Development of Distress Index Prediction Models for Rehabilitation Treatments in Louisiana Using Advanced Machine Learning Techniques

Performance prediction models are used by state agencies to predict future trends in distress indices, hence, determining the required maintenance and/or rehabilitation treatment as well as the deterioration rate and remaining pavement service life. However, most of these models are based on a limited number of parameters and cannot predict the performance distress indices reliably. Such limitation resulted in having, most of the time, a maximum prediction period of five years. As a solution and coping with the ever-increasing size of pavement data, machine learning techniques have become a promising alternative. The objective of this study was to develop a machine-learning-based framework for states with a hot and humid climate that can predict the long-term field performance (for 11 years) of their asphalt (AC) overlays based on their key project conditions. Two machine learning algorithms were examined, namely Random Forest (RF) and CatBoost, and the one yielding a higher accuracy was considered. In this study, the well-known pavement condition index (PCI) was used as the pavement performance indicator. A total of 892 log miles of AC overlay data were obtained from the Louisiana Department of Transportation and Development (LaDOTD) Pavement Management System (PMS) database. Based on the collected data, six models were trained (for each algorithm) and validated to predict the future PCI of AC overlays for up to 11 years. Results indicated that the RF algorithm yielded higher accuracy than the CatBoost Algorithm and thus the RF-based models were considered in the proposed decision-making framework.各州交通运输主管部门常采用性能预测模型，预判路面破损指标的未来变化趋势，以此确定所需的养护与修复处置方案，同时推算路面劣化速率及剩余服役寿命。然而，现有多数模型仅基于有限参数构建，无法可靠预测路面性能破损指标；受此局限，这类模型的最大预测时长通常仅为5年。为解决上述问题并应对持续增长的路面数据体量，机器学习技术已成为颇具前景的替代方案。本研究旨在为湿热气候地区的州级主管部门开发一套基于机器学习的分析框架，该框架可依托核心工程条件，预测沥青混凝土（asphalt concrete，简称AC）罩面长达11年的长期现场性能表现。研究选取两种机器学习算法开展测试，即随机森林（Random Forest，简称RF）与CatBoost，并选用其中精度更高的算法进行后续研究。本研究采用通用的路面状况指数（Pavement Condition Index，简称PCI）作为路面性能评价指标。研究数据取自路易斯安那州交通运输与发展部（Louisiana Department of Transportation and Development，LaDOTD）路面管理系统（Pavement Management System，PMS）数据库，共包含892车道英里的沥青混凝土罩面相关数据。基于采集到的数据集，本研究针对两种算法分别训练并验证了6个模型，用于预测沥青混凝土罩面最长达11年的未来路面状况指数。实验结果表明，随机森林算法的预测精度优于CatBoost算法，因此本研究提出的决策框架将采用基于随机森林的模型。