Date of Award
Doctor of Philosophy in Transportation - (Ph.D.)
Civil and Environmental Engineering
I-Jy Steven Chien
Janice Rhoda Daniel
Lane closures due to road reconstruction and maintenance have resulted in a major source of non-recurring congestion on freeways. It is extremely important to accurately quantify the associated mobility impact so that a cost-effective work zone schedule and an efficient traffic management plan can be developed. Therefore, the development of a sound model for predicting delays or road users is desirable.
A comprehensive literature review on existing work zone delay prediction models (i.e., deterministic queuing model and shock wave model) is conducted in this study, which explores the advantages, disadvantages, and limitations of different modeling approaches. The performance of those models seems restricted to predict congestion impact under space-varying (i.e., road geometry, number of lanes, lane width, etc.) and time-varying (i.e., traffic volume) conditions. To advance the delay prediction accuracy, a multivariate non-linear regression (MNR) model is developed first by incorporating big data to capture the relationship of speed versus the ratio of approaching traffic volume to work zone capacity for work zone delay prediction. The MNR model demonstrates itself able to predict spatio-temporal delays with reasonable accuracy.
A more advanced model called ANN-SVM is developed later to further improve the prediction accuracy, which integrates a support vector machine (SVM) model and an artificial neural network (ANN) model. The SVM model is responsible to predict work zone capacity, and the ANN model is responsible to predict delays. The ultimate goal of ANN-SVM aims to predict spatio-temporal delays caused by a work zone on freeways in the statewide of New Jersey subject to road geometry, number of lane closure, and work zone duration in different times of a day and days of a week. There are 274 work zones with complete information for the proposed model development, which are identified by mapping data from different sources, including OpenReach, Plan4Safety, New Jersey Straight Line Diagram (NJSLD), New Jersey Congestion Management System (NJCMS), and INRIX. Big data analytics is used to examining this massive data for developing the proposed model in a reliable and efficient way.
A comparative analysis is conducted by comparing the ANN-SVM results with those produced by MNR, RUCM (NJDOT Road User Cost Manual approach), and ANN-HCM (the ANN model with work zone capacity suggested by Highway Capacity Manual). It is found that ANN-SVM in general outperforms other models in terms of prediction accuracy and reliability. To demonstrate the applicability of the proposed model, an analysis tool, which adapts to ANN-SVM, is developed to produce graphical information. It is worth noting that the analysis tool is very user friendly and can be easily applied to assess the impact of any work zones on New Jersey freeways. This tool can assist transportation agencies visualize bottlenecks and congestion hot spots caused by a work zone, effectively quantify and assess the associated impact, and make suitable decisions (i.e., determining the best starting time of a work zone to minimize delays to the road users). Furthermore, ANN-SVM can be applied to develop, evaluate, and improve traffic management and congestion mitigation plans and to calculate contractor penalty based on cost overruns as well as incentive reward schedule in case of early work competition.
Du, Bo, "An artificial neural network model for predicting freeway work zone delays with big data" (2017). Dissertations. 2.