A Smart Traffic Incident Management (TIM) System for Estimating Highway Incident Duration and Impacts with and without Surveillance Sensors

Abstract

Highway incidents are major contributors to traffic congestion, causing significant delays for daily roadway users and reducing the reliability and productivity of transportation systems. To mitigate the negative impacts of these incidents and quickly restore highway operations, it is crucial for highway agencies to implement an efficient incident management system. However, providing the public with real-time information about the impacts of incidents at a desired level of precision is challenging due to the complexities involved in obtaining sufficient data and understanding the intricate relationships among the factors that influence these impacts. To address this challenge, this study proposes a Smart Traffic Incident Management (TIM) system that delivers robust and reliable information on estimated clearance durations, resultant queue lengths, time-varying traffic information, and traffic detouring volumes from freeways to adjacent arterials. This initiative aims to improve the effectiveness of incident response, thereby enhancing the resilience and functionality of the transportation network.

The proposed system consists of four primary modules. Module 1 aims to robustly predict incident clearance duration through the proposed Knowledge Transferability Analysis (KTA) model, featuring its automated process for assessing, selecting, and transferring existing prediction rules from pre-existing Incident Duration Prediction Models (IDPM). This strategic utilization obviates the necessity for integrating field operators' expertise in formulating prediction rules, thereby alleviating the dependency on an ample volume of incident records for prediction rules calibration. The evaluation results, using I-70 in Maryland for the case study, have demonstrated the effectiveness of the proposed KTA model in not only expediting the development process of an IDPM but also improving the resulting accuracy of the prediction rules.

Module 2 endeavors to robustly predict incident queue length by introducing the Real-time Incident Queue Prediction (R-IQP) system. This system's principal model enhances the formulations for queue propagation dynamics by incorporating the influences of incoming drivers' perceptions and responses to progressively constrained traffic conditions. Additionally, two supplementary models are proposed to precisely estimate flow rates, leveraging probe speed information, to accommodate different surveillance environments characterized by varying levels of data availability. The evaluations of the proposed R-IQP system with both the field-collected data and the well-calibrated simulator’s data have proven the capability of the R-IQP system on predicting time-varying queue lengths for incidents with various clearance durations and types of lane blockage statuses.

Module 3 introduces a traffic flow model specifically designed for traffic incident management. The proposed Incident-oriented METANET (I-METANET) enhances the widely used METANET model with three key improvements: 1) reflecting the merging behaviors incurred by incidents and their significant yet diminishing effects on speed propagation over upstream segments; 2) incorporating the simultaneous effects of upstream traffic flows and downstream incident-induced queue waves on the speed of a subject segment; and 3) integrating the combined effects of ramp-flow weaving maneuvers and the presence of incident queues on traffic conditions at interchange segments. The proposed I-METANET model, calibrated and validated using field data from I-4 in Florida, has demonstrated its effectiveness in predicting time-varying speeds and flow rates over roadway segments during incident clearance periods.

Module 4 focuses on assessing the impact of freeway incidents on nearby local roads. To achieve this, the study developed a Real-time Detour Volume Estimation (R-DVE) system, designed to estimate the volume of traffic diverted from the freeway mainline to its adjacent arterials, even when freeway traffic sensors are unavailable. This system leverages a set of offline speed-flow models developed in Module 2 to estimate traffic flow using probe speed data as input. Additionally, the proposed R-DVE incorporates a Quality Assessment Mechanism (QAM) that integrates a robust customized dynamic speed-flow relations (CDSFR) developed in Module 3 to continually examine the applicability of the estimated flow rates and update the offline speed-flow models. The performance evaluation, based on real-world incident cases on I-95 in Maryland, demonstrated that the R-DVE system can accurately estimate real-time detouring volumes, highlighting its practical applicability.

The proposed Smart Traffic Incident Management (TIM) system, delineated through its comprehensive modules, embodies several key features aimed at enhancing incident management efficacy, including 1) providing a systematic decision-making framework for incident clearance duration prediction, particularly valuable for highways lacking sufficient incident records to calibrate prediction rules; 2) incorporating predicted clearance duration to generate timely estimates of incident queue length, with the adaptive capability particularly beneficial for highways under varying levels of traffic sensor availability; 3) predicting time-varying traffic information to faciltiate better incident management and responsive strategies; 4) generating real-time estimates of detour volume originating from each interchange within the impact area, facilitating the execution of appropriate responsive operations contributing to efficient incident management; and 5) exhibiting a dynamic nature by updating estimated information when additional data become available or when there are changes in traffic dynamics or incident clearance operation to ensure the continuous relevance and accuracy of the provided information.