Seite - 84 - in Algorithms for Scheduling Problems

Algorithms 2018,11, 68 Figure10.Displacementmutationoperator. Figure11.Exchangemutationoperator. Figure12. Insertionmutationoperator. 6.TuningUptheParameters 6.1. ParameterCalibration Calibration is aprocedure for choosing thealgorithmparameters thatprovide thebest result in theresponsevariables.Acomputationalexperiment forcalibration includes the followingsteps: (a)eachinstanceorworkloadisrunwithallpossiblecombinationsofparameters; (b)thebestsolutionis obtained;(c)therelativedifferenceofeachalgorithmonthebestsolutioniscalculated; (d)multifactorial AnalysisofVariance (ANOVA)witha95%confidence level isapplied tofindtheparameter thatmost influences thesolution; (e) thesetof thebestvalues is selectedforeachparameter. Table 9 shows the parameters used for calibration. A total of 3× 2× 3× 3× 3 = 162 differentalgorithmsorcombinationsareconsidered. Thirtyruns foreachcombinationareperformed, with162×30=4860experiments in total. Foreachof the30runs,adifferentworkloadwillbe taken. Foreachof thefive jobs,30batchesaregeneratedrandomlyfromauniformdistributionaccordingto their limits (Table4). Eachbatch isassignedtoaworkload,obtaining30 loads from5 jobs. Thevariationofprocessing capabilities (inminutes)of themachines inStages2and5 isgeneratedaccording toTables1and2. Stage2has6machineswithdifferentprocessingcapabilities inkilograms. Stage5has6machines. 84