项目(2016YFB0303100)。

摘　　要：通过DDQN强化学习的方法开展路面养护决策分析,以路面长期效益费用比的最大化为目标构建养护决策模型,计算出效益费用比更优的养护方案。模型以道路条数和使用年限为状态特征,以四种养护措施为动作空间,以路面养护效益与资金比值作为奖励,构建了一种动作选择策略,使养护方案满足最低使用要求。结果表明:基于DDQN养护决策模型的收敛速度比DQN模型快1倍,计算出的养护方案具有较高效益费用比,路面处于优良状态。This paper employs a Double Deep Q-Network(DDQN)reinforcement learning approach to analyze pavement maintenance decisions,aiming to maximize the long-term benefit-cost ratio of the pavement.A maintenance decision model is constructed to calculate a more cost-effective maintenance plan.This model uses the number of road segments and years as state features,four maintenance measures as the action space,and the ratio of pavement maintenance benefits to costs as the reward.An action selection strategy is proposed,which ensures that the pavement meets operational requirements.Practical engineering data is used as a case study.The results indicate that the convergence speed of the DDQN-based maintenance decision model is twice as fast as the Deep Q-Network(DQN)model.The calculated maintenance plan demonstrates a higher benefit-cost ratio,keeping the pavement in excellent condition.

关 键 词：沥青路面 路面养护决策 深度强化学习 养护方案 

分 类 号：U418.6[交通运输工程—道路与铁道工程]