Add MDVs Vehicle Type

[mkmahajan]

We currently have trouble matching historical data on gasoline vs. diesel consumption in freight HDVs. I looked into this and found that according to the AEO, gasoline freight HDVs travel about 50% more per year than diesel vehicles, but both are assigned the same BAADTbVT value in the EPS. Subscripting this variable by vehicle technology should fix the issue. I have not checked on this yet, but I suspect AEO may also show differences between the amount traveled by electric and gasoline LDVs, for example.

If this edit makes sense, I'm happy to try implementing it before having Jeff review.

[jrissman]

Yes, this is straightforward to do. Megan, please go ahead and make the subscripting changes, if you like.

Note that the "develop" branch currently has a non-running version of the model in it, but you should always test your changes in a running copy of the model, both to catch errors and to make sure the new feature resolved whatever issue you were seeing. BAADTbVT was updated in "develop," so there is no version of the model right now that has the latest BAADTbVT data and that runs. You may choose to wait until the AEO 2021 data update is done and the model runs again, or you could copy the updated BAADTbVT file from develop into a branch made from the end of #149 (the last running version of EPS.mdl), make sure it works, and then merge into "develop."

[mkmahajan]

I have been working on this issue in branch #150, and it has turned out to be a bit more complicated than I originally expected. I've updated BAADTbVT by taking VMT in AEO and dividing by vehicle stock for passenger LDVs, freight LDVs, and freight HDVs (the only vehicle types where AEO provides enough data to do this calculation). I then subscripted BAADTbVT and also updated the cargo distance allocation. The model had previously allocated cargo distance to vehicle technologies by the share of vehicle stock, but I added in a new variable to help allocate by annual distance traveled instead, which means we match fuel consumption more closely.

However, there is a remaining issue because we no longer exactly match AEO on BAU cargo distance traveled. That's because the EPS first calculates the number of new vehicles needed, then feeds that into the new logit function to figure out which vehicles are deployed. Since the different vehicle technologies have different BAADTbVT values, the resulting cargo distance may vary based on which vehicles are cheaper to deploy. For example, the average annual distance traveled by electric freight HDVs is about half the average annual distance traveled by diesel freight HDVs (because they're mostly Class 3 or 4 vehicles as opposed to the Classes 7-8, which are predominately diesel and travel much further on average). Robbie suggested we could feed total cargo distance into the logit function instead, then determine new vehicles based on the cost per mile. But this isn't actually representing what we want either - for example, if we put on an EV mandate for HDVs, we'd likely want to represent 1 Class 7/8 diesel vehicle being replaced by 1 Class 7/8 electric vehicle, rather than 1 Class 7/8 diesel vehicle being replaced by 2 Class 7/8 electric vehicles that each travel half the annual distance of an average diesel vehicle.

One thing that may help with some of what we're seeing is to move vehicle Classes 3-6 into freight LDVs and to leave the freight HDV category as only Classes 7-8. Since Classes 3-6 have travel patterns that are more similar to freight LDVs, this should make the delta between the average annual distance traveled within each vehicle type smaller. Even so, there are still noticeable differences between the average distance traveled by vehicle technologies. For example, the AEO shows electric and natural gas light commercial trucks actually travel more on average than gasoline or diesel light commercial trucks. @jrissman - before doing anything else on #150, I wanted to check in to see how you might approach the issues noted here. It may be that we should just try reallocating Classes 3-6 and tabling the structural updates, or perhaps there are additional structural elements we could add.

[jrissman]

I see the issue.

I do not think that using present-day BAADTvBT values to represent vehicle electrification is a good idea. I agree with you when you say:

if we put on an EV mandate for HDVs, we'd likely want to represent 1 Class 7/8 diesel vehicle being replaced by 1 Class 7/8 electric vehicle, rather than 1 Class 7/8 diesel vehicle being replaced by 2 Class 7/8 electric vehicles that each travel half the annual distance of an average diesel vehicle.

The whole point of vehicle electrification is to replace fossil fuel-using vehicles with electric equivalents, performing essentially the same duties. If you find that today, electric trucks travel only half as far as diesel trucks, and you allocate cargo distance to vehicle technologies, you'd end up finding you need twice as many electric trucks to meet the same demand as diesel trucks do today, which would be much higher-cost than buying one longer-haul EV. A major cost would be paying two drivers' salaries and benefits, which is a cost the EPS isn't capturing, so a per-mile cost calculation for two EV trucks vs. one diesel truck would miss a huge component of the cost increase in running two trucks. As you say, that isn't what we're trying to model. We expect that future electric trucks - the kind that will replace long-haul diesel trucks - would be capable of performing as well as the diesel trucks they replace, which means achieving similar BAADTvBT values, and hence one truck and one driver can do the job. So this is a case where historical BAADTvBT data by vehicle technology type is not a good guide to the annual distance driven by EVs deployed by the model.

This also might lead to problems - there is tricky calibration going on in the transport sector, and the procedure with trans/SoCDTtiNTY is a refined version that tries to make what were very difficult problems (in earlier EPS versions) tractable. I'm worried that breaking these assumptions around BAADTvBT may also break something there.

I the changes you are looking at may be overkill for the problem you are looking to solve. Is it simply that you get a different balance of gasoline consumption vs. diesel consumption in year 1 than transportation fuel data show? That doesn't have anything to do with the way the model picks what to deploy in years later than 1, which I think we should not touch, except to tweak to the Logit function's time-series shareweights to limit deployment of technologies until they are mature enough to handle the need the vehicle buyer (e.g. trucking company) is looking for.

Yes, you can base the boundary line between "freight HDVs" and "freight LDVs" on duty cycle. The model should allow you to freely assign whichever of the 8 truck classes into "freight HDV" and "freight LDV" categories. The model doesn't know anything about truck classes and there is no special reason why the line between these two categories has to be in any particular place with respect to truck weight/size. I think it makes sense for you to put this division wherever you find it more useful for distinguishing between the types of trucks that have certain characteristics, including BAADTvBT, but not limited to that variable, because you also have to consider where it makes sense to put the dividing line in terms of Average Vehicle Loading AVLo.

Within the Freight LDVs category, you write: "Even so, there are still noticeable differences between the average distance traveled by vehicle technologies. For example, the AEO shows electric and natural gas light commercial trucks actually travel more on average than gasoline or diesel light commercial trucks." As with freight HDVs, I think this is a historical artifact and is not necessarily something we believe will be true of future electric freight LDVs versus gas or diesel freight LDVs. As we expect EVs to take over more and more trucking roles, the ways in which they are used should come to resemble the ways diesel and gas trucks are used today. So as with the HDV case above, I think we should keep the current approach and not assume that these historical differences will be reflected in the vehicles the model deploys in the future.

Maybe picking the dividing line between "freight HDVs" and "freight LDVs" carefully will result in a good enough fit between gasoline and diesel fuel consumption in year 1.

I think the way to get this better would be to understand the roles the vehicles are being used in. For instance, imagine a new subscript, "Vehicle Role," that has elements such as "Long-Haul Transport," "Intra-City Deliveries," and so one for each major use case for trucks. BAADTvBT is determined not be the vehicle's technology, but by the vehicle's role. Then we have a certain demand for each role to be performed, and vehicles compete separately for each role. Maybe EVs have a lower shareweight in the "long-haul transport" role in early years due to battery size limitations, and we relax that restriction later. But maybe EVs have a high shareweight in "intra-city deliveries" because the short distances and stop-and-go traffic play to EVs' strengths.

This gets at the fundamental reason why we see differences in BAADTvBT by vehicle technology. It's not intrinsic to the technology - it's based on the role in which the truck is used. Accordingly, if the vehicles are competing to fill demand for certain roles, instead of just competing to fill demand for generic freight ton-miles and passenger-miles, we could simulate this better, using an extension of the current structure with the new subscript. This might benefit our ability to closely match some other vehicle types - freight ships come to mind, where we have an exceptionally wide range of vehicle roles being grouped together (say, ocean-going tankers and tugboats).

Adding a new "Vehicle Role" subscript would be a big task and not in-scope for 3.2, or necessarily 3.3. I'm not sure it's worth it for the benefits. But it's the thing that comes to mind as a way to better capture what's actually happening, and what's actually driving these differences in vehicle usage.

[jrissman]

Sorry, I erroneously included some text at the end of the earlier comment that I meant to delete prior to posting the comment. I've deleted the extraneous text on GitHub, so please view my comment in the GitHub thread in your web browser, not in your email program.

[jrissman]

Perhaps Vehicle Type could become Vehicle Role. It would be easier than adding an additional subscript, and I'm not sure we'd need Vehicle Type anymore, because Vehicle Role would be more disaggregated and might effectively imply given Vehicle Types.

[robbieorvis]

Thanks, Jeff. I propose we just do the grouping differently for now as suggested, with Light and Light-Medium going into LDV Freight and Heavy going into HDV Freight. I think that will mostly (though not fully) solve the issue.

I am not proposing at all we do this, but the way this is handled in other models I’ve seen is to group trucks by class or weight, which is effectively the same as the role you outlined. Trucks tend to be used within a class/weight category similarly, and so their distance should be similar.

I think for now and perhaps forever we can just be careful how we categorize input data without having to modify the structure.

---

Robbie Orvis Director of Energy Policy Design Phone: 415-799-2171 98 Battery Street, Suite 202 San Francisco, CA 94111 www.energyinnovation.orghttp://www.energyinnovation.org/ @.***D71FC8.7BF6A180]

---

Check out our book, Designing Climate Solutions: A Policy Guide for Low-Carbon Energyhttps://www.amazon.com/Designing-Climate-Solutions-Policy-Low-Carbon/dp/1610919564 Available wherever books are sold

[Policy Design book cover]

From: Jeff Rissman @.> Sent: Monday, March 22, 2021 5:30 PM To: Energy-Innovation/eps-us @.> Cc: Subscribed @.***> Subject: Re: [Energy-Innovation/eps-us] Subscript BAADTbVT by vehicle technology (#150)

Perhaps Vehicle Type could become Vehicle Role. It would be easier than adding an additional subscript, and I'm not sure we'd need Vehicle Type anymore, because Vehicle Role would be more disaggregated and might effectively imply given Vehicle Types.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/Energy-Innovation/eps-us/issues/150#issuecomment-804408262, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AK5N6SJATETQPIGADRFDEULTE6ZD5ANCNFSM4Y2HFHLQ.

[mkmahajan]

This sounds good. I'll work on the data fix in the develop branch today, but leave this issue open in case we want to revisit in the future. I'm removing the 3.2 tag for this issue.

[jrissman]

Sounds good. I renamed this issue "Add MDVs Vehicle Type" under the assumption that going from two vehicle classes to three vehicle classes would be enough. Splitting it into too many categories is undesirable because the demand for each vehicle type is completely isolated from other types in the EPS. For instance, just as you can't use a truck to do the job of an airplane, you wouldn't be able to use an MDV to do the job of an HDV. I think that's probably desirable with three vehicle classes, but if we had eight classes of trucks, it might become a poor assumption, because I think there are situations where you could use a Class 6 truck to do the same work as a Class 5 truck, and giving them distinct properties like different BAADTbVT values wouldn't make sense. Also, collecting the data becomes more miserable the more truck classes we break things into. So I think a three-way split into LDVs/MDVs/HDVs provides a good balance that offers better accuracy than a two-way split, while avoiding the two downsides I mentioned that happen if you split things into too many weight classes.

I agree with not adding MDVs for 3.2 but it can be considered for future model versions.