Capturing uncertainty intuition in road maintenance decision-making using an evidential neural network

Abstract

Decision-making of project-level road maintenance is the process of mapping road information into a maintenance plan. Even though benefitting from deep learning, the decision-making still faces the problem of maintenance data uncertainty. The data uncertainty derives from imperfect road information collection and arbitrary selection of maintenance plans. Such uncertainty always leads to unreasonable maintenance decision-making. This study proposes an evidential approach using information entropy (IE) and Dempster–Shafer theory (DST) to capture and handle uncertainty in the decision-making of project-level road maintenance. The approach first uses an IE-based judgment method (IE-based method) to capture and observe quantitative data uncertainty. The DST-based method is then developed to handle maintenance data uncertainty through utilizing evidential neural network and set-valued decision-making. A numerical experiment is performed on the maintenance data with 280 km of semi-rigid base highways in China. The results indicate that the IE-based method can measure the data uncertainty in the information of road sections. The DST-based method captures the cautious intuition on the selection of maintenance plans, thereby reducing the decision error rate by over 14% under specific conditions when facing data uncertainty.