Page - 186 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Figure4. Single-PointCrossover 5. EvaluationExperiment 5.1. ExperimentalEnvironment Table1. InitialResources Generation Time(s) Money(yen) Stamina low 500 15000 7500 1000 medium 500 30000 15000 2000 high 500 60000 30000 4000 Wehaveimplementedouralgorithm inPython3andexecutedtheimplementedalgorithm on aPCwith Intel Core i7-8550U1.8GHz, 16GBRAMandWindows 10HomeOS. Further,weexecutedouralgorithm for threedifferent initial resourceassignmentsshown in Table 1, to investigate the correlations between the derived solutions and the initial resources. In this experiment,we derived the information on the resource consumption onmoneyand time for eachpath (solution) byusingGoogleMapAPI. Since it is diffi- cult toknowactual staminaconsumptionandsatisfactiongotonapath,weset the imag- inaryvalues empirically.Modelingof stamina consumption and satisfaction acquisition (proposed in [15]) isour futurework. In this experiment, we targeted 30 PoIs in Higashiyama-area of Kyoto, Japan as showninFig.11, andwedetermined thestart and thegoalpoints inadvance.Moreover, becauseof the shortdistancebetweenPoIs,weusedcar (taxi), busandwalkingas types of transportation. 5.2. Relationshipbetween InitialResourceandSolutions Fig. 5 shows the scattered plot of initial solutions generated at random (at 0-thGener- ation). These solutions are paths consisting of 10 randomly selectedPoIs including the start and goal points. Furthermore, Fig. 6, Fig. 7 andFig. 8 plot 100 solutions that our algorithm calculated at 500-th generationwith low,medium and high initial resources. Table2 showsapart of thederived solutions, andvalues represent consumed/remaining resources and satisfaction value. This result supports that our proposed algorithm can derive diverse solutions considering the trade-offs between resources and satisfaction values. MoreoverFigs. 6–8 indicate relationships betweenmoney, time, stamina and satis- faction.Specifically, inFig. 6, solutions aredense in the left-bottom area, because solu- tions are limitedby low initial resources in this case.On theother hand,medium initial resource case (Fig. 7), solutions aremore distributed inwider area than the low initial Y.Hiranoetal. /AMethod forGeneratingMultipleTourRoutes186