We are facing some trouble with the time-stepped algorithm.
Indeed, we are running a constrained scenario (net emissions to be 0 in 2050 in Europe by 2050) under myopic foresight. Two windows of optimization are used (20 years and 10 years) and are compared to the perfect foresight model.
The problem is that under the most constrained scenario regarding the myopic foresight (10 years foresight), the demand of some end use sectors becomes very high than the input values (the attached figure is a result of the processing of the three scenario). the only change between the three scenarios are the steps of the time-stepped algorithm and the overlapping years (set to half the value of the steps).
we don't know why the demand diverges from our inputs.
Do you have any advice for me at this situation as we appear to be all out of ideas ?
Best regards,
17-05-2020, 08:06 PM (This post was last modified: 17-05-2020, 08:08 PM by Antti-L.)
Well, without seeing the model it is very hard to say what is causing such a behaviour. Apparently it is the least cost solution satisfying your constraints after the previous solution step, but the artificial increases in the demand commodities indicate that your constraints may be tightening too quickly for the model to be able to adapt to it, with so limited foresight. There could be many reasons for such rigidity in the model, but for example, if you have demand technologies with fixed AFs, I guess that the investments into those technologies in the previous periods (around 2040) might make it almost impossible to achieve large subsequent changes in the very short term, which your constraints for 2050 may require. Why your constraints are then satisfied by increasing the demands is not clear to me, if your only relevant policy constraint is "net emissions to be 0 in 2050 in Europe by 2050". But if you have other constraints, such as imposing a high renewable share, they might well explain it. Did you try 10 years foresight without the emissions constraint and did that work without such a problem?
overproduction of demand can happen if there are share constraints that can't be met otherwise. Maybe you can allow the techs in this segment to retire.
18-05-2020, 02:04 PM (This post was last modified: 18-05-2020, 02:39 PM by canismajoris.)
Thank you for your answers.
Yes I've already tried the times stepped algorithm with a less constrained scenario ; a little hard to interpret the results but still it seems to be working. The objective of using the time stepped algorithm in our study is to see if the model invests in more conventional sources in the short and medium term (2030 typically) and in which sectors given the constraints specified.
The model we are working on is the JET Model (https://ec.europa.eu/jrc/en/scientific-t...chnologies).
as a first attempt, I allowed early retirement for all the technologies of the model (attached Excel file), but demands are still over produced in 2050.
Even with less myopic foresight (20 years foresight), total final consumption of the industry sector is more than is higher than in the perfect foresight (the difference is almost 10% of the overall consumption of industry).
I am not sure if the way I allowed early retirement is correct or not.
Could you please enlighten me on that ?
PS: PF_MF10 is the difference between the myopic foresight 10 and the perfect foresight.
18-05-2020, 03:28 PM (This post was last modified: 18-05-2020, 03:59 PM by Antti-L.)
Yes, the time-stepped algorithm simulates "short-sighted" decision-making, and 10 years is very short-sighted. Therefore, I think it is not surprising that you may get a notably larger final energy consumption than under perfect foresight. So, if you have a 10% difference in industrial final consumption, it would not sound unreasonable to me.
But concerning the final demands, it is interesting if you are still getting any notable increases compared to the COM_PROJ projections, if you let all technologies retire before the end of their lifetime. However, you did not show any results concerning the demands now, only for final energy.
Are you perhaps using elastic demands as well in the time-stepped runs? If you are, that might well be causing problems ...
Anyway, the time-stepped algorithm does not add any new constraints into the model, it basically only fixes in each step the variable levels of earlier periods to the solution in the previous steps, and then solves the model again with the horizon extended according to the timestep.
18-05-2020, 04:06 PM (This post was last modified: 18-05-2020, 07:14 PM by canismajoris.)
Unfortunately, I am still getting notable increases, mainly for bunkers, non residential street lightning, non residential multimedia and the pulp and paper industry. The other thing that surprises me is that the increase in TFC (mainly in industry) does not match the increase in demand which is mainly caused by the increase in the transport sector.
Sorry for not including demands. Please find them attached to this post.
I tried with/without elasticity and the results are the same.
18-05-2020, 08:39 PM (This post was last modified: 18-05-2020, 08:41 PM by Antti-L.)
Very interesting. If the increase in the "demands" (actually not the demands but the production of the demand commodities, e.g. TBunk, if I am reading the chart correctly) is not reflected in the final energy, how is that large increase in the supply then produced? In other words, which processes are then producing the increasing demand supply, without causing corresponding increases in the final energy? And, if it is costly (as one would expect), which model constraints are causing those large increases?
If I am reading the chart correctly, the TBunk supply is increasing by 3 × 106 PJ = 3000 exajoules, is that correct?! That should certainly be reflected in the final energy or at least in the primary energy (if bunkers are not counted in your final energy). I guess the total world primary energy is today about 600 EJ.
