Different solution in TIMES versions
#1
Dear Antti,

I am developing TIMES model and recently I have faced a problem of different solution of the model in two versions of TIMES source code. All TIMES versions up to v430 gives one solution and others starting from v431 - another. And I found that caused by new version of the file "fillvint.gms" in version v431. So when I run the model with TIMES code v431 but with the "fillvint.gms" file from v430, the model gives same results. But I didn't have any idea why this happens.

Could you help me with this issue? 

Thanks and best regards,
Bakytzhan
Reply
#2
Hmm... unless you have used the VINTOPT switch, on the basis of a quick look I don't see a reason for such difference.

If you have not used the VINTOPT switch, could you provide me the model input files (*.DD and *.RUN) to investigate the issue (privately if you prefer, e.g. by sending me a Dropbox/Google drive link in a private message)?

Thanks.

[EDIT:] I have now tested with several models, whether using the "fillvint.gms" file from v430 would make a difference, but I didn't obtain any different results. Therefore, your help by providing a reproducible test case would be appreciated.
Reply
#3
Thanks for providing the model input files.

I was able to reproduce the difference, which was a rather delicate issue:

In the older version, while interpolating the cost attributes, TIMES added the vintage year 2016 for the following technologies: EBIG20, EWOD20, EHCO_SCPC20. This addition was done just to have one preceding data point for the time-series. Because the cost attributes NCAP_FOM / ACT_COST had an interpolation option 5, the costs were therefore set to zero in 2016. The resulting densely interpolated/extrapolated costs thus increased from zero to the 2020 value between the years 2016 and 2020.

In the newer version, while interpolating the cost attributes, TIMES only added the year 2016 for those technologies into a temporary vintage set (for the same purpose as above), and only while interpolating flow-related attributes. Therefore the costs were not set to zero in 2016, because the technologies had START=2020. The resulting densely interpolated/extrapolated costs were thus extrapolated backwards from the 2020 value, and so the 2020 value was set between the years 2016 and 2020.

The issue was thus caused by the interpolation option 5 being used for some cost attributes of vintaged technologies that also had START=2020. I believe the new version can be considered correct, and, and more importantly, I think your model did not work correctly with the earlier version, due to those strange interpolation options.  Can you explain why you had specified those interpolation options for the NCAP_FOM of EBIG20 and EWOD20, and for the ACT_COST of EHCO_SCPC20?  What was the intended purpose of them?  

By default, all cost attributes are fully interpolate/extrapolated, which should be in most cases appropriate.
Reply
#4
Dear Antti,

Thanks a lot for your help.

When assigning this interpolation rule I forgot about this point about default value. That's why there was interpolation rule "5" with start in 2020. I will remove this interpolation rules "5", this will resolve my problem.

So if I understand correctly it was the problem of not being interpolated to 2016 in combination with "ILED", it was cost free to invest in it earlier than 2020, despite starting point "2020"?

Thank you very much for your help,
Bakytzhan
Reply
#5
No, you didn't understand it correctly. It was certainly not cost-free to invest in it earlier than 2020, despite starting point "2020".  The START=2020 was by no means getting ignored. It was not possible to invest before period 2020 in either case.

But the interpolation between 2016 and 2020 affects the costs of the 2020 vintage.  The beginning year of the period 2020 was 2018, and with ILED=1 the costs would be taken from 2019!  And your interpolation option 5 caused the costs in 2019 to be too low.

And for the activity costs similarly:  As the period 2020 is from 2018 to 2022, the ACT_COST interpolation between 2016 and 2020 affects the costs in 2018 and 2019.  

But thanks anyway for the report; it was good to see what the issue was.
Reply


Forum Jump:


Users browsing this thread: 2 Guest(s)