[Q&A] Introducing DRT: Scoring Parameters and Impact on Modal Share

[ArezHK]

Provide the MATSim version you are using. You can find the version in the pom.xml. E.g. 2025.0.

16.0-2024w15

Did you build a minimal example to reproduce the issue?

• Yes!

Describe the problem you are facing.

Hello,
I have a MATSim model for a specific area and have already completed the calibration for the following modes: car, pt, walk, bicycle, and pedelec. The current parameters are as follows:

	<parameterset type="scoringParameters">
	<param name="earlyDeparture" value="0.0" />
	<param name="lateArrival" value="-18.0" />
	<param name="marginalUtilityOfMoney" value="1.0" />
	<param name="performing" value="6.88" />
	<param name="utilityOfLineSwitch" value="-1.0"/>
	<param name="waiting" value="-0.0" />		
	<param name="waitingPt" value="0.0" />
											 										 
	<parameterset type="modeParams">
		<param name="mode" value="car"/>
		<param name="constant" value="-1.66" />
		<param name="dailyUtilityConstant" value="0.0" />			
		<param name="marginalUtilityOfDistance_util_m" value="0.0" />			
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<!-- fixed annual costs like insurance, parking around 2000 euros: 5.3 euro daily -->				
		<param name="dailyMonetaryConstant" value="-5.3"/>
		<!-- fuel cost 0.2 cents per km, 0.0002 per meter -->				
		<param name="monetaryDistanceRate" value="-0.0002"/>
	</parameterset>

	<parameterset type="modeParams">
		<param name="mode" value="pt"/>
		<param name="constant" value="-8.58" />
		<param name="dailyUtilityConstant" value="0.0" />		
		<param name="marginalUtilityOfDistance_util_m" value="0.0" />				
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<param name="monetaryDistanceRate" value="0.0"/>
		<!-- db ticket 58x12=696, 1.9 daily -->				
		<param name="dailyMonetaryConstant" value="-1.9"/>
	</parameterset>
	<parameterset type="modeParams">
		<param name="mode" value="walk"/>
		<param name="constant" value="0.0"/>			
		<param name="dailyMonetaryConstant" value="0.0" />
		<param name="dailyUtilityConstant" value="0.0" />			
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<param name="marginalUtilityOfDistance_util_m" value="0.0"/>
		<param name="monetaryDistanceRate" value="0.0"/>
	</parameterset>

	<parameterset type="modeParams" >
		<param name="mode" value="bicycle" />		
		<param name="constant" value="-3.08" />
		<!-- 30 euro insurance: 0.082 daily -->				
		<param name="dailyMonetaryConstant" value="-0.082" />		
		<param name="dailyUtilityConstant" value="0.0" />			
		<param name="marginalUtilityOfDistance_util_m" value="0.0" />
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<param name="monetaryDistanceRate" value="0.0" />
	</parameterset>

	<parameterset type="modeParams" >
		<param name="mode" value="pedelec" />		
		<param name="constant" value="-4.84" />
		<!-- 60 euro insurance, 40 euro maintenace: 0.27 euro daily -->
		<param name="dailyMonetaryConstant" value="-0.27" />
		<param name="dailyUtilityConstant" value="0.0" />			
		<param name="marginalUtilityOfDistance_util_m" value="0.0" />
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<!-- 0.03 euro per km and 0.00003 per meter -->
		<param name="monetaryDistanceRate" value="-3E-5" />
	</parameterset>

Now, I want to add a new mode called 'drt' and define its parameters in a way that is logical and realistic, but I’m unsure which parameters to modify and what values to assign. Should I also edit drt fare parameters?

	<parameterset type="modeParams">
	        <param name="mode" value="drt" />
		<param name="constant" value="0.0"/>			
		<param name="dailyMonetaryConstant" value="0.0" />
		<param name="dailyUtilityConstant" value="0.0" />			
		<param name="marginalUtilityOfTraveling_util_hr" value="0.0" />
		<param name="marginalUtilityOfDistance_util_m" value="0.0"/>
		<param name="monetaryDistanceRate" value="0.0"/>
	</parameterset>	

Should I also consider adjusting the parameters of existing modes (e.g., to reflect inconvenience, competitiveness, or other behavioral shifts) when introducing DRT?

Additionally, I have configured the following strategy settings:

SubtourModeChoice weight: 0.2

BestScore weight: 0.9

My goal is to analyze the effect of introducing DRT in the area and observe the resulting changes in modal share.

I would appreciate your guidance or suggestions on how to set appropriate parameters for the DRT mode.

Thank you in advance!

If possible, provide a link to your minimal example. This might be a GitHub repository.

No response

If you want, provide some background information about you and your project. This might help us understand your issue better.

No response

[paulheinr]

@luchengqi7 @nkuehnel could you help as you are our DRT experts?

[luchengqi7]

Hello,

1. You don't need to adjust parameters for the calibrated mode. You just need to specify the parameters for DRT mode (e.g., ASC, disutility for travel). The parameter set for DRT has the same format as other modes.

2. You may refer to this repository for some DRT examples: https://github.com/matsim-vsp/drt-scenario-library

3. Usually we use Change Exponential Beta or Select Exponential Beta as the non-innovative strategy (instead of Best Score). You may consider to switch to one of the two strategies.

Hope this helps!

[ArezHK]

Dear @luchengqi7 ,

Thank you for your response.

I have a few additional questions I hope you can help me with as well:

-----1. About ChangeExpBeta and SelectExpBeta:
If I choose ChangeExpBeta or SelectExpBeta, should I first re-run the previous calibration based on the real modal share to find updated ASCs, and then add the DRT mode to the current modes?

Also, could you briefly explain the difference between BestScore, ChangeExpBeta, and SelectExpBeta?

-----2. Adjusting DRT Cost Parameters:
I noticed two places where DRT cost parameters can be adjusted:

A. In the multiModeDrt module:

B. In the scoring module:

Some of these parameters (e.g., dailySubscriptionFee and dailyMonetaryConstant) appear quite similar. I'm a bit confused about which ones should be changed depending on the perspective (user vs. DRT provider). I want to test multiple scenarios, such as:

Cost per km
Cost per hour
One-time usage payment independent of time and distance
Annual subscriptions

Which module and parameters should be used to reflect each of these perspectives correctly?

-----3. On Adjusting ASC for DRT:
After adding the DRT mode in the config file and defining parameters in the scoring module, is it advisable to also change the ASC for DRT? Or would it be better to tune other parameters instead?
If adjusting the ASC is acceptable, could you please suggest a reasonable starting value for it?

----- 4. On Disutilities and mode Parameters:
As Mr. Rieser mentioned parameters related to walk and bike disutilities, do you have any suggestions for changes to my other parameters—such as those related to cost, disutility, etc.—that could help make the model more realistic?

Sorry for the long and numerous questions, and thank you in advance for your help.

[luchengqi7]

1. Yes, you may need to re-run the calibration, as the result may change.
   Both "Select exponential beta" and "Change exponential beta" follow the logit selection model. There are some minor difference between the two (for more details you can refer to the MATSim book). You can use either of them. We usually use "Change Exponential Beta".

Select best can be a little bit too strong. And we need some agents to be able to choose not the best plan, such that the system may move away from a local optimum and converge to a better point.

2. Both parameters will impact the DRT users. The "modeParams" is the same as other modes. The fare is the fare (which will also be converted to the disutility) on top of the travel disutility specified in "modeParams". There will be no charge on the DRT operators.

3. It is advisable to fix the other parameters in the "modeParams" (e.g., time and distance disutility) and fares to a reasonable value. Then adjust the ASC, until you reach your desired mode share. We should try to avoid extreme values of ASC when calibrating the system. If ASC is too large or too small (i.e., too negative), then you may consider to adjust other "modeParams" and re-do the ASC calibration.

4. We (VSP) usually set walk to 0.0 (ASC and time/distance utility), as many other modes also have walk leg (e.g., walk to PT stops, walk to the road to drive car). The long walk will be penalized by the opportunity cost (i.e., performing, which is 6.0 by default*), and we believe that should be adequate. For Bike, we usually have some penalty for distance to simulate the effect of cyclist getting tired after riding for long distance.

@mrieser Do you have other suggestions on this point (i.e., what are the suggested typical parameters for walk mode from your side)?

*: By the way, 6.0 per hour means 0.1 point per minute. The other cost parameters can be adjusted according to this scale. The moeny will also be converted to score, and, by default, 1 unit of money = 1 score.

[ArezHK]

Dear @luchengqi7 ,

Thank you for your response.

----1. As you mentioned, the DRT fare is added on top of the travel disutility specified in modeParams. Would it be acceptable if I set the DRT fare values to zero and instead just adjust the scoring parameters for drt? (For other modes like PT, I won’t be adding fare parameters either.)

----2. Additionally, I reviewed the scenarios you previously mentioned. After checking the config files from the Leipzig 1.1 scenario, I noticed that the marginalUtilityOfTraveling_util_hr values vary across modes. However, I’m unsure whether these values were derived from any specific calculation or underlying logic. Here are the values I found:

Mode "walk": param name="marginalUtilityOfTraveling_util_hr" value="-1.5"
Mode "bike": param name="marginalUtilityOfTraveling_util_hr" value="-10.0"
Mode "pt": param name="marginalUtilityOfTraveling_util_hr" value="-2.0"
Mode "car": param name="marginalUtilityOfTraveling_util_hr" value="-0.5"
Mode "drt": param name="marginalUtilityOfTraveling_util_hr" value="-2.0"
param name="waitingPt" value="-2.0"
param name="utilityOfLineSwitch" value="-1.0"

In contrast, in the Berlin 6.3 scenario, most of these values are set to zero.

In your experience, which approach is more appropriate for realistically representing the discomfort or disutility of different travel modes? So far, I have followed the Berlin scenario and have not set marginalUtilityOfTraveling_util_hr values for different modes. However, from a logical perspective, using public transport (PT) is generally more unpleasant than a door-to-door DRT service.

----3.Finally, since I am introducing a new DRT mode that does not currently exist in my study area, there is no real-world modal share data available. In this case, which modal share values or simulation results can be considered reliable or trustworthy?

I appreciate your insights in advance and thank you for your time and support.

[mrieser]

Looking at your scoring parameters, I notice a few things:

• your parameters for mode walk are all set to 0.0. So even if the agent walks 50km or even 100km, the walk-leg would get a score of just 0.0, and the score is completely independent of the distance or time the agent spends walking.
• the bicycle mode only defines constant values, but no distance- or time-based values. Agents will receive the same amount of (negative) utility for a bicycle-leg no matter if they travel 100 meters or 100 kilometers with the bicycle.

While technically the agents would still occur opportunity costs (they could not spend time at an activity and gain utility there while they are walking/cycling), I still think the constant values for walk- and bike-legs pose a risk for strange side-effects and potentially strange mode choice behavior for individual agents.

[ArezHK]

Dear @mrieser ,

Thank you for your response.

As you mentioned the disutility parameters, I would like to mention a few points. For each mode, the following parameters are defined:

marginalUtilityOfDistance_util_m
marginalUtilityOfTraveling_util_hr

Before starting the calibration of my model, I was unsure about the appropriate values for these parameters. I reviewed several scenario configurations available on GitHub and noticed that in many cases, these parameters were left unchanged or set to zero. Based on that, I also left them unchanged.

However, I’ve always had concerns about how to represent the discomfort associated with different transport modes, such as biking or public transit. Now that you’ve pointed this out, I understand it's important to incorporate these aspects more accurately.

Since this topic is a bit confusing to me, I would really appreciate your guidance. Based on the parameters I shared in my first message, do you have any suggestions on appropriate values? Or could you point me to helpful resources or example scenarios that might serve as a reference?

Additionally, I currently have the following parameters unchanged. Do you think any adjustments are necessary?

Thank you in advance for your support.

[mrieser]

Hi @ArezHK
I've written down how I think about the scoring parameters in the attached slides. Hopefully it helps to set suitable parameters in your case.

MATSim_Scoring.pdf

[ArezHK]

Dear @mrieser ,

Thank you for your response and for sharing the helpful data.

I tried connecting that data with the parameters used in the Kelheim and Leipzig scenarios. Here's what I found:

In Kelheim 3.1:
marginalUtilityOfTraveling_util_hr for car, pt, drt and walk are zero and for bike is -3 and for ride is -12, additionaly waitingPt is -1.6

In Leipzig 1.1 scenario:
marginalUtilityOfTraveling_util_hr for car is -0.5
marginalUtilityOfTraveling_util_hr for pt is -2.0
marginalUtilityOfTraveling_util_hr for bike is -10.0
marginalUtilityOfTraveling_util_hr for walk is -1.5
marginalUtilityOfTraveling_util_hr for drt is -2.0

And in Berlin scenario these values are almost all set to zero.

After reviewing all these values and your instructions, I still couldn’t identify a clear or logical method for setting realistic discomfort (disutility) parameter numbers. Are these values typically chosen based on the modeler's assumptions, or is there a specific criterion—for example, assuming that public transport is two times less pleasant than a door-to-door DRT service?

Also, one point I found a bit confusing was the diagram you shared in the PDF file. Intuitively, I would expect the discomfort of biking to be higher than that of driving a car—so the marginalUtilityOfTraveling_util_hr for bike should be a lower value than for car (i.e., more negative). But in the diagram you provided, it appears to be different: the multiplier for bike is -6 (F(x) = -6x) and for car it's -2 (F(x) = -2x). Am I understanding this correctly, or might I be misinterpreting the diagram?

I appreciate your insights in advance and thank you for your time and support.