Machine learning-based stocks and flows modeling of road infrastructure

Abstract

This paper introduces a new method to account for the stocks and flows of road infrastructure at the national level based on material flow accounting (MFA). The proposed method closes some of the current shortcomings in road infrastructures that were identified through MFA: (1) the insufficient implementation of prospective analysis, (2) heavy use of archetypes as a way to represent road infrastructure, (3) inadequate attention to the inclusion of dissipative flows, and (4) limited coverage of the uncertainties. The proposed dynamic bottom-up MFA method was tested on the Norwegian road network to estimate and predict the material stocks and flows between 1980 and 2050. Here, a supervised machine learning model was introduced to estimate the road infrastructure instead of archetypical mapping of different roads. The dissipation of materials from the road infrastructure based on tire–pavement interaction was incorporated. Moreover, this study utilizes iterative classified and regression trees, lifetime distributions, randomized material intensities, and sensitivity analyses to quantify the uncertainties.