Page - 346 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1282 Table6.Cont. Parameter Value Thebiasmatrix βik= ⎛⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎝ −5.12 −5.12 −5.12 5.12 3.19 5.12 −1.84 −1.37 2.81 −2.42 5.12 3.19 5.12 −1.84 −1.37 2.81 −2.42 5.12 3.19 5.12 −1.84 −1.37 2.81 −2.42 −5.12 3.61 −5.12 3.61 −5.12 3.61 ⎞⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎠ Theoutputweightmatrix ρ= ( 0.34 −0.45 −0.48 0.38 0.41 −0.28 0.40 −0.23 −0.21 0.24 ) Figure6.The iterationsprocessof thebatalgorithm(BA). 4.CaseStudy Inorder to testify the feasibilityof theproposedmodel, the24-hpower loaddataofYangquan City are selected for twomonths. It can be seen that there is nearly no apparent regularity to be obtained fromtheactual loadcurves showed inFigure7which represents the four classesof load curve. Asmentioned above, the three testing days belong to classes 4, 1, 3 respectively and the predictionmodel isbuilt for thepower loadforecastingat thesametime. Class 1 Class 2 Class 3 Class 4 Figure7.Thefour typesofpower loadcurve. Theprogramruns inMATLABR2015bunder theWIN7system. The short-termelectric load forecastingresultsof threedaysof theBA-ELM,ELM,BPandLSSVMmodelsareshowninTables7–9, respectively. For thepurposeofexplaining theresultsmoreclearly, the forecastingvaluescurveof the proposedmodelandcomparisonsareshowninFigures8–10. Inaddition,Figures11–13reflect the comparisonsof relativeerrorsbetweentheproposedmodelandtheothers.AccordingtoFigures8–10, thedeviationcanbecapturedbetweentheactualvalueandthe forecastingresults. It canbeseenthat 346