简介:ThispaperdevelopsanewlowerboundmethodforPOMDPsthatapproximatestheupdateofabeliefbytheupdateofitsnon-zerostates.ItusestheunderlyingMDPtoexploretheoptimalreachablestatespacefrominitialbeliefandselectactionsduringvalueiterations,whichsignificantlyacceleratestheconvergencespeed.Also,analgorithmwhichcollectsandprunesbeliefpointsbasedontheupperandlowerboundsispresented,andexperimentalresultsshowthatitoutperformssomeofthestate-of-artpoint-basedalgorithms.
简介:Usingfinite-timecontrolapproach,thispaperproposesanewdesignmethodfornonlinearrobustexcitationcontrolofawidelyused5th-ordermodelofsynchronousgenerators.Thefinite-timeexcitationcontrollerachievedherecanimprovethesystem'sbehaviorsinsomeaspectssuchasquickconvergenceandrobustnessforuncertainties.Simulationsdemonstratethatthefinite-timeexcitationcontrollerismoreeffectivethansomeotherexcitationcontrollers.