Research On Project Level Decision-making Of Highway Asphalt Pavement Maintenance Based On Reinforcement Learning

Under the influence of traffic load and environment,the performance of asphalt pavement deteriorated every year which leads to the more and more severe maintenance task.The lack of accurate prediction of pavement performance and the unreasonable application of maintenance measures will result in a large amount of waste of maintenance funds.Based on the highways in Jiangsu Province,this paper systematically studies the project-level decision-making problem of highway asphalt pavement based on the research results of asphalt pavement maintenance management and decision-making support system(PMS).Firstly,the current condition of highway asphalt pavement in Jiangsu Province was evaluated,including the scale of highway network,road ages,traffic volumes,pavement structures,pavement performance and pavement structure conditions.Then the indicators of performance prediction and maintenance decision-making were determined.Secondly,a data quality control method was proposed.According to the different characteristics of rutting,roughness,skid-resistance and pavement distress,two different strategies,the longest subsequence method and the direct deletion method,were established respectively.The pavement detection data Jiangsu Province in 2018 was evaluated and the resurveyed data were compared.Then,five different pavement prediction models were established by using neural network method to predict rutting,roughness,skid-resistance,transverse cracking and pavement distress respectively.Various influential factors were considered.The mean influence value algorithm was used to evaluate the sensitivity of each input variable.The accuracy of models was verified.Finally,the reinforcement learning method is introduced into the maintenance decision-making to maximize the long-term cost-effectiveness of the full width of the pavement.Under the principle of minimizing human intervention,the method of finding out the best maintenance strategy is proposed by defining a reasonable reward function.The implementation process and decision-making results were introduced by taking Ningchang and Zhenli Expressway as an example.