Evaluating the Resilience of Two-dimensional Genetic Algorithm-based Centralized Traffic Signal Control versus Maxpressure-based Decentralized Control When Operating with Input Error.

抄録

Investigations on impact of errors in traffic input data on the resilience of real time traffic signal control systems are scarce. The present paper investigates the impact of underestimating vehicular volumes and queue length on performance of a real time decentralized as well as a centralized traffic signal control system. The decentralized system applies Maxpressure algorithm to determine control plans. The centralized system relies on a link model to predict traffic flow, integrates a rolling horizon framework, and uses a two-dimensional genetic algorithm to obtain optimal phase plans, green times, and offsets. Findings from the study highlight advantages of using centralized over decentralized control, integration of predictive model and rolling horizon framework while designing optimal controls, as well as the necessity of early interventions to avoid spillback conditions. These observations are beneficial to traffic engineers, urban planners and researchers invested in designing and implementing real time traffic signal control systems.