Parallel Proximal Bundle Methods for Stochastic Electricity Market Problems

Publication TypeConference Paper
Year of Publication2015
AuthorsF.-Javier Heredia; Antonio Rengifo
Conference Name27th European Conference on Operational Research
Conference Date12-15/07/2015
Conference LocationGlasgow, UK.
Type of Workinvited
Key Wordsresearch; MTM2013-48462-C2-1; mixed-integer nonlinear programming; proximal bundle methods; multimarket electricity problems; parallelism
AbstractThe use of stochastic programming to solve real instances of optimal bid problems in electricity market usually implies the solution of large scale mixed integer nonlinear optimization problems that can't be tackled with the available general purpose commercial optimisation software. In this work we show the potential of proximal bundle methods to solve large scale stochastic programming problems arising in electricity markets. Proximal bundle methods was used in the past to solve deterministic unit commitment problems and are extended in this work to solve real instances of stochastic optimal bid problems to the day-ahead market (with embedded unit commitment) with thousands of scenarios. A parallel implementation of the proximal bundle method has been developed to take profit of the separability of the lagrangean problem in as many subproblems as generation bid units. The parallel proximal bundle method (PPBM) is compared against general purpose commercial optimization software as well as against the perspective cuts algorithm, a method specially conceived to deal with quadratic objective function over semi-continuous domains. The reported numerical results obtained with a workstation with 32 threads show that the commercial software can’t find a solution beyond 50 scenarios and that the execution times of the proposed PPBM are as low as a 15% of the execution time of the perspective cut approach for problems beyond 800 scenarios.
URLClick Here
ExportTagged XML BibTex
Heredia_EURO15.pdf978.91 KB