Research On State Prediction And Maintenance Decision Optimization Technique Of Subway Track Irregularity Based On Machine Learning

In recent years,with the acceleration of urbanization,the subway has become one of the main modes of transportation for citizens.As the basis for the operation of subway trains,the track irregularity seriously affects the safety of train operation and the comfort of passengers.It is of great significance to study subway track irregularity state prediction and maintenance decision-making optimization technology to scientifically and reasonably grasp the track irregularity state and its changing law,optimize the preparation of maintenance plans,and realize dual control of the safety and maintenance cost of rail line equipment.To realize “condition-based maintenance” and “preventive maintenance” of track irregularities in subway,this paper studies the two aspects of the state prediction of track irregularity and the optimization of maintenance decision.Besides,this paper designs and develops corresponding software based on the above research results.The main contents of the dissertation are as follows.(1)A machine learning-based subway track irregularity state prediction model is proposed.In this paper,the Stacking ensemble learning method is used to construct an ensemble model of track irregularity prediction based on Stacking.The model takes the overall irregularity degradation trend of the 200 m track unit of subway line as the research object,and uses the multiple historical track irregularity detection data of subway lines to make a personalized prediction of the track irregularity state.The validity of the model was verified by the inspection data,maintenance data and related line equipment account data of the 174 rail inspection cars of Beijing Metro Line 1 and Line 2 from 2013 to 2020.The results show that the model can accurately predict the state of track irregularity.(2)A machine learning-based subway track irregularity maintenance decision optimization model is proposed.In this paper,200 m track unit of subway line is used as the research object,and Markov reinforcement learning technology is used to construct an optimization model for track irregularity maintenance decision based on Adaptive Learning Markov Decision Process(AL-MDP).Taking the minimum total cost in the planning period as the optimization goal,and taking the state of the 200 m track unit section as the constraint,the optimal maintenance decision for the track irregularity of the track unit for the long period is formulated.An adaptive learning mechanism based on the stacking-based track irregularity state prediction ensemble model is set in the model to provide the track irregularity state prediction results,so that it can adaptively learn the track irregularity state degradation process to optimize the maintenance decision of track irregularity delicately and personally.Taking the 200 m track unit with mileage of B102～B104 on Beijing Metro Line 1 as an example,a case study is carried out.The results show that compared with the traditional MDP model that lacks an adaptive learning mechanism,the model can better save track irregularity maintenance costs and improve the quality of maintenance decisions.(3)A sub-system for prediction and maintenance decision-making optimization of Beijing subway track irregularity is designed and developed.Based on the above research results,the author participated in the design and development of the relevant functional modules of the track irregularity prediction and maintenance decision optimization subsystem in the scientific research project “Beijing Subway Rail Smart Operation and Maintenance Management Information System”,and used actual production data of Beijing subway to display the application.