Would it be possible to account for both long-term and short-term (recurring) uncertainty in stochastic TIMES?
One possible implementation (though I do not have the complete knowledge of the code) would be to allow the user to define the stage after which the capacity-related variables are fixed, which can be >=1, via a $ option. By default, the stage after which the capacity related variables are fixed, could be set to 1.
Do you think that this does make sense? Is it too difficult to implement?
In the attached image are three event trees: the first considers long-term uncertainties, the second short-term recurring uncertainties and the third is the concept that I am suggesting.
Thanks a lot, Vangelis
PS. Apologies for writing to this forum but I do not have sufficient permissions to create new topic to the TIMES forum.
Vangelis Wrote:I do not have sufficient permissions to create new topic to the TIMES forum.
That would be very strange. Please contact Amit about it.
Small improvements to TIMES can indeed be requested directly on the ETSAP Forum, but larger enhancement proposals, like this apparently is, should be submitted the ETSAP Project Head for consideration. What you propose would seem to require quite a lot of redesign and rewriting large parts of the code to support the generalization.
As you may know, the support for modeling recurring uncertainties ("SPINES") is an experimental feature, which was implemented because there was interest for it and the implementation could be done without too much effort. However, even for that an approval from the project head was required beforehand.
To start I would like to mention that a paper has been published were SPINES was used for recurring uncertainties. "Short-term uncertainty in long-term energy system models — A case study of wind power in Denmark", from P. Seljom and A. Tomasgard.
Regarding your question I understand you want to run a model were some of the inputs are uncertain in the long run and some of the inputs are "variable" or recurring.
The user has with the current code the option to use the standard stochastic approach or SPINES. In the SPINES methodology, all capacity related variables are independent from the State Of the World. In other words, they are not dependent on the uncertain events. This is useful when one want to analyse variability such as fuel prices or renewable availabilities.
However ,I do not see a reason why you cannot build an event tree that has a mix of both long term and short term uncertainties and use the standard stochastic approach ?
Basically, my question is why you would want the capacity related variables to be fixed at some point ?
It would be interesting if you could further explain the event tree with a specific example.
Dear Antti and dear Wouter, Thank you very much both for your reply. I don't want to overtake this forum so I will be short in my replies:
@Antti: You are right, now I saw also the complexity, perhaps if there is some interest from other ETSAP members to model something like this we can submit a proposal to the ETSAP Project Head.
@Wouter: Thank you very much for the paper it is very interesting (I was not aware about it). In the paper the capacity related variables are fixed in the 2nd stage to the values of the 1st stage (SPINES option) to consider uncertainty regarding the operational decision. My wish was to be able to extend the 1st stage to be a stochastic tree accounting for e.g. demand uncertainty having let's say N stages and then once the investments were there for each demand path up to the Nth stage the operational uncertainty is revealed. So, the investments in the N+1 stage should be fixed to the investment of the previous stage, which in this case is not the 1st. It is quite complex, as Antti wrote. If there is need for something like this from you or from other members I would be happy to submit a joint proposal.
I got confused with the terminology used in the paper and in the trees you uploaded. What was puzzling me is the statement that the capacity related variables of STAGE 2 are fixed to STAGE 1. Now I realize this refers to the mathematical formulation of the two-stage stochastic problems with two distinct stages. Now, when there is a limited number of scenarios, two-stage stochastic problems can be modeled as one big LP problem, as is done in TIMES. As written in the paper, the two stages are optimized simultaneously.
The confusing element is that in the TIMES code, a Stage does not refer to a certain level of problem solving but simply refers to the time where the tree splits into two or more branches or SOWs (States Of the World). In other words, even when you use SPINES, a capacity related variable in stage 2 can differ from stage 1, for example when an investment is done. In TIMES terminology, saying that "the capacity related variables are fixed to the previous stage" is similar to saying "the capacity related variables are the same for each SOW within a stage."
This helped me to understand what you want to do. Using the meaning of Stages and SOWs from TIMES, I think following example can help. Assume there is a problem where in Stage 2 you have 6 SOWs (arbitrary example). SOW 1,2 and 3 could refer to a long term uncertainty, such as a high demand. SOW 4, 5 and 6 could refer to a world with low demand. The difference between SOW1-2-3 and SOW 4-5-6 could refer to recurring short term uncertainties like high, medium and low oil prices. If I understand correctly, you would like the capacity related variables to be fixed within the group SOW1-2-3 and withing the group SOW4-5-6. The model user could solve this maybe by creating constraints that force the capacities to be equal within a group of SOWs. However currently, I do not think we can define user constraints that involve the SOW dimension.
As Antti wrote before this would require some reformulation of TIMES code, which is a quite complex task and can be done only in the context of a proposal to ETSAP Head. So if you are also interested in something like that then we can submit a proposal to the next call.
I think combining both short- and long-term uncertainty will be limited to relative small models due to solution time. What is your opinion on this? I am personally finding it challenging to solve models using only short-term uncertainty. As I understand, the solver does not apply any stochastic decomposition techniques...
To start I would like to mention that a paper has been published were SPINES was used for recurring uncertainties. "Short-term uncertainty in long-term energy system models — A case study of wind power in Denmark", from P. Seljom and A. Tomasgard.
Regarding your question I understand you want to run a model were some of the inputs are uncertain in the long run and some of the inputs are "variable" or recurring.
The user has with the current code the option to use the standard stochastic approach or SPINES. In the SPINES methodology, all capacity related variables are independent from the State Of the World. In other words, they are not dependent on the uncertain events. This is useful when one want to analyse variability such as fuel prices or renewable availabilities.
However ,I do not see a reason why you cannot build an event tree that has a mix of both long term and short term uncertainties and use the standard stochastic approach ?
Basically, my question is why you would want the capacity related variables to be fixed at some point ?
It would be interesting if you could further explain the event tree with a specific example.
Thanks,
Wouter
Dear Wouter,
I am using the document authored by you about integrating recurring uncertainties in TIMES (Congratulations by the way, it is brilliant).
I am using the portoflio theory approach to model recurring uncertainties of fuel prices and investment cost of largy hydropower technologies. The Montecarlo simulations and all the auxiliary dummy flows have been created, however the problema comes when I introduce negative upabsdev commodity flows (when the real cost is lower than the expected cost). The model starts operating weird and does not otpimize properly. It seems that the model doesn't process the negative financial commodity flows......is there a workaround for this? In the user manual you use negative values for the upsdev scenarios (see attached snapshot).
