Specifically, can we do car-following simulations? I have read in different articles e.g. Optimization of Wiedemann and Fritzsche car-following models for emission estimation that a car-following model can be simulated in MATLAB. What are the prerequisites for running such a simulation? What is the essential knowledge required for writing a program to run such simulation?
http://www.sciencedirect.com/science/article/pii/S1361920914001862
Hi,
Don't know much about traffic simulation programming but it typically runs like any physics based simulation: 1) for every dt (time step) 2) you evaluate all the possible conditions and 3) determine the transformations for all objects on scene.
Car-following models are (as far as I'm aware) nothing more than the set of conditions inside step 2). The context of your simulation however will determine what kind of knowledge you need to perform your algorithm.
For example: if you want the type of possible collisions between cars (or any other object) as a criteria you'll need to use (or build) a physics engine. If this is not important and you just want to determine areas with greater vehicle density maybe you just need to build distance based events (between objects I mean).
I can't really state the difficulty for the Traffic Simulation (microsimulation) task but the programming will require you to have, at least, some base knowledge about objects (class), statements (if/elif/else) and loops (for/while).
It may make your life a lot easier if you look for a library that already does much for your project. See if this helps you: http://jfietkau.github.io/Streets4MPI/
Thank you very much Pedro. I'll explore the Streets4MPI soon. One thing that I didn't understand is the 'physics engine' thing you mentioned. If I have an equation that describes the acceleration of a vehicle for each time step where the input parameters are vehicle's own speed and distance to lead vehicle, then would I need the physics engine?
By physics engine I meant a full working collision and movement framework. For instance if "crash" is a possible event in your simulation than you would have to build a collision criteria for your loop.
So if you have 10 vehicles (let's say 10 rectangles) within two road physical barriers (let's say each is composed of 20 squares) you'll have to test for each "dt" (time step) if any of the vehicles collided with either barriers or other vehicles (so this would imply at least another loop). This is a widely debated subject in Game Development and you'll find plenty of examples online.
Only consider this if you really need it since it's another layer of complexity. As I said I don't know much about Traffic Simulation but it seems to me you don't need this (meaning the algorithm itself would stop the collision from being possible).
Simulation of car-following models can be done in excel (see attachment), so surely they can be done in more elaborated environments such as R or Python. See also my microsimulation at http://www.mtreiber.de/MicroApplet_html5 which is written in html5/JavaScript
Thanks Martin Treiber. The attached excel sheet is very useful. I have also started reading your book 'Traffic Flow Dynamics' which is also a very useful resource.
Thanks Hong Yang for the suggestion. Would you mind sharing your R code? Also one question to everyone: What is the initial speed, desired speed and desired accelerations for a random vehicle in the simulation? In VISSIM they are defined explicitly by speed distributions and acceleration functions. How can we control these in our own simulation?
Basically, the parameter set of a car-following model defines the driving characteristics of a certain driver-vehicle combination (technically also called a "driver-vehicle unit", DVU). This gives hints for initializing/controlling them:
* Since the driving style of a given driver changes only insignificantly over time (the "intra-driver variations" are much less than the "inter-driver variations") the parameters characterizing the style, e.g., desired acceleration, deceleration, and time gap, should be set at vehicle creation time (e.g., as a realization of a random variable) and be kept constant for the lifetime of this DVU in the simulation
* The desired speed depends strongly on the type of road and on the location. Therefore, in contrast to the style parameters mentioned above, this parameter should be changed also during the lifetime of a DVU whenever it passes an inhomogeneity (e.g., temporarily decreased when a passage through a village is simulated).
* The distributions to draw the parameters from represent the inter-driver variations. Generally, they are correlated, e.g., a high deceleration often comes with a low desired time gap since both are indicators of an "aggressive" style. Alternatively, one can define several DVU classes (e.g., moderate car drivers, aggressive drivers, and truck drivers) each having a constant parameter set, and assigning new DVUs with certain probabilities to one of these types. Thus, correlations are automatically included
* You have to carefully distinguish between model parameters such as desired speed representing the driving characteristics and initial values of actual state variables such as the initial speed. There is no rule for the latter: It just depends what you want to simulate. In open systems, the initial values often are irrelevant anyway, since one regularly needs a "warmup period" before producing the actual results.
Yes very much in python. We have used it in our learning sessions in lab. We used tkinter
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators