Thank you for the videos. You are a wonderful teacher. I was wondering what are the purposes of doing dynamic assignment? Is it because dynamic assignment helps to calibrate the model? You discussed about hyperparameter tuning in dynamic assignment (surcharges, costs, logit, kirchoff exponent). What results from VISSIM help us understand when to stop with hyperparameter tuning in dynamic assignment? At what point in the simulation are we satisfied with the current settings?
Hi, - Dynamic models allow you to use matrices, so saves you lot of time not setting up static routes - Big models where alternative routes exist, it is essential to use dynamic, so vehicles can take into account other options. - In the future you are testing a scenario where you build a new road, widen an existing one, so you want to see how many vehicles are attracted by the extra capacity - In real life traffic data usually comes from strategic model cordons as a matrix, that you calibrate against counts. So you must use matrices in this case as well. - Many times you survey traffic by ANPR cameras (licence plate recognition), and that also gives you matrices - Very rarely static routes are adequate for modelling. Only for very simple cases. - For individual small intersection models other software is used, and in Vissim usually we model smaller areas. That is Vissim the best for. It can model interaction between intersection well. -If you have static routes and you do any change in your model (move a link/update a connector...) you need to set up the static routes again. There are some other aspects as well and the combination of the above mentioned ones.
He Matyas, Thanks for the video. Got 1 question. How do you setup the settings when you put it in scenario management? I usually use Morning and evening rush hour matrices and it changes with each modification
Hi, not sure what kind of settings do you refer to? Could you elaborate please? Between scenarios, lets say AM and PM peak, normally, you should not have differences besides signals and traffic demand.
Hi Matyas! I wonder what is the difference between using dyn1_019.bew and dyn1.bew (of course, same question for weg). Thank you so much in advance for the answer and for all your videos!
Hey, each bew and weg file represent a different run. These are cost and path files. Costs and paths can change between each simulation run when you use dynamic assignment. So '019' shows the number of simulation run. You b, using that I am basically re running simulation run no 19. As you run multiruns vissim can save the path and cost file so you can go back and pick the run you need. Dynamic assignment mainly used when there are alternative route choices available. You need to converge (balance out vehicles across alternative runs iteratively), and when this happens you take that run
It is gonna be a long answer :) - Usually models run for quite a long time and it would take ages to undertake 100 runs or more - During a year, traffic levels change a lot, there are seasonality factors to consider - Usually models are built to represent a relatively busy, if not the busiest periods - Models in most of the cases need to follow guidelines set by local/national authorities, councils and they in many cases have their own requirements on number of runs (see Transport for London modelling guidelines for example) - After 5-10-20 runs, you don't expect any material change in the results/outputs, because by increasing the number of runs the results should just get closer to the 'ultimate average', basically you would just confirm more and more decimal number in delays/queue length results... - models are just models, and include lots of assumptions usually, so spending lot of more time running it just would not make sense, considering the result of 20 run and the result of 21 run, and 50 runs are nearly identical anyway. - and lastly, the longer the model runs take, the more it costs for the client in fees, or less the profit the modeller would make. So it is a balance between cost/output quality
@@MatyasLemberTutorials Thank you for the answer. In 6:30 , you were saying that you added 5 runs to represent daily variability, and then you picked the average. So, I thought in order to get an average value with (say, 95% confidence) level, we need to get more samples, hence more runs. In statistics we usually try to get the confidence level when we finally present an output for research. So, that was my line of thought. But I understand you said in your comments to follow government regulations and suggested to go for 20 runs.
Hi, yes of course, what you say is true and can be useful tool for congested network. There is also an output called Standard Deviation for all results. That indicates how different the runs are within the average. I added some related info below for this topic, you might find it useful. Usually after 10-20 runs the average does not/should not change considerably. If it does, then the model itself usually is not good enough. Calibration is not stable for example, or the network operates over capacity, or there is something not coded well. In these cases further analysis/work is needed. In case of a congested network, more runs to average maybe :). We try to achieve higher confidence level in addition to enough simulation runs, by expanding survey programs, achieving better validation levels, getting higher convergence levels (higher over more consecutive runs) and so on. These are all important aspects. Please note that my opinion is mainly based on consultancy work and impact assessments and not on research. In research you might have difference objectives and requirements.
Those guidelines are lots of fun :D If interested, check out TfL's (Transport for London) modelling guidelines as a start. Online available, version 4 is the latest I think. It discusses multiple modelling tools in it, Vissim is one of them.
Thank you for the videos. You are a wonderful teacher. I was wondering what are the purposes of doing dynamic assignment? Is it because dynamic assignment helps to calibrate the model? You discussed about hyperparameter tuning in dynamic assignment (surcharges, costs, logit, kirchoff exponent). What results from VISSIM help us understand when to stop with hyperparameter tuning in dynamic assignment? At what point in the simulation are we satisfied with the current settings?
Hi,
- Dynamic models allow you to use matrices, so saves you lot of time not setting up static routes
- Big models where alternative routes exist, it is essential to use dynamic, so vehicles can take into account other options.
- In the future you are testing a scenario where you build a new road, widen an existing one, so you want to see how many vehicles are attracted by the extra capacity
- In real life traffic data usually comes from strategic model cordons as a matrix, that you calibrate against counts. So you must use matrices in this case as well.
- Many times you survey traffic by ANPR cameras (licence plate recognition), and that also gives you matrices
- Very rarely static routes are adequate for modelling. Only for very simple cases.
- For individual small intersection models other software is used, and in Vissim usually we model smaller areas. That is Vissim the best for. It can model interaction between intersection well.
-If you have static routes and you do any change in your model (move a link/update a connector...) you need to set up the static routes again.
There are some other aspects as well and the combination of the above mentioned ones.
He Matyas,
Thanks for the video. Got 1 question. How do you setup the settings when you put it in scenario management? I usually use Morning and evening rush hour matrices and it changes with each modification
Hi, not sure what kind of settings do you refer to? Could you elaborate please?
Between scenarios, lets say AM and PM peak, normally, you should not have differences besides signals and traffic demand.
Hi Matyas!
I wonder what is the difference between using dyn1_019.bew and dyn1.bew (of course, same question for weg).
Thank you so much in advance for the answer and for all your videos!
Hey, each bew and weg file represent a different run. These are cost and path files. Costs and paths can change between each simulation run when you use dynamic assignment.
So '019' shows the number of simulation run. You b, using that I am basically re running simulation run no 19.
As you run multiruns vissim can save the path and cost file so you can go back and pick the run you need.
Dynamic assignment mainly used when there are alternative route choices available. You need to converge (balance out vehicles across alternative runs iteratively), and when this happens you take that run
Hello, 5:55 why use only 5 runs. Why not use 100 runs. Or maybe 365 runs to represent one year?
It is gonna be a long answer :)
- Usually models run for quite a long time and it would take ages to undertake 100 runs or more
- During a year, traffic levels change a lot, there are seasonality factors to consider
- Usually models are built to represent a relatively busy, if not the busiest periods
- Models in most of the cases need to follow guidelines set by local/national authorities, councils and they in many cases have their own requirements on number of runs (see Transport for London modelling guidelines for example)
- After 5-10-20 runs, you don't expect any material change in the results/outputs, because by increasing the number of runs the results should just get closer to the 'ultimate average', basically you would just confirm more and more decimal number in delays/queue length results...
- models are just models, and include lots of assumptions usually, so spending lot of more time running it just would not make sense, considering the result of 20 run and the result of 21 run, and 50 runs are nearly identical anyway.
- and lastly, the longer the model runs take, the more it costs for the client in fees, or less the profit the modeller would make. So it is a balance between cost/output quality
@@MatyasLemberTutorials Thank you for the answer. In 6:30 , you were saying that you added 5 runs to represent daily variability, and then you picked the average. So, I thought in order to get an average value with (say, 95% confidence) level, we need to get more samples, hence more runs. In statistics we usually try to get the confidence level when we finally present an output for research. So, that was my line of thought. But I understand you said in your comments to follow government regulations and suggested to go for 20 runs.
Hi, yes of course, what you say is true and can be useful tool for congested network. There is also an output called Standard Deviation for all results. That indicates how different the runs are within the average.
I added some related info below for this topic, you might find it useful.
Usually after 10-20 runs the average does not/should not change considerably. If it does, then the model itself usually is not good enough. Calibration is not stable for example, or the network operates over capacity, or there is something not coded well. In these cases further analysis/work is needed. In case of a congested network, more runs to average maybe :).
We try to achieve higher confidence level in addition to enough simulation runs, by expanding survey programs, achieving better validation levels, getting higher convergence levels (higher over more consecutive runs) and so on. These are all important aspects.
Please note that my opinion is mainly based on consultancy work and impact assessments and not on research. In research you might have difference objectives and requirements.
@@MatyasLemberTutorials Thank you for the answer. Yes, I want to learn about standard practices in the industry and consultancy now.
Those guidelines are lots of fun :D
If interested, check out TfL's (Transport for London) modelling guidelines as a start. Online available, version 4 is the latest I think. It discusses multiple modelling tools in it, Vissim is one of them.