Uncertainty Propagation from the Cell Transmission Traffic Flow Model to Emission Predictions: A Data-Driven Approach

Abstract

Road traffic exhaust emission predictions are used to inform transport policy and investment decisions aimed at reducing emissions and achieving sustainable mobility. Emission predictions are also used as inputs when modelling air quality and human exposure to traffic-related air pollutants.To be effective, such policies and/or integration must be based on robust models that not only provide point-based predictions, but also inform these with an interval of confidence that properly accounts for the propagation of uncertainties through the complex chain of models involved. This paper develops a data-driven methodological framework which enables calculating the uncertainty in average speed-based emission predictions induced by uncertainty in its traffic data inputs which are most often predictions (or outputs) of traffic flow models. An ensemble-based optimisation approach is used to estimate both calibration and validation errors arising from uncertainty in the structure and parameterisation of the Cell Transmission Model (CTM); a discretised first-order macroscopic traffic flow model that is often integrated with average speed-based emission models. A Monte Carlo sampling approach is proposed to propagate the uncertainty in traffic flow inputs to emission predictions. To ensure transferability of findings, this methodology has been tested using multiple real data sets on three motorway road networks, one of which operates under Variable Speed Limits (VSL).

Microscopic Traffic Modeling Inside Intersections: Interactions Between Drivers

“…The traditional procedure to validate a traffic flow model contains two steps: first, the model parameters are calibrated using the collected field data (i.e., find the best fitting parameters), and then the quality of the model with these parameters is tested on another data set, which is the validation step (Treiber and Kesting 2012, Daamen, Buisson, and Hoogendoorn 2014, Spiliopoulou et al 2014, Treiber and Kesting 2014, Punzo and Montanino 2016, Mariotte et al 2020. In calibration and validation, the macroscopic performance indicators are commonly used to measure the overall performance of the system, such as the traffic flow (Xie, Nie, andLiu 2017, Tang et al 2020), flow rate (Jiang et al 2017, Yuan, Knoop, and, speed (Chiu, Zhou, andSong 2010, Tian et al 2017), density (Ma, Dong, andZhang 2007, Sayegh, Connors, andTate 2018), queue length (Zhang et al 2020), travel times (Hollander and Liu 2008), and their combinations (Kim and Mahmassani 2011, Ni et al 2016, Han et al 2017, Kontorinaki et al 2017. One may notice that, although calibration of the model can involve finding individual vehicle parameters (Treiber and Kesting 2013), in validation often aggregate variables are used (Treiber and Kesting 2012).…”