Revisit ODME, Need guidance.

3 posts / 0 new
Last post
jabunch
Revisit ODME, Need guidance.

We have recently started using ODME in different Travel Forecasting Packages to refine the outputs from a regional travel demand model to improve their peak period vehicle trips in and out of a subarea or corridor.  We have primarily been using a Least Squares ODME estimation method.  We have learned a lot but are starting a new project and their is a  debate going on amongst staff  on how to use ODME to create the trip tables that are input to subsequent meso and micro assignments.  I need some advice/guidance on recommended practices and what to avoid.  Also, there was a healthy discussion here in 2014/2015 on ODME, and since use of ODME is becoming standard practice (as recommended by NCHRP 765) if we have learned anything more in the 6 to 7 years that have followed.  Has any guidance on using ODME (National, State, Local) been established that you refer to? NCHRP 765 is too high level.

What we know from our initial investigation, is that good ODME output is really dependent on starting with a validated assignment method that captures the relative impedances on alternative paths, and different bottlenecks within the system, A network and zone system that represents the major routes and traveler options within the study area, and the number and quality of the counts themselves.  We recognize that any ODME only produces 1 possible origin to destination trip table solution among many that can produce the same count values.

So here are my questions/issues. They focus on using ODME for more complex subarea studies that encompass alternative paths along a corridor, or neighborhoods/activity centers/malls and business complexes  versus more simple (conceptually) estimating the ramp to ramp volumes on access controlled facilities or networks.

  • What insights can you provide on the number of counts, their location, and mixing turning movement counts with mid-link counts? 
    • Should counts be "balanced"?  Should they be seasonally adjusted?
    • Can you over-specify the number and locations of the counts?  for example, inputting for every signalized intersection in the subarea?
    • What are your thoughts on where counts should be located?  In our previous work we started with screenline/cordon locations, and then added additional locations in subsequent iterations.
  • We have read about carrying out ODME in iterations where one pass feeds another pass.  Is it good practice to add additional count locations between passes?  When this is done how do you preserve the general patterns that come out of Pass 1 so Pass 2 does not distort the results significantly (Pass 2 is conceptually refining Pass 1)?  Should we simply start with all count locations and somehow weight the screenline and cordon count locations higher?
  • We are having a discussion on whether to use a seed trip table from the regional travel demand model, versus a table of 1's on all connected OD pairs.  I can see how a table of 1's can sort of work from ramp to ramp ODs for a controlled access facility, but is it a reasonable approach for more complex subarea or corridor networks?  There is also the question of leaks in the network due to driveways, parking lots, etc. in the subarea in question and obtaining counts for them.
  • How do you evaluate the quality of the ODME output?  In our previous work the R2 and % RMSE from the assignments would be great. However, we noticed that we would input counts at connectors to zones, and then find that this would cause wild swings in the origins/destinations to other zones.  Or there would be a backup in the network prior to the count location and the ODME would continue to try to match the count and actually increase the error.   This showed us that having a validated assignment process prior to ODME is critical.
  • Related to the above, how do you address oversaturated condition's and/or peak spreading within the ODME process?  This is especially important when not all network options exist within the operations simulation model, e.g. only the major arterials and freeways/interstates are coded (locals, collectors, and minor arterials are not coded). 

Just as an aside, we are looking for practitioner level guidance, not academic papers on the algorithms. Yes I have read Hans Speiz's and other paeprs.

Thanks for your help on this.

Jim Bunch, Senior Transportation Planner
Mead & Hunt.

Pilo Willumsen

Dear Jim,

These are important and relevant questions. Britain and other countries have been using these techniques for some time because of the emphasis in obtaining the best possible matrices for the base year (and apply incremental modelling for future years).

There is a document with useful experience and guidance for this task: https://www.gov.uk/government/publications/matrix-building-for-transport...

Luis

……………………………………………

Luis Willumsen

Director

Willumsen Advisory Services

Nommon

London & Madrid

T: +44 7979 53 88 45

www.nommon.es

…………………………………………..

From: on behalf of jabunch
Date: Wednesday, 24 March 2021 at 17:29
To: TMIP
Subject: [TMIP] Revisit ODME, Need guidance.

We have recently started using ODME in different Travel Forecasting Packages to refine the outputs from a regional travel demand model to improve their peak period vehicle trips in and out of a subarea or corridor. We have primarily been using a Least Squares ODME estimation method. We have learned a lot but are starting a new project and their is a debate going on amongst staff on how to use ODME to create the trip tables that are input to subsequent meso and micro assignments. I need some advice/guidance on recommended practices and what to avoid. Also, there was a healthy discussion here in 2014/2015 on ODME, and since use of ODME is becoming standard practice (as recommended by NCHRP 765) if we have learned anything more in the 6 to 7 years that have followed. Has any guidance on using ODME (National, State, Local) been established that you refer to? NCHRP 765 is too high level.

What we know from our initial investigation, is that good ODME output is really dependent on starting with a validated assignment method that captures the relative impedances on alternative paths, and different bottlenecks within the system, A network and zone system that represents the major routes and traveler options within the study area, and the number and quality of the counts themselves. We recognize that any ODME only produces 1 possible origin to destination trip table solution among many that can produce the same count values.

So here are my questions/issues. They focus on using ODME for more complex subarea studies that encompass alternative paths along a corridor, or neighborhoods/activity centers/malls and business complexes versus more simple (conceptually) estimating the ramp to ramp volumes on access controlled facilities or networks.
What insights can you provide on the number of counts, their location, and mixing turning movement counts with mid-link counts?
Should counts be "balanced"? Should they be seasonally adjusted?
Can you over-specify the number and locations of the counts? for example, inputting for every signalized intersection in the subarea?
What are your thoughts on where counts should be located? In our previous work we started with screenline/cordon locations, and then added additional locations in subsequent iterations.
We have read about carrying out ODME in iterations where one pass feeds another pass. Is it good practice to add additional count locations between passes? When this is done how do you preserve the general patterns that come out of Pass 1 so Pass 2 does not distort the results significantly (Pass 2 is conceptually refining Pass 1)? Should we simply start with all count locations and somehow weight the screenline and cordon count locations higher?
We are having a discussion on whether to use a seed trip table from the regional travel demand model, versus a table of 1's on all connected OD pairs. I can see how a table of 1's can sort of work from ramp to ramp ODs for a controlled access facility, but is it a reasonable approach for more complex subarea or corridor networks? There is also the question of leaks in the network due to driveways, parking lots, etc. in the subarea in question and obtaining counts for them.
How do you evaluate the quality of the ODME output? In our previous work the R2 and % RMSE from the assignments would be great. However, we noticed that we would input counts at connectors to zones, and then find that this would cause wild swings in the origins/destinations to other zones. Or there would be a backup in the network prior to the count location and the ODME would continue to try to match the count and actually increase the error. This showed us that having a validated assignment process prior to ODME is critical.
Related to the above, how do you address oversaturated condition's and/or peak spreading within the ODME process? This is especially important when not all network options exist within the operations simulation model, e.g. only the major arterials and freeways/interstates are coded (locals, collectors, and minor arterials are not coded).
Just as an aside, we are looking for practitioner level guidance, not academic papers on the algorithms. Yes I have read Hans Speiz's and other paeprs.

Thanks for your help on this.

Jim Bunch, Senior Transportation Planner
Mead & Hunt.

--
Full post: https://tmip.org/content/revisit-odme-need-guidance
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13565

jabunch

Thanks very much.  The DFT guidance is proving to be very helpful.