Seite - 344 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1282 Figure4.Thepartialautocorrelationresultof theoverallpower load. 01˖0000˖00 02˖00 03˖00 04˖00 05˖00 06˖00 07˖00 09˖0008˖00 10˖00 11˖00 12˖00 13˖00 14˖00 15˖00 16˖00 17˖00 18˖00 19˖00 21˖0020˖00 22˖00 23˖00 Figure5.Thepartialautocorrelationresultof the loadwith thesameinterval. 3.4. ClusteringwithAntColonyAlgorithm Selecting the exogenous features as input directlymay lead the predictionmodel to a slow convergenceandtopoorpredictionaccuracy. Thus, thepaperemploys thesimilarday loadwhich is clusteredbytheantcolonyclusteringalgorithmfor thepredictionsoas to improve the forecasting accuracy.Accordingto the loadeverydayandthesix factorsextractedfrom22variables, the60days from1May2013 to 30 June2013 arenamedwithnumbers from1 to 60 andaredivided into four clustersby theant colonyalgorithm. Theparametersof theACCalgorithmcanbeseen inTable3, andtheclusteringresult isexpressed inTable4.Asaconsequence, it canbeknownthat the three test dayswhosenumbersare58,59,and60belongtoclass4, class1,andclass3, respectively. Table3.Parametersof theantcolonyclusteringalgorithm. Parameter m Alpha Beta Rho N NC_max Value 30 0.5 0.5 0.1 4 100 344