摘要
科学合理的吊索维护策略对保障索承桥梁安全运营具有重要作用。针对易损吊索的维护更换决策问题,考虑吊索外观性与结构性损伤状态,提出了一种以桥梁全寿命周期吊索系统维护成本及风险成本之和最低为目标的预防性维护决策方法。根据维护决策问题构建了优化目标函数,以桥梁吊索服役场景为环境,桥梁运维管理系统为智能体,建立了状态空间、动作空间、状态转移概率矩阵和奖励函数,以累计折减奖励的期望代替维护优化问题目标函数,构建了基于马尔可夫决策过程的状态预测和维护决策模型。然后,以吊索系统维护决策模型为基础,基于融合目标网络、经验回放机制的竞争双深度Q网络(Dueling Double Deep Q-Network, D3QN)算法,建立了吊索系统预防性维护决策方法。最后,通过状态预测模型和预防性维护决策方法构建了吊索系统维护决策框架,以一座悬索桥为例进行了分析,通过状态预测模型使智能体与环境不断交互,模拟吊索的劣化及维护过程,产生神经网络训练所需数据,基于交互所得数据训练D3QN算法网络模型,进而获得最优维护策略,并与传统策略进行了对比。结果表明:所提方法综合考虑了桥梁吊索维护成本与结构风险,可动态自适应调整维护策略,与传统策略相比,该方法所得策略可减少维护成本12%以上。
Scientific and reasonable suspender maintenance policies play a major role in ensuring the safe operation of cable-supported bridges.This study addressed the decision-making challenges associated with the maintenance and replacement of vulnerable suspenders by considering their appearance and structural damage state.Accordingly,a preventive maintenance decision-making method that minimizes the combined costs of maintenance and risk throughout the bridge's lifecycle was proposed.First,an optimization objective function was constructed based on the maintenance decision-making problem.The suspender service context was defined as the environment,and the bridge operation and maintenance management system acted as the agent.In addition,the state space,action space,state transition probability matrix,and reward function were established.The expectation of the cumulative discount reward replaced the objective function of the maintenance optimization problem,and state prediction and maintenance decision models based on the Markov decision process were constructed.Then,a preventive maintenance decision method for the suspender system was established based on the suspender system maintenance decision model and dueling double deep Q-network(D3QN)algorithm,which incorporates both a target network and an experience replay mechanism.Finally,a maintenance decision-making framework for the suspender system was constructed using the state prediction model and preventive maintenance decision-making method.With a suspension bridge used as a case study,the state prediction model enabled continuous interaction between the agent and environment,simulating the degradation and maintenance processes of the suspenders while generating the necessary data for training the neural network.Based on the interaction data,the D3QN algorithm network model was trained to obtain the optimal maintenance policy,which was then compared with traditional policies.The results show that the proposed method comprehensively considers the maintenance cost and structural risk and dynamically and adaptively adjusts the maintenance policy.Compared with the traditional policy,the maintenance cost of the policy obtained under the proposed method can be reduced by more than 12%.
作者
马亚飞
晏鹏
何羽
王磊
MA Ya-fei;YAN Peng;HE Yu;WANG Lei(School of Civil Engineering,Changsha University of Science&Technology,Changsha 410114,Hunan,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2024年第11期64-75,共12页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2021YFB2600900)
国家自然科学基金项目(52178107)
湖南省杰出青年科学基金项目(2024JJ2003)。
关键词
桥梁工程
吊索
马尔可夫决策过程
维护决策优化
深度强化学习
bridge engineering
suspender
Markov decision process
maintenance policy optimization
deep reinforcement learning
