具有共同吸引力限制的多类目的地选择.pdf

1、“Destination”choiceismoregenerallylocationchoice.“Multiclass”isreferringto4step andotheraggregatemodelswithhouseholdsegmentations.Thisisnotreallyaboutaninnovation,butaboutapplicationofawellestablishedmodelconcept.Thepurposeofattractionconstraintsinlocationchoicemodels.Constrainedattractionshavelongb

2、eenjustifiedaxiomatically(employmentisgiven),byequilibriumtheories(thefreechoicesofshopperssustaintheobservedretailemployment),ordatafusion(trafficcounts,tripgenerationestimates).Locationsandamountsofemploymentbyindustryoroccupationareseldomderivedfromanyonescommutepreferences,butfromproximitytocust

3、omers,supplies,shipping,etc.Multinomial logit(mostcommonbutnottheonlykind).Ifsharesorvolumesofchoiceofthealternativesaregiven,ratherthanpredicted,thenmustsolveanendogenousattributeforeach,usuallyanalternativespecificconstant(ASC).Shadowpricesrepresentunobservedmarketclearingforces.Endogenous ASCshav

4、eauniquesolution.IPFandtheequivalentlogarithmicadjustmentarewellknown,simpletoapply,androbust.TherearedifferentmethodstosolvefortheASCsolution.TheseincludeNewtonRaphson(asinAlogit andsimilarestimation),butthisisinfeasiblewithhundredsorthousandsofalternatives.Thereareimprovedlargeproblemsolutions one

5、tobementionedlater.Shadowpriceupdateisexactlythesameasrecalibratingalternativespecificconstants.SomeIPFapplicationsadjustmatrix elementsdirectly,insteadoffactors(butthisshufflesmorestorageandaccumulatesroundofferror.)Veryfewhavereportedtoinvestigateconstraintasahypothesistoteststatistically.ThedePal

6、mastudyaccountedforprice,andfoundtheobservedpricesinsufficienttoexplainmarketclearing.Bernardin usedageneticalgorithm agoodchoiceforrobustnessforlackofyetknowngradientorcurvatureoflikelihood.Presentedatthe2014InnovationsinTravelModelingconference.MyrecentpaperinTRR2677(5)identifiedthegradientandthem