Download latest Version of VEDA-FE 4.5.8. The Change Log for this version.

Veda Application Installation guide


TIMES stochastic: help!
#1
Hi there!

I'm attempting to run TIMES in stochastic mode for the first time and whilst I've had some limited success, I'm really struggling to achieve the results I had hoped to obtain - a minimax regret solution.  I found the following in the etsap forum:


Quote:applying the Minimax criterion requires the following steps:

1.      Solve the deterministic perfect information problem for all sow, obtaining the objective values M(sow) ;

2.      Solve the Minimax problem, minimizing the worst regret (difference from the perfect information solution), obtaining the optimal hedging strategy;

3.      Solve the deterministic problem for all sow again, fixing the initial periods to the optimal hedging strategy.

The first (1) step above can be done by using the sensitivity analysis facility of TIMES. The "middle" step (2) above can be carried out by defining the following parameters for the stochastic problem:

S_UC_RHS('OBJ1','FX','2',sow) = M(sow)



I've been able to produce credible results for step 1 and have obtained the objective values for each state of the world.  However, I'm stuck at step 2 in that I don't know how or where to define the last statement "S_UC_RHS('OBJ1','FX','2',sow) = M(sow)".  Does anyone have an example they could share?
Reply
#2
Welcome to the VEDA Forum.

The savage Step 2 means running a stochastic problem, and for running it you should thus:

1) Define all the parameters needed for setting up the stochastic problem;
2) Define S_UC_RHS('OBJ1','FX','2',sow) = M(sow) (from step 1);
3) Include in the run the scenario(s) where you have defined the above;
4) Activate stochastic mode in the VEDA control panel.

For an example stochastic scenario, see the attachment in http://forum.kanors-emr.org/showthread.php?tid=360
For S_UC_RHS, OBJ1 would go to Other_Indexes, FX into LimType, 2 into Stage, and sow into SOW.

Important remark: The problem of step 2 has an unfortunate characteristic: since all that matters when computing the MMR strategy is indeed the value of the Minimax Regret, all other regrets are left free to take any values, as long as these values remain below the MMR. This remark is equivalent to saying that the problem is highly degenerate.
Reply
#3
Thanks Antti-L. Unfortunately, I think I may have misinterpreted your advice above as when I attempt to import this new scenario for step 2, I receive a "scenario with no data" error message. Can you confirm that once the first step has been performed and run successfully, the values of OBJ1 and SOW should be available to VEDA_FE? and don't need to be manually defined?

Thanks for your patience!
Reply
#4
See:
http://iea-etsap.org/forum/showthread.php?tid=65

The values M(sow) are obtained from Step 1.
For Step 2, you need to define S_UC_RHS('OBJ1','FX','2',sow) = M(sow).
There is no automatic way of defining these values, and so you have to enter the numerical values yourself. But because you have the Step 1 results in VEDA-BE, copying the values M(sow) to the Scenario file shouldn't be too difficult?

Quote from the earlier post:
As you can see, even if the formulation needed in step (2) is supported by TIMES, using the Minimax criterion requires considerable manual effort.

However, I have no idea why you get a "scenario with no data" error. You should be defining not only the S_UC_RHS values, but also the SW_START, SW_SUBS and SW_SPROB/SW_PROB, as well as the uncertain parameters for your stochastic problem. Do you have all these in the "scenario with no data"?

[EDIT:] 'OBJ1' is a symbol, and so you just enter the string OBJ1 into the Other_Indexes column. SOW is the set of states of the world, which are numbered 1,...,N. You already stated that you have obtained the objective values for each state of the world, and so you already know N and you know the values M(sow), sow=1,...,N.
Reply
#5
(17-11-2017, 10:28 PM)Do you happen to have published work, using this model formulation, for me to read? It would be interesting to get a better understanding on the motivation behind this approach. Pernilleandrewmortimer Wrote: Hi there!

I'm attempting to run TIMES in stochastic mode for the first time and whilst I've had some limited success, I'm really struggling to achieve the results I had hoped to obtain - a minimax regret solution.  I found the following in the etsap forum:


Quote:applying the Minimax criterion requires the following steps:

1.      Solve the deterministic perfect information problem for all sow, obtaining the objective values M(sow) ;

2.      Solve the Minimax problem, minimizing the worst regret (difference from the perfect information solution), obtaining the optimal hedging strategy;

3.      Solve the deterministic problem for all sow again, fixing the initial periods to the optimal hedging strategy.

The first (1) step above can be done by using the sensitivity analysis facility of TIMES. The "middle" step (2) above can be carried out by defining the following parameters for the stochastic problem:

S_UC_RHS('OBJ1','FX','2',sow) = M(sow)



I've been able to produce credible results for step 1 and have obtained the objective values for each state of the world.  However, I'm stuck at step 2 in that I don't know how or where to define the last statement "S_UC_RHS('OBJ1','FX','2',sow) = M(sow)".  Does anyone have an example they could share?

Reply
#6
The minimax regret approach is to minimize the worst-case regret.

Published work, using this model formulation:

Loulou R. & Kanudia A. "Minimax regret strategies for greenhouse gas abatement: methodology and application", Operations Research Letters, 25: 219–230. They explain the motivation as follows:

The Minimax Regret criterion, also known as Savage criterion, is one of the more credible criteria for selecting decisions under uncertainty, i.e. when the likelihoods of the various possible outcomes are not known with sufficient precision to use the classical expected value or expected utility criteria.

The only risk-avert formulation fully supported in TIMES is the "Expected utility criterion with linearized risk aversion". The simplistic experimental support for using the Minimax criterion in TIMES was added only because it was easy to add, and some of the primary designers of TIMES, prof. Richard Loulou and Dr. Amit Kanudia, have advertised this criterion in some of their papers.
Reply
#7
[quote pid='3254' dateline='1525768211']
Thank you for the information Antti!  It seems like the minimax regret approach is inspired by robust optimisation - https://www.sciencedirect.com/science/ar...5417303459

I believe the method can give quite conservative solutions, and I hope to compare this method with a more standard stochastic formulation in the future.

Pernille


The minimax regret approach is to minimize the worst-case regret.

Published work, using this model formulation:

Loulou R. & Kanudia A. "Minimax regret strategies for greenhouse gas abatement: methodology and application", Operations Research Letters, 25: 219–230.  They explain the motivation as follows:

The Minimax Regret criterion, also known as Savage criterion, is one of the more credible criteria for selecting decisions under uncertainty, i.e. when the likelihoods of the various possible outcomes are not known with sufficient precision to use the classical expected value or expected utility criteria.

The only risk-avert formulation fully supported in TIMES is the "Expected utility criterion with linearized risk aversion". The simplistic experimental support for using the Minimax criterion in TIMES was added only because it was easy to add, and some of the primary designers of TIMES, prof. Richard Loulou and Dr. Amit Kanudia, have advertised this criterion in some of their papers.
[/quote]
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)