Asking MARKAL-STOCHASTIC problems

[Kelly Chou]

I am Kelly Chou from Industrial Technology Research Institute in Taiwan. I am studying MARKAL-STOCHASTIC and have some problems about the model results analysis.  I hope I can get some help by your modeling experiences. My problems of modeling are below:

(1)How to get the Expected Value of Perfect Information (EVPI) from model results? As I learn the formula for EVPI=COST_hedge-Σp_i l COST_PFi , I am not clear how to find the "COST_hedge" and "COST_PFi" in MARKAL result parameters. Do they mean the "D.TOTCOST"(Discounted Total System Cost, Net of Taxes & Subsidies) in MARKAL result parameters?

(2)The constraints of market share, setting in model, will not be effected while running MARKAL-Stochastic but they are effective in running Standard MARKAL. It results in incorrect technologies portfolio in running MARKAL-Stochastic. Have you ever run into this kind of problems in running MARKAL-Stochastic? If you did, could you tell me how to deal with the problems?

Thank you for your attention.

Best Regards.

Kelly Chou

[UCL_Will]

Dear Kelly,

Thank you for the question.

The first thing to work out is what results you should be looking at and how to compare runs.

Are you running Stochastic MARKAL with Elastic Demand or in Least Cost mode?

For example, if you have a simple stochastic model with the following upper bound on CO2 emissions:

1990      2000     2010
   10         10        20     SOW1
   15         20        30     SOW2

...and your resolution of uncertainty (START_STG2) parameter is set to the period 2010, your hedging strategy will be evident in 1990 to 2000 and recourse strategy in 2010.  However, note that the SOW1 constraint OVERRIDES all other values in other states of the world in the hedging strategy.  So, the values highlighted in red are overwritten.  This may be the problem you are seeing in (2) in your message.  An alternative issue could be that recently a bug was discovered that meant rule-based constraints were ignored in Stochastic MARKAL.  This has been corrected in a recent code update, so you may need to update your version of the MARKAL code.

So, to calculate EVPI, be very careful to ensure that your deterministic run is specified correctly.  Using our example above:

DET1 = 1990 10, 2000 10, 2010 20;
DET2 = 1990 10, 2000 10, 2010 30;

If you are running in MARKAL Elastic Demand, the expected cost of a stochastic run is as follows: Exp_Cost_[sow] = SUM((D.MED.TESCST + D.TOT.TAXSUB + D.MED-SURF.RED - D.MED-SURF.GRO)_[sow] X PROBABILITY_[sow]))

This might also be useful:

Answer Parameter
D.MED-TESCST	=	- D.MED.OBJ - D.MED-SURF.RED + D.MED-SURF.GRO	[1]
D.MED-TESCSTDIF	=	- D.MED-REF.OBJ - D.MED-SURF.RED + D.MED-SURF.GRO + D.MED-OBJ	[2]
D.MED-TESCSTDIF	=	- D.MED-REF.OBJ + D.MED.TESCST	[3]
D.MED.CPSURPLUS	=	- D.MED-OBJ + D.MED-REF.OBJ	[4]
D.MED.CPSURPLUS	=	- D.MED-SURF.RED + D.MED-SURF.GRO - D.MED.TESCSTDIF	[5]
D.MED.CPSURPLUS	=	- D.MED-SURF.RED + D.MED-SURF.GRO - D.MED.TESCST + D.MED-REF.OBJ	[6]
Objective Function for each SOW	=	D.MED.TESCST+ D.TOT.TAXSUB + D.MED-SURF.RED - D.MED-SURF.GRO	[7]
or by substituting [1] into [7]	=	- D.MED.OBJ + D.TOT.TAXSUB	[8]

I hope that helps.

Kind regards,

UCL_Will
Research Associate in Energy and Modelling
UCL Energy Institute

[Kelly Chou]

Dear Will,

Thank you for your response.

I make a test run for  the SOW1 constraint  NOT OVERRIDES all other values in other states of the world in the hedging strategy. But the constraints of market share setting in Stochastic Markal are still ignored .So I think this problem may be a bug on Markal code,as you said.

I am studing Stochastic MARKAL with Least Cost mode, and next study will extend with Elastic Demand. Thank you for your suggestion of Answer parameters.

Best regards,

Kelly

[UCL_Will]

Hi Kelly,

I realised that my posting wasn't very clear.  During the hedging strategy, your stochastic parameters for SOW 1 should override any parameter settings in other SOWs.  Are you seeing this?

In the recourse period, the parameter values in each state of the world should then take hold.

Will

[Kelly Chou]

Dear Will,

Thank your for your reminding. Yes, I make a mistake for the "OVERRIDE" of SOW1.

I am studding about the uncertainty of energy price, using MARKAL-STOCHASTIC with Least cost mode. I set the resolution of uncertainty on 2015 and the scenarios are p1=Low, p2=Low+50%, p3=Low+100%, for example:

Energy price    2005    2010   2015

p1                    10        12       14

p2                    15        18       21

p3                    20        24       28

So the correct setting on STOCHASTIC should be:

Energy price    2005    2010   2015

SOW1                20       24      28

SOW2                10       12      14

SOW3                15       18      21

And the deterministic run setting should be:

DET1 = 2005 20, 2010 24, 2015 28;
DET2 = 2005 20, 2010 24, 2015 14;

DET3 = 2005 20, 2010 24, 2015 21

Is it right?

[UCL_Will]

Hi Kelly,

Yes, that is correct.  However, if p1 = SOW 1, p2 = SOW 2 and p3 = SOW 3, then your deterministic runs would be:

DET1 = 2005 10, 2010 12, 2015 14;
DET2 = 2005 10, 2010 12, 2015 21;

DET3 = 2005 10, 2010 12, 2015 28;

So, you need to be careful when you specify the stochastic run and understand why you are specifying SOW1 as the LOW + 100% scenario.  Perhaps it would be better to use the LOW + 50% case as your SOW1 - this may be equivalent to a central case?

Will

[Kelly Chou]

Hi Will,

I learn correctly about the setting of SOW.

Thank you very much.