Network-wide signal timing stochastic simulation optimization with environmental concerns

• Published in: Applied soft computing Vol. 77; pp. 678 - 687
• Main Authors: Zheng, Liang, Xu, Chengcheng, Jin, Peter J., Ran, Bin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.04.2019
• Subjects: Signal timing optimization; Simulation-based optimization; Total delay; Vehicular emissions; Vehicular emissions; Signal timing optimization; Total delay; Simulation-based optimization
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2019.01.046

This study addresses a network-wide signal timing optimization problem with environmental concerns by a bi-objective stochastic simulation-based optimization (BOSSO) method. In this method, the global samples evaluated by costly simulation are used to build a type of surrogate model named the regressing Kriging model, which are then employed to predict bi-objectives of untested samples or filter noises from the evaluated samples. An adaptive selector is incorporated to determine which samples in the local trust-region are evaluated by costly simulation and which ones by the built regressing Kriging model. This helps to balance computational costs and accuracies of three quadratic regression models, especially when the variable dimension is high. The non-interactive role of a decision maker is taken to generate more non-dominated solutions around the desired bi-objective point. In the field experiments, an urban road network with 15 signalized and five non-signalized intersections in Changsha, China, is modeled as the simulation scenario by VISSIM. Then, the traffic simulation model is firstly calibrated from two aspects by the BOSSO method, which can well reproduce the reality. After that, the network-wide bi-objective signal timing optimization problem is also solved by the BOSSO method. Numerical results show that compared with the real-field traffic states, the total delay and vehicular emissions are reduced by at most 16.90% and 32.22% respectively under the budged number of simulations. Balance analyses also show the existence of a competing relationship between bi-objectives. Finally, the BOSSO method is validated to outperform three other counterparts (NSGA-II, BOTR and BOEGO) from various aspects. •Bi-objective stochastic simulation-based optimization (BOSSO) method is introduced.•Bi-objective calibration and signal timing problems under uncertainties are well solved.•BOSSO outperforms three other counterparts from various aspects.