Page - 92 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2019,12, 57 systemreliabilitywasbetter thanthehouseandthelaboratoryloadsat thisstep,becausetheofficeload profilewasbasicallysynchronizedwith the irradiationandwindcurvesandthesolarenergycouldbe useddirectly tosupply the loads. Therefore,weomittedStep3 that representedtheoptimizationwith w=2andLPSP=0inTables5and6. ȱ (a)ȱwȱ=ȱ2ȱandȱ(SOClow,ȱSOChigh)ȱ=ȱ(40%,ȱ50%)ȱ (b)ȱwȱ=ȱ0ȱandȱ(SOClow,ȱSOChigh)ȱ=ȱ(30%,ȱ40%)ȱ Figure8.Thereferenceplots for theoffice load. Table7.Theoptimaldesignprocedures for theoffice load. (b,s,w) (SOClow,SOChigh) LPSP (%) J ($/kWh) Step1 (8,10,2) (40%,50%) 5.08% 1.107 Step2 (23,11,2) (40%,50%) 0% 1.106 Step3 - - - - Step4 (29,17,0) (40%,50%) 0% 0.818 Step5 (29,17,0) (30%,40%) 0% 0.817 Step6 (26,17,0) (30%,40%) 0% 0.791 Step7 (26,17,0) (30%,40%) 0% 0.791 Optimal (26,17,0) (30%,40%) 0% 0.791 Settingw=0gaveasignificantcost reductionto J=0.818$/kWhwithLPSP=0bysetting(b, s,w) = (29, 17, 0) (seeStep4ofTable7). The iterativeprocedures then further improvedthesystemcost to J=0.817$/kWhwithLPSP=0byadjustingthepowermanagementas (SOClow,SOChigh)= (30%, 40%),andthecostfinallyconvergedto J=0.791$/kWhwithLPSP=0bysetting(b, s,w)= (26,17,0) and(SOClow,SOChigh)= (30%,40%).Whencomparedwith theoriginalcost, thecostwasreducedby 28.6%whilemaintainingcompletesystemreliability. 3.4. Cost andEnergyDistributions Theoptimal systemdesigns for the three loads,basedonthereferenceplots, are illustrated in Tables5–7.Wefurtheranalyzedthecostandenergydistributionsof thesesystems,asshowninTable8. First, thelaboratoryachievedthelowestunitenergycostbecauseitsaveragedailyenergyconsumption was the largest; therefore, the initial costswereshared. Thehousehold loadshowedanopposite result. Second, thelaboratoryusedthemostsolarpanelsandbatteries,whilethehouseholdappliedthefewest solarpanels andbatteries, toproduceandstore sufficient energy for the loadrequirements. Third, theoptimalbatteryunits forall loadsdidnotdiffermuch(23–27units); thiswasnot intuitivebecause the laboratory loadwasmainlyatnight,while theoffice loadwasmainly indaytime. Thereasonfor thiswas that thebattery lifewas shortened if onlya small amountof thebatteryenergywasused. Therefore, using a large amount of the battery energy increased the initial cost but it also helped toextend thebattery life, thereby reducing thebattery costs. For instance, for the laboratory load, thebatterycostwas the lowesteven thoughthe laboratory loadusedthe largestamountofbattery energy. Because the initial batterySOCwasset as80%in the simulationmodel, anegativeenergy supplydistributionofbatterymeans thebatterySOCishigher than80%at theendof thesimulation, 92