General Pattern and Neighborhood Search Algorithms for Non-Convex Unit Commitment

(整期优先)网络出版时间:2014-02-12
/ 1
Thispaperpresentsanewapproachtosolvemixed-variableunitcommitment(UC)problemswithnon-smoothcostfunctionsbasedonageneralizedpatternsearchfilter(GPS-filter)algorithm.AGPS-filteralgorithmdoesnotrequireanyinformationaboutthegradientoftheobjectivefunctionwhilesearchingforanoptimumsolution.Atthesametime,itisavailableforsolvingmixed-variableoptimizationproblems,whichisverysuitableforUC.AnewsuitablediscreteneighborhoodstructurewithUCcharacteristicsisproposedtoimproveGPS-filterefficiently.Alotofmultipleunits'statesarefixedbeforesearch;hence,thepollingsearchofdiscretevariableisefficientforafewuncertainunits.Numericalexperimentsareincludedtodemonstratetheproposedapproach'sabilitytohandlethehighlynonlinear,discontinuous,non-smoothcostfunctionsandmixedvariablesoftheUCproblem.