Research on pavement maintenance planning using reinforcement learning release_gphrosiozbgbfkx6eq6bx5db54

Abstract

<jats:title>Abstract</jats:title> A mature theoretical model is an integral component of pavement management systems. Typically, such models encompass pavement performance prediction, cost–benefit analysis of maintenance measures, and decision-making frameworks for maintenance strategies. This paper primarily investigates the economic analysis and decision planning aspects of pavement maintenance measures. Initially, considering greenhouse gas emissions, a comprehensive cost evaluation is conducted for three typical maintenance measures (microsurfacing, slurry sealing, and milling with overlay), encompassing carbon emission costs, delay costs, and construction costs, to establish comprehensive cost indicators for subsequent decision optimization models. Subsequently, an optimization algorithm with Clip pruning is introduced for near-end strategy optimization, along with the development of suitable evaluation functions for maintenance measure effectiveness, thereby establishing a reinforcement learning-based optimization model for pavement maintenance decisions. Regarding the model results, the costs of microsurfacing, 4 cm AC slurry seal, 4 cm SMA slurry seal, and 4 cm milling with overlay are 10 ¥(m2), 50 ¥(m2), 66 ¥(m2), and 60 ¥(m2), respectively. Furthermore, by comparing the reinforcement learning decision model developed in this study with the Q-learning algorithm, the model demonstrates a 60% improvement in convergence speed and a 20% increase in average reward.
In application/xml+jats format

Archived Content

There are no accessible files associated with this release. You could check other releases for this work for an accessible version.