Regulating Traffic Flow and Speed on Large Networks: Control and Geographical Self Organizing Map (Geo-SOM) Clustering

TR Number

Date

2021-06-09

Journal Title

Journal ISSN

Volume Title

Publisher

Virginia Tech

Abstract

Traffic growth and limited roadway capacity decrease traveler mobility and increase traffic congestion and fuel consumption. Traffic managers employ various control techniques to mitigate the aforementioned problems. One well-known network-wide control strategy is perimeter control (or gating). Perimeter control is based on the Network Fundamental Diagram (NFD). NFD-based perimeter control techniques are used to solve congestion problems in transportation networks. One well-known method used in the literature is Proportional Integral Control (PIC). PIC solves the congestion problem, but suffers from sensitivity to parameter tuning and the need for model linearization. A weather-tuned perimeter control (WTPC) and a jam density-tuned perimeter controller (JTPC) were developed to cope with parameter sensitivity for different weather conditions and jam densities, respectively. In an attempt to overcome PIC problems, a sliding mode controller (SMC) was developed. SMC does not require model linearization and parameter tuning. It is also robust to varying demand patterns. SMC computes the flow that needs to enter a protected network and converts it to corresponding traffic signal timings to achieve the desired control strategies. Another approach to implementing the sliding mode controller is to control vehicle speeds on the links entering the protected network. Coupling speed harmonization (SH) with sliding mode control (SMC), an SMC-SH was developed and implemented in the INTEGRATION microscopic traffic simulator. The mentioned controllers are all tested on a mid-size grid network replicating downtown Washington DC. SMC-SH improved different performance metrics on the whole grid network compared to the no control case. Specifically, it improved average travel time, total delay, stopped delay, fuel consumption, CO2 emissions by 17.27%, 18.18%, 12.76%, 5.91%, and 7.04%, respectively. In order to test the SMC-SH on a real large-scale network, the downtown Los Angeles (LA) network is used. The LA network is known for its congested freeways, so a development of a Freeway-SMC-SH controller is performed and tested. It shows good results in improving the performance not only of freeways, but also the overall LA network performance. Particularly, the network-wide average travel time, total delay, stopped delay, fuel consumption and CO2 emissions improved with respect to the no control case by 12.17%, 20.67%, 39.58%, 2.6%, and 3.3%, respectively. An identification of a homogeneously congested area is needed to apply SMC-SH on LA roads (not freeways). The geographical self organizing maps (GeoSOM) clustering algorithm is applied and tested on the LA network. The clustering goal is to identify a geographically connected region with small density variance. GeoSOM is able to achieve that objective with better performance than the state-of-the-art Kmeans and DBSCAN clustering algorithms. The enhancements reached up to 15.15% for quantization error, 61.05% for spacial quantization error, and 43.96% for variance. Finally, the SMC-SH is tested on the protected region of the LA network identified by the GeoSOM algorithm. SMC-SH succeeds in improving network-wide vehicle travel time, total delay, stopped delay, fuel consumption and CO2 emissions by 6.25%, 9.4%, 16.47%, 1.7%, and 2.19%, respectively.

Description

Keywords

Traffic flow control, speed harmonization, network fundamental diagram, macroscopic fundamental diagram, GeoSOM clustering

Citation