Stochastic with SOW on limited variables

[wnijs]

Dear all,

Is it possible in TIMES to have an extra dimension "SOW" on a limited set of variables ?

If I understand correctly, TIMES allows us today to have a split in "before" and "after" a certain time period. Before this period, TIMES creates variables that are similar for all SOW. After this period, all variables are SOW specific.

I wonder if it is possible to have a split in the type of variable rather than based on the period. When for example having variable costs SOW specific, the model could choose optimal investment strategy for all these SOW.

An example in Gams shows what i mean (modified from an existing model):

http://dl.dropbox.com/u/18523560/Powerplants.gms

Thanks

WOuter

[Antti-L]

The stochastic mode of TIMES has been designed to support only event trees where the uncertainties will be resolved at certain periods in the future, i.e. hedging against long-term uncertainties. 

But I guess it could be extended to support also the handling of short term uncertainties within each period, if that's what you mean? For example, do you think one could perhaps use such an approach for analysing the impact of uncertainties in wind power conditions within each period, on the optimal power plant investments?

If you think that such a new way of running stochastic programs would be useful in TIMES, I encourage you to make the full design for it in writing (in conceptual and mathematical terms), and then submit the proposal to the ETSAP project head to be considered for implementation. Or, you could implement it yourself, and make it available for all ETSAP members.

[wnijs]

Dear Antti, you point out exactly what I mean. There is a link with the work of KUL and VITO on fuel prices variation and risk aversion, presented last week. I also know that IFE is interested in a way to incorporate uncertainties in wind power conditions. So, yes the short term uncertainties could maybe have a large impact on investment choices. First, I will think about the concept and see if I can implement it myself. My idea untill now is that I make only a very limited set of variables SOW specific. What might work is to put the "resolution time" upfront (basic year or close) and than to add constraints so that all variables are equally treated over the SOW, except the ones envisaged. I keep you informed. Wouter

[Antti-L]

Dear Wouter,

The Stanford presentations have just been uploaded at the ETSAP site, and I had a look at your presentation on the fuel price variation analysis. It was quite interesting, although I didn't fully understand all the details of the formulation. The presentation seemed to imply that you have already implemented the approach into TIMES, and so it would be nice to have a closer look at it, and see if the approach could be incorporated into the standard code.

I think that at least an option of restricting of all the VAR_NCAP and VAR_CAP variables to a single SOW (SOW=1) under the stochastic mode could indeed be implemented into the common TIMES code. Consequently, also the objective function components related to capacity costs (OBJINV, OBJFIX, OBJSALV) would then also require only a single SOW. I guess such an option might already make it possible to analyse hedging against shorter-term uncertainties, which are recurring by nature, and in that sense basically remain unresolved throughout the model horizon? If there is general interest (and approval by the ETSAP Project Head), I might be able to include such an option in the code by September.  If other users have any comments on this, please post here or on the ETSAP Forum.

[Pernille.S]

Dear Antti,

As Wouter has mentioned before, IFE is very interested in including stochastic of short term uncertainties to the TIMES code. We are particularly interested to use the availability of wind as a stochastic parameter. For us it would be great to get this feature!

 

Pernille

[UCL_Will]

I've just had a quick discussion about this with some colleagues.  We have some reservations as to whether implementing aleatory [inherently random] uncertainties into TIMES would actually bring added value to the model results.

The largest effect of the random variation in wind speed is whether one can cover peak demand with a given penetration of intermittent renewables.  This then filters through to the optimal investment in back-up capacity.  Given that a system operator will wish to reduce the Loss of load probability to a very low value, the back-up capacity needed to cover will be enough for the worst case (high demand, low wind availability) or for the 90th percentile of cases.  This is treated probabilistically in most situations, but a conservative approach that does not require stochastics should be sufficient.

Other aleatory uncertainties, such as fuel price volatility, can be expressed through increasing hurdle rates for such technologies (the risk premium of the investment is included in the required return).

I'd be interested to hear of the application of these short term uncertainties within TIMES that you're thinking of implementing.

Regards from sunny London.

Will

[wnijs]

Thanks Antti, I am very pleased that the methodology is of interest to you and that it can be spread to other ETSAP participants. What is the status of this ?

 

It is good if I also mention that two approaches have been worked out but some work still need to be done. A first approach is what I would like to call the “non adaptive” approach and the second is the “adaptive” approach.

 

In the non-adaptive approach, a “portfolio like approach” was implemented  into TIMES, together with Denise van Regemorter, Wim Benoot and Joris Morbee. This enables the modeller to put an extra cost on upward variation of a combination of costs. It is to include risk aversion and the results of the Stanford presentation were produced with this method. It is implemented in TIMES via Excel files, but not into the TIMES code. In this approach,  both the investments and the usage of these investments (fuel use for example) are fix for the total horizon and thus non-adaptive.

 

In the second approach,  I wanted to include adaptiveness (as in real options theory), based on the difference fix/variable as explained above. I have not implemented this into TIMES yet, but I was planning. So, yes it would be great to implement it since it could help models with variable conditions like rain, wind, solar and prices. The method you suggest seems OK and in line with my suggestion to treat some variables equally (I think VAR_NCAP and VAR_CAP are sufficient). Risk aversion also can be used into this adaptive approach with the already existing DEVUP on stochastic.

 

More time is needed (or some ETSAP funding)  for a better description, for working out the ideas and to get an overview of the PROs and CONs.

To be continued.

Wouter

[Antti-L]

Thanks for the follow-up, Wouter.

I have now asked the ETSAP Project Head whether I could include the new option in the next release of TIMES, but my offer was rejected. The Project Head said that he considers it best that other TIMES modellers contribute to the debate and make proposals for the design and eventual implementation. He also suggested that for the modeling of the stochastic behaviour of intermittent renewables, the first step should probably be to add a new stochastic dimension "State of the Weather" for a set of VAR_ACT or VAR_FLO.

Consequently, as I didn't get the approval, I cannot make the new option available for other ETSAP members.  I also don't see the point in implementing a new "State of the Weather" dimension for TIMES variables, and will thus leave the implementation to other partners.

However, if you can provide a more complete description at some point, I would certainly be interested to look at it.

[Antti-L]

Status Update:
After some discussions, the ETSAP Project Head has now decided to approve the inclusion of the new option in the common code. So, I will make it available in the next main Release of TIMES, before October. The new option will thus make it possible to activate the SOW index only for the non-capacity-related variables. With another sub-option you would also be able to make any remaining cumulative and dynamic equations "SOW-neutral", thereby ensuring independence of recurring uncertainties in successive periods.