不等式约束优化新型sqcqp算法分析-analysis of new sqc qp algorithm for inequality constrained optimization.docxVIP

不等式约束优化新型SQCQP算法研究摘要不等式约束优化是数学规划领域的一个重要分支,在经济计划、电网规划、实时控制、工程设计和交通运输等实际问题中有着相当广泛的应用.不等式约束优化研究的焦点在于探索设计各类快速有效、实用的数值优化方法.本学位论文针对不等式约束优化问题,借助序列二次约束二次规划(SQCQP)方法、罚函数法和积极识别集精确识别技术等思想对此类问题的新型求解算法展开研究.首先,提出了一个新的求解不等式约束优化问题的罚函数型全局收敛SQCQP算法.该算法初始点任意选取,每次迭代只需求解一个恒有最优解的凸二次约束二次规划(QCQP)子问题,并且采用积极识别集精确识别技术降低子问题的规模和计算成本.算法在没有函数凸性及SlaterCQ等条件下实现了全局收敛性.此外,算法在数值测试过程中具有稳定性和有效性.其次,建立了一个新的快速收敛SQCQP算法.通过构造新型QCQP子问题,并设计适当的罚参数修正策略和新型线搜索,算法不仅能从任意初始点开始,而且能保证迭代点在有限步迭代后落入可行域.并在不需要线性无关约束规格(LICQ)、严格互补等较温和的条件下,证明了算法的弱全局收敛性、全局收敛性以及超线性收敛性.最后,通过一定规模的数值试验测试了算法的有效性.关键词：不等式约束优化序列二次约束二次规划新型线搜索算法积极识别集全局收敛性超线性收敛性RESEARCHONNEWSQCQPALGORITHMSFORINEQUALITYCONSTRAINEDOPTIMIZATIONABSTRACTInequalityconstrainedoptimizationisoneofthemostimportantbranch-esinthefieldofmathematicalprogramming,whichiswidelyappliedinprac-ticalproblems,suchaseconomicplan,powernetworkplanning,real-timecontrol,engineeringdesign,transportationandsoon.Thefocusofthein-equalityconstrainedoptimizationresearchistoexploreanddesignallkindsofquick,effectiveandpracticalnumericaloptimizationmethods.Thethe-sisstudiesnewnumericalalgorithmsforsolvinginequalityconstrainedop-timizationproblemsbymeansoftheideasofsequentialquadraticallycon-strainedquadraticprogramming(SQCQP)method,penaltyfunctionmethodandanaccurateidentificationtechnologyofactiveset.Firstly,anewpenalty-functiontypegloballyconvergentSQCQPalgo-rithmforinequalityconstrainedoptimizationproblemsisproposed.Theal-gorithmcanchooseaninitialiterationpointarbitrarily,ateachiterationonlyaconvexquadraticallyconstrainedquadraticprogramming(QCQP)subprob-lemwhichalwayshasanoptimalsolutionneedstobesolvedandthescaleofthesubproblemsandcomputationalcostarereducedbyemployingtheactivesetaccuratelyidentificationtechnology.Withoutconditionssuchasthecon-vexityoffunctionsandSlaterconstraintqualification,globalconvergenceofthealgorithmisproved.Inaddition,thealgorithmisstableandeffectiveintheprocessofnumericaltests.Secondly,anewfastconvergentSQCQPalgorithmisestablished.ByconstructinganewQCQPsubproblemanddesigningaapp