Research Information System
2nd International Conference on Applied Engineering and Natural Sciences, Konya, Turkey, 10 - 13 March 2022, pp.1-3, (Summary Text)
The Importance and Process of Calibration in Microscopic Traffic Simulation
Abdullah Erdem Boztaş* and Halit Özen2
1Faculty of Transportation and Logistics, Istanbul University, Turkey
2Faculty of Civil Engineering, Yıldız Technical University, Turkey
*ae.boztas@istanbul.edu.tr Email of the corresponding author
Abstract – A system can be defined as a complex unity formed of many often-diverse parts subject to a common plan or serving a common purpose. Transportation is by this definition a system that has a major impact on the daily life of the community. Hence the importance of and need for transportation planning. The prior reflections of transportation planning are on city planning, enabling economic activities, promoting community interaction, and enhancing quality of life by improving air quality and minimizing the impact on climate change. Considering the scale of transportation systems varying from basic segments of highway network to complex integrated public transit systems, it is not feasible to conduct tests nor to improve infrastructure without certainty on site. Therefore, the necessity of identifying problems and issues in the current transportation system and finding solutions and considering potential changes that can bring improvement in the current plan requires a different approach. This physical impossibility of testing a new and improved transportation system in the field can be overcome with traffic simulation.
Traffic simulation is the mathematical and visual modeling of transportation systems with input data of network attributes, demand, signal timing, public transit systems, etc. In order to evaluate different solutions and improve the current system, the model must first simulate real life conditions (travel times and traffic volume) within acceptable limits. It is expected for the model outputs and observed data to vary as a result of stochastic character of traffic. At this stage what needs to be done to establish the desired reliability level is calibration and validation of the model. Calibration is the process of fine tuning the model parameters to reflect the real life conditions as approximate as possible. These model parameters are variables to represent the driving behavior in two major approaches: car following and lane change. In PTV Vissim which is the microscopic simulation software used in the study; Wiedemann 99 is used as the car following model. Wiedemann 99 is a traffic flow model developed by R. Wiedemann (1999) that imitates the physical and psychological behaviors of drivers. Since Wiedemann 99 has the greater number of assessable model parameters among the car following model options, the possibility for tuning the model to match observed driving behavior is higher. The advantageous number of car following model parameters is 10 for Wiedemann 99 model and these are as listed below:
CC0: Standstill distance
CC1: Gap time distribution
CC2: Following distance oscillation
CC3: Threshold for entering following
CC4: Negative speed difference
CC5: Positive speed difference
CC6: Distance dependency of oscillation
CC7: Oscillation acceleration
CC8: Acceleration from standstill
CC9: Acceleration at 80 km/h
Since these parameters are developed to simulate a generalized driving behavior pattern, the default values often return unacceptable distant traffic values from the observed data. This is caused by the differentiation of the driver characteristics between the study’s model and Wiedemann 99’s model. The calibration which is the answer to this problem is done by systematically changing the values and comparing the results of traffic volume and average travel time outputs with each other and the observed data. This procedure takes 40-60 simulation runs for each of the parameters separately to reach the optimum value (the acceptable value that returns the closest outputs to real life).
As explained before, the calibrated values of the parameters considered to be validated if they return acceptable values of both traffic volume and travel time results. And this results in two optimum values of the parameter instead of one. In order to overcome this issue, another evaluation parameter (P) is chosen. Hence the default car following model parameters results in lower traffic volume and higher average travel time than observed data, P is the chosen as the division of total number of vehicles by average travel time. ( P=(total number of vehicles)/(average travel time) )
When the model of this study is simulated using the default Wiedemann 99 car following model, the evaluation parameter P is returned as 3,744 (1490/397,947). The calibration of the parameters (CC0-CC9) has the potential to optimize the model by significant increases in P. As seen in the graphic below, we have managed to achieve a convincing 44% increase in P by the calibration of the parameter CC2: Following distance oscillation.
In conclusion, it is shown that traffic simulation is crucial to transportation planning and the most significant part of traffic simulation is the calibration process where the developed model gets validated to be able to reflect the observed data and to evaluate potential changes to the current system. In this study, steps of calibration process is discussed and with a combination of optimum values of each parameter the real life conditions were simulated within an acceptable range.
Keywords – Transportation planning, traffic engineering, microscopic traffic simulation, PTV Vissim, car following model, calibration and validation