Documents

2013 Aghabayk Sarvi Young Kautzsch

Categories
Published
of 15
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Description
Australasian Transport Research Forum 2013 Proceedings 2 - 4 October 2013, Brisbane, Australia Publication website: http://www.patrec.org/atrf.aspx 1 A NOVEL METHODOLOGY FOR EVOLUTIONARY CALIBRATION OF VISSIM BY MULTI-THREADING Kayvan Aghabayk 1 *, Majid Sarvi 1 , William Young 1 & Lukas Kautzsch 2 1 Institute of Transport Studies, Department of Civil Engineering, Monash University, Australia 2 PTV Planung Transport Verkehr AG, Karlsruhe, Germany * Corresponding Author: Kayva
Transcript
   Australasian Transport Research Forum 2013 Proceedings 2 - 4 October 2013, Brisbane, Australia Publication website: http://www.patrec.org/atrf.aspx  1 A NOVEL METHODOLOGY FOR EVOLUTIONARY CALIBRATION OF VISSIM BY MULTI-THREADING Kayvan Aghabayk 1 *, Majid Sarvi 1 , William Young 1 & Lukas Kautzsch 2 1  Institute of Transport Studies, Department of Civil Engineering, Monash University, Australia 2  PTV Planung Transport Verkehr AG, Karlsruhe, Germany * Corresponding Author: Kayvan.Aghabayk@monash.edu Abstract   Traffic micro-simulation models have become very popular in transport studies and are extensively used in research and by industry. Traffic simulation models are especially very useful in reflecting the dynamic nature of transportation system in a stochastic manner, which is beyond the capability of some classic methods. Nevertheless, one of the major concerns of micro-simulation users has been the appropriate calibration of this software. An inappropriate calibration may end in a wrong conclusion and decision which could result in irreparable costs or problems. This paper develops an efficient methodology to improve the calibration procedure of traffic micro-simulation models. It applies the method to VISSIM. It provides a methodology for auto-tuning of VISSIM as one of the well-known and commercially available traffic micro-simulations. More specifically, it looks at the car-following and lane changing models as they form the main component of any traffic micro-simulation. This approach uses particle swarm optimisation (PSO) method as an evolutionary algorithm through the VISSIM COM interface and parallel optimisation technique to reduce the cost of auto-tuning and calibration. This paper could be of interest to transport experts in particularly those who are using traffic micro-simulation and looking for auto-calibration approach. Keywords:  Traffic micro-simulation, Calibration, Auto-tuning, Evolutionary algorithm, Parallel optimisation, Multithread, Particle Swarm Optimisation (PSO) 1. Introduction Traffic simulation models have become an important and popular tool in modelling transport systems, in particular, owing to advent of fast and powerful computers. One of the supreme advantages of using such tools is to assess different alternates and scenarios prior to their implementations. Traffic simulation models could be divided into three categories including microscopic, macroscopic, mesoscopic simulation models. First category simulates the movement of individual vehicles in a traffic stream. Car-following and lane-changing models are the two fundamental components in traffic micro-simulations. The second (macroscopic) category simulates transportation network section-by-section rather than by tracking individual vehicles. The relationships between flow, speed, and density of traffic stream form the fundamental basis of this category. Mesoscopic traffic simulation models combine the properties of the first and second models.  2  Along with the increasing popularity and use of traffic simulations, an essential concern has been raised about their proper applications in the study they are used for; or more specifically their appropriate calibration and validation. As no single model can comprise the whole universe of variables, every model must be adapted for local conditions using real world data. The performance of the model should be also evaluated through independent data sets. These processes are known as calibration and validation. More specifically, calibration of the traffic simulation normally refers to computing the magnitude of the parameters embedded in the simulation models to match the real traffic and local driving behaviour. The outputs of traffic simulations may not be accurate and reliable without appropriate calibration. However, the calibration process, especially for microscopic simulations, could be a complex and time-consuming task because of the large number of unknown parameters (Toledo et al. 2004). Some studies (e.g. Gardes et al. 2002, Chu et al. 2003, Park and Schneeberger 2003, Moridpour et al. 2012) used the generic procedure to calibrate traffic micro-simulations using sensitivity analysis and trial-and-error which could be very resource-intensive and time-consuming. Some other studies (e.g. Lee et al. 2000, Ma and  Abdulhai 2002, Park and Qi 2005, and Menneni et al. 2008) used an evolutionary algorithm like Genetic Algorithm (GA) for calibration purposes. However, the process may be still time-consuming due to the significant computational load associated with large-scale traffic simulation runs. This study introduces a methodology to calibrate traffic micro-simulations based on an evolutionary algorithm known as Particle Swarm Optimisation (PSO). To overcome the long running time, it applies multi-thread technique and implements Parallel PSO algorithm. This method can use several CPUs and runs several simulation instances in parallel which can shorten the run time significantly. VISSIM (2012) traffic micro-simulation was used in this study for implementation of the algorithm. Section 2 explains the VISSIM interface and defines the important parameters that should be considered for calibration procedure. In particular, the parameters related to the driving behaviours are discussed in this section. Section 3 explains about the optimisation techniques and the particle swarm optimisation (PSO) algorithm is discussed in details. Section 4 presents the implementation and parallelisation of the PSO algorithm for auto-calibration of the traffic micro-simulation. The paper is closed by providing some conclusions and further remarks for future work in Section 5. 2. VISSIM and Calibration Parameters The micro-simulation which was used in this study is VISSIM (2012), version 5.40. The name is derived from “Verkehr In Städten - SIMulationsmodell” (German for “Traffic in cities - simulation model”). The software was developed at the University of Karlsruhe, Karlsruhe, Germany, during the early 1970s. Commercial distribution of VISSIM began in 1993 by PTV Transport Verkehr AG, which has continued to distribute and maintain VISSIM till now. VISSIM is one of the latest traffic micro-simulations available and provides significant enhancements in terms of driver behaviour, multi-modal transit operations, interface with planning / forecasting models, and 3-D simulation. VISSIM (2012) is a microscopic, time step and behaviour based simulation model developed to analyse private and public transport operations under constraints such as lane configuration, vehicle composition, traffic signals and so on. Access to model data and simulation is provided through a COM interface, which allows VISSIM to work as an  Automation Server and to export the objects, methods and properties. The VISSIM COM interface supports Microsoft Automation and thus the program can be implemented in any of the RAD (Rapid Application Development) tools ranging from scripting languages like Visual  A NOVEL METHODOLOGY FOR EVOLUTIONARY CALIBRATION OF VISSIM BY MULTI-THREADING  3 Basic Script or Java Script to programming environments like Visual C++ or Visual J++.  Also, internal driving behaviour can be replaced by a fully user-defined behaviour using the External Driver Model DLL Interface of VISSIM. The accuracy of the traffic flow simulation model is highly related to the accuracy of estimating vehicles’ movements in the network. The driving beha viour in the micro-simulation is linked to each link by its behaviour type. The traffic flow model in VISSIM is a discrete, stochastic, time step based, microscopic model with driver-vehicle-units as single objects. The model contains a psycho-physical car-following model for longitudinal vehicle movement and a rule-based algorithm for lateral movements. The model is based on the continued work of Wiedemann (1974) for car-following process and Wiedemann and Reiter (1992) for lane-changing manoeuvres. The car-following and lane changing models and their associate parameters are explained in the following sub-sections. These parameters affect the vehicle interactions directly and cause substantial differences in simulation results and thus should be considered in the calibration procedure particularly. However, the calibration parameters are not limited to the parameters associated with the car-following and lane changing model. Some further parameters which could be considered are used to define the maximum and desired acceleration/deceleration of vehicles, desired speed distributions, and signal control. Here two fundamental components within the traffic micro-simulations are presented as an example, but the rest of parameters can be also considered in calibration procedure. The proposed algorithm allows setting as many parameters as desired for calibration. However, more parameters will result in more complex process and thus will require more running time. 2.1. Car-following model VISSIM uses the psychophysical car-following model developed by Wiedemann (1974). The concept of this model is that the faster moving vehicle drivers approaching slower vehicle start decelerating when they reach their own individual perception threshold. However, the speed may become smaller than the lead vehicle speed as the results of driver’s imperfection in the estimation of the lead vehicle speed. This means the driver will accelerate slightly again after reaching another threshold. This results in an iterative process of acceleration and deceleration due to drivers’ imperfections to determine the exact speeds of the lead vehicles. Figure 1 shows a typical car-following behaviour of a vehicle based on the logic explained above.   There exist two car-following models in VISSIM: Wiedemann74 and Wiedemann99 (VISSIM 2012). The former one is suggested to be applied for urban arterial roads and the later one is more suitable for freeways. The basic idea of the models is the assumption that a driver is in one of the four driving modes: Free driving, Approaching, following or braking. These modes are determined by the following six thresholds (also shown in Figure 1):   AX: the desired distance between two stationary vehicles   BX: the minimum following distance which is considered as a safe distance by drivers   CLDV: the points at short distances where drivers perceive that their speeds are higher than their lead vehicle speeds   SDV: the points at long distances where drivers perceive speed differences when they are approaching slower vehicles   OPDV: the points at short distances where drivers perceive that they are travelling at a lower speed than their leader   SDX: The maximum following distance indicating the upper limit of car-following process  4 Figure 1: a typical car-following behaviour of a vehicle (VISSIM 2012) More details about these thresholds can be found in Wiedemann (1974) or VISSIM (2012). For each mode, the acceleration could be determined as a result of speed, relative speed, space headway and the individual characteristics of driver and vehicle. This study explains the Wiedemann99 car-following model with more details as it is more suitable for freeways and contains more parameters causing more difficulties for calibration. Further, the relation between the calibration parameters and the perceptual thresholds are required to be investigated in details. Figure 2 is a snapshot of VISSIM interface showing the parameters associated with Wiedemann99 car-following model. The parameter explanations are presented below. The ‘Look ahead distance’  parameters determine the minimum and maximum distances as well as the number of vehicles in front of a driver in the same link that can be observed and thus influence driver’s reaction accordingly. The ‘Look back distance’  parameters determine the maximum and minimum distances that a driver can see backwards within the same link in order to react to other vehicles behind. The ‘temporary lack of attention’  parameters determine the probability and time duration in which a driver does not react to the behaviour of lead vehicle (except for emergency braking). The ‘smooth closeup behavior’  parameter determines whether or not drivers slow down more smoothly when approaching standing obstacles. If it is checked, drivers will prepare to stop behind the obstacle from the maximum look ahead distance. If it is not checked, drivers will have the normal following behaviour and will not consider the obstacle from the long distance. The ‘s tandstill distance for static obstacles ’ parameter determines the distance that d rivers keep while standing in upstream of all static obstacles. The distance can be fixed by checking the box otherwise it would be a random value following a normal distribution with the mean of   and variance of      .

For some

Jul 23, 2017

Egan

Jul 23, 2017
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks