978-1-5386-9484-8/19/$31.00 ©2019 IEEE Improving the calibration time of traffic simulation models using parallel computing technique Nima Dadashzadeh Graduate school of Science Engineering and Techniology, Istanbul Technical University, Istanbul, Turkey Traffic Technical Institute, University of Ljubljana, Ljubljana, Slovenia dadashzadeh@itu.edu.tr Marijan Žura Traffic Technical Institute, Civil and Geodetic Engineering Faculty University of Ljubljana Ljubljana, Slovenia marijan.zura@fgg.uni-lj.si Murat Ergun Transportation Engineering Division, Civil Engineering Faculty Istanbul Technical University Istanbul, Turkey ergunmur@itu.edu.tr Ali Sercan Kesten Civil Engineering Division, Department of Engineering Işık University Istanbul, Turkey sercan.kesten@isikuni.edu.tr Abstract—The calibration procedure for traffic simulation models can be a very time-consuming process in the case of a large-scale and complex network. In the application of Evolutionary Algorithms (EA) such as Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) for calibration of traffic simulation models, objective function evaluation is the most time-consuming step in such calibration problems, because EA has to run a traffic simulation and calculate its corresponding objective function value once for each set of parameters. The main contribution of this study has been to develop a quick calibration procedure for the parameters of driving behavior models using EA and parallel computing techniques (PCTs). The proposed method was coded and implemented in a microscopic traffic simulation software. Two scenarios with/without PCT were analyzed using the developed methodology. The results of scenario analysis show that using an integrated calibration and PCT can reduce the total computational time of the optimization process significantly - in our experiments by 50% - and improve the optimization algorithm’s performance in a complex optimization problem. The proposed method is useful for overcoming the limitation of computational time of the existing calibration methods and can be applied to various EAs and traffic simulation software. Keywords—Calibration, Parallel Computing, Genetic Algorithm, Particle Swarm Optimization, Parallel Hybrid GAPSO, Parallel Hybrid PSOGA, VISSIM I. INTRODUCTION Parallel computing technique (PCT) and known multithreading technique have been introduced in the field of modeling and calibration when most of the modelers were suffering from weak performance of sequential (serial) computing methods in terms of its long computational time. The basic idea behind this technique is to divide a large problem into smaller tasks solved simultaneously on multiple processors in a process called parallel execution or parallelization. There are two different kinds of PCT:  Sharing computation work among available cores of one computer, called PCT in this study.  Distributing computational work among a cluster of several computers. Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) have been used mostly because of their ease of implementation, especially in the calibration procedure of traffic simulation models. Microscopic and macroscopic traffic simulation models can be considered large-scale and complex networks. When applying EA to such calibration problem, EA has to run a simulation and calculate its corresponding objective function value once for each set of parameters, causing a time-consuming step. Sequential use of GA [1]–[6], PSO [7]–[10], and hybrid GA and PSO (GAPSO) [11] have been studied thoroughly; however, in the existing literature there is no study regarding implementation of PCT on hybrid GAPSO and hybrid PSO and GA (PSOGA) models. The limitations of using single GA and single PSO algorithms is explained and solved in authors’ previous work by proposing a combination of the GA and PSO named hybrid GA and hybrid PSO [11]. Although it showed successful computational performance, one might consider using parallel computing PCT to decrease the computation time of the proposed calibration procedure. The main contribution of this study has been to develop a quick calibration procedure for the selected parameters using EA and PCT. In this study, we used VISSIM [12] as traffic microsimulation software, and GA, PSO, and hybrid GAPSO as optimization algorithms. The proposed method was programmed using MATLAB and implemented via the VISSIM - MATLAB COM interface on selected case study. Further parts of article are structured as follows: Section 2 presents the methodology of research and detailed information of coding of optimization algorithms in both serial and parallel mode. Section 3 defines the case study area and relevant observed data collection procedures. Results of different scenarios (with/without PCT) of proposed methodology on case study area are discussed in detail in section 4. Lastly, section 5 concludes the research findings and contribution. II. METHODOLOGY Selection of appropriate objective function and stopping criteria are two common steps of all optimization algorithms. There are many single and multi-objective functions used to minimize the error of simulated and observed data. The Mean Absolute Normalized Error (MANE) [13]–[16] falls among several multi-objective functions used in previous studies for the calibration of VISSIM simulation model parameters and is widely used around the world. The developed code, here, can perform the optimization process based on both single (e.g. speed-only, volume-only, and occupancy rate) and multi-