I’m working with large scale unit commitment models of the European electricity market, formulated as mixed-integer linear programs (MILP). The model is written and generated in GAMS 24.7 and solved by the ILOG CPLEX 12.5 solver. Preprocessing of input data and output data is executed in Matlab.
An issue concerning large scale MILP unit commitment problems is the long run time of these models. With respect to this issue, I wonder what the different solver options can contribute in reducing the run time. Currently, all CPLEX solver options are default, except the number of threads which is set to 4. The CPLEX solver manual shows different options with regard to MIP (node selection, branching, cutting planes, preprocessing, etc.).
Has anyone experience with this? There are too many solver option to test them one by one.
In attachment an example of a MILP unit commitment problem I've developed.
Thank you in advance.
NOTICE: developerWorks Community will be offline May 29-30, 2015 while we upgrade to the latest version of IBM Connections. For more information, read our upgrade FAQ.
This topic has been locked.
3 replies Latest Post - 2013-03-07T22:44:23Z by SystemAdmin
Pinned topic CPLEX MIP solver options for unit commitment model
Answered question This question has been answered.
Unanswered question This question has not been answered yet.
Updated on 2013-03-07T22:44:23Z at 2013-03-07T22:44:23Z by SystemAdmin
Re: CPLEX MIP solver options for unit commitment model2013-02-26T22:15:35Z in response to SystemAdminYou might try the CPLEX tuning tool. It evaluates predetermined sets of parameter settings that are known to work well for some models. Information on the tool can be found in the documentation at http://pic.dhe.ibm.com/infocenter/cosinfoc/v12r5/topic/ilog.odms.cplex.help/CPLEX/UsrMan/topics/progr_consid/tuning/01_tune_title_synopsis.html
Re: CPLEX MIP solver options for unit commitment model2013-03-07T22:44:23Z in response to SystemAdminMy experience with this kind of model was, that it takes quite long so find feasible solutions.
It helps to first solve a reduced model, fixing a subset of the cheapest in cost per MWh units to on,
changing the model by relaxing the fixed bounds and continuing with the full model.
Another measure helping is: up to some limit (nodes or time) set emphasis to 3, use probing 3 and
set cuts to aggressive, then continue (retaining the tree) with emphasis 4 (hidden feasible solutions).
Using only default settings normally gives computing times of weeks or memory exhaustion.