Page - 102 - in Algorithms for Scheduling Problems
Image of the Page - 102 -
Text of the Page - 102 -
Algorithms 2018,11, 55
agraphG, thealgorithmhas tovisit all conflictpairs, andby replacing theedgeswithdirectedarcs,
clarify thepriorities forusing the resources (i.e., the sections andassociated tracks). In this research,
the objective is tominimize the sumof the total final delay that each train experiences at its final
destination (i.e., TFDj = Max (
0,xendj,last−einitialj,last )
, j ∈ J ). We alsomeasure the correspondingwhen
the delay threshold is threeminutes, i.e., only delays larger than threeminutes are considered,
i.e.,(
∑TFD+3j ,if TFDj≥3min )
. The reasonwhy the thresholdof threeminutes is selected is that this
thresholdisusedtologtraindelaysinSweden.Delayssmallerorequaltothreeminutesarenotregistered
inthesystemoftheresponsibleauthority,Trafikverket (theSwedishNationalTransportAdministration).
Furthermore,delayslargerthanthreeminutesmaycausetrainconnectionstobemissed,whilebelowthree
minutes theconsequencesfor transfers inthestudiedregional-national trainsystemarenot that likely.
Tovisit theconflictingoperations,wefollowedthestrategypresentedbyGholamiandSotskovfor
thehybrid job-shopschedulingproblem[27]. In thisstrategy,a listofpotential candidateoperations
for conflict resolutionwill begenerated. This list is a collectionof ongoingoperations, or thefirst
operationof thenext trains.Afteranyconflict resolution, the listwillbeupdated, somecandidates
will beaddedand, if it isnecessary, somewill bedeleted. Anoperationwill beadded to the list if
the in-degreevalueof the relatedvertexbecomeszero (thenumberofarcends toavertex is called
the in-degreevalueof thevertex). The in-degreevaluedecreaseswhentheconflict resolution isdone
for aparentvertex. When the list is regenerated, the shortest release timeof all vertices in the list
will be calculatedorupdatedagain, and theminimumonewill be selected for conflict resolution.
Are-orderingmaycauseavertex tobedeletedfromthe list. Bythisapproach,all theverticeswillbe
visitedonlyonce,unlessare-orderinghappens. Thisstrategydecreases thecomputational timeof the
algorithm.Thesecondbenefitof thisapproach is thataddinganewarc (oj,i, oj′,i′ )doesnotaffectany
previouslyvisitedverticesunlessare-orderingoccurs.Dynamicupdateofdata ispossibledueto this
feature. InAlgorithm1, thepseudocodeforconflict resolution ispresented.
Thealgorithmstarts fromavertexos andfindsall theneighbourhoodverticesandaddsthemto
a list of candidatevertices. At each iteration, avertexwithaminimumrelease timewill be selected
fromthecandidate list forconflict resolution. If there isaconflictbetweenthecandidateeventandthose
events thatpreviouslyusedthat trackorsection(checkList),a localre-routingwillbeapplied. For local
re-routing,aplatformortrackwithminimumavailability timewillbeselected.Meanwhile, if there is
still a conflict, thealgorithmtries touse re-timingor re-ordering. For conflict resolutionbetween the
candidateoperationoj′,i′andanoperationoj,i fromthechecklist,anewarcwillbeaddedtothegraphG.
Afteraddingthearc(oj,i,oj′,i′ ), thestart time(x begin
j′,i′ )ofoperationoj′,i′willbepostponedtothefinishing
time(xendj,i )ofoperationoj,iontheconflictingtrackorplatform. If thealgorithmaddstheoppositearc
(oj′,i′,oj,i ), thenoperationoj′,i′havetobeservedbeforetheoperationoj,i,whichmeansthat thepredefined
order of the trains forusing this section is changed. This conditionoccurswhen the algorithm tries
topreventdeadlocks (unfeasiblesolutions),or theoperationoj,ihasahightardiness (adelayedtrain).
Sixdispatchingrulesareprovidedfor theconflict resolutionprocess. Table2 isadescriptionof those
dispatchingrules.
In the trainschedulingproblem,a trainblocks itscurrentsectionuntilobtainingthenextsection
orblock. Therefore, swappingisnotpossible in the trainschedulingproblem.Accordingly, theaddArc
functionhas tosatisfy theblockingcondition in thegraphG (thealternativegraphapproach). Tofulfil
theblockingcondition, insteadofaddinganarc fromtheoperation oj,i to theoperation oj′,i′, anarc
fromthenextoperationoj,i+1withazeroweightwillbeaddedtotheoperationoj′,i′. Thisdirectedarc
meansthat theoperationoj′,i′ cannotproceedwith itsexecutiononthemachinemuntil theoperation
oj,i+1 starts (x begin
j′,i′ ≥ x begin
j,i+1,meanwhile, theoperationoj,i is finished). Thisconditionblocks job j′onthe
previousmachine,evenif it iscompleted. If theoperationoj,i is the lastoperationof job j, thenanarc
withaweightdj,ihas tobeadded.
102
back to the
book Algorithms for Scheduling Problems"
Algorithms for Scheduling Problems
- Title
- Algorithms for Scheduling Problems
- Authors
- Frank Werner
- Larysa Burtseva
- Yuri Sotskov
- Editor
- MDPI
- Location
- Basel
- Date
- 2018
- Language
- English
- License
- CC BY 4.0
- ISBN
- 978-3-03897-120-7
- Size
- 17.0 x 24.4 cm
- Pages
- 212
- Keywords
- Scheduling Problems in Logistics, Transport, Timetabling, Sports, Healthcare, Engineering, Energy Management
- Categories
- Informatik
- Technik