Date on Master's Thesis/Doctoral Dissertation
Civil and Environmental Engineering
Civil Engineering, PhD
Li, Dr. Zhixia (Richard)
Committee Co-Chair (if applicable)
Connected vehicles; signalized intersections; queue length estimation; traffic volume estimation; shockwave analysis; deep learning
The development of Connected Vehicles (CV) facilitates the use of trajectory data to estimate queue length and traffic volume at signalized intersections. The previously proposed models involved additional information that may require conducting different types of data collection. Also, most models need higher market penetration rate or more than a vehicle per cycle to provide adequate estimation. This is mainly because a significant number of the estimation models utilized only queued vehicles. However, the state of motion in non-queued vehicles, particularly slowed-down vehicles, provides much information about the traffic characteristics. There is a lack of exploiting the information from slowed-down vehicles in identifying the last queued vehicle to improve the estimation models. The importance of this work is to propose a cycle-by-cycle queue length and traffic volume estimation models by utilizing the slowed-down vehicles. It proposes a sophisticated model to estimate the queue length and traffic volume from connected vehicles with low market penetration rate (MPR) by utilizing shockwave theory and deep learning technique (artificial neural network). The work starts with establishing a relationship between the slowed-down vehicles and last queued vehicles based on shockwave theory and traffic flow dynamics. Then, the queue estimation algorithm is modeled based on the capacity state and deep learning technique. The traffic volume algorithm modeled is based on the queue length information. Four experiments were conducted to investigate the performance of the queue length and traffic volume estimation models on dataset from simulation environment and real-world data. The queue length results of the simulation experiment demonstrated an adequate mean absolute percentage error (MAPE) of 13.44%, which means an accuracy of 86.56%. The highest MAPE was 19.16% (80.84% accuracy) and the lowest was 7.36% (92.64%). The results of the queue length algorithm applied on real-world data demonstrated an MAPE of 21.97% (78.03% accuracy). The performance of the traffic volume algorithm on simulation data demonstrated an excellent MAPE of 11.8% (88.2% accuracy). The performance of the algorithm based on real-world data from demonstrated an MAPE of 23.57% (76.43% accuracy). Although the previous models can provide similar accuracy rates, they require higher MPR. With the low MPR of 10%, the proposed models revealed an adequate estimation accuracy of the cycle-by-cycle queue length and traffic volume.
Algomaiah, Abdulmaged, "Utilizing shockwave theory and deep learning to estimate intersection traffic flow and queue length based on connected vehicle data." (2022). Electronic Theses and Dissertations. Paper 3813.