Page - 154 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2080 4.7.NumberofLags Thenumberof lags ineachmodel is changedfrom0to25 inorder toexposehowthisparameter affects theaccuracyofeachmodel. TheresultsarecategorizedbytypeofdayonTable8. TheARmodel obtainsa less accurate forecast than theNNwhen the lagsarebelow7days. However, the results beyondthis thresholdbenefit theARmodelclearly. TheARmodelseemstocontinue its improvement upto lagnumber21(threeweeks)but theNNreachesaplateauat lag7.Onceagain, theNNmodel performsmoreaccuratelywhenlittle information(in thiscase lags) isavailablebut it isoutperformed bytheARmodelwhenthe limitation is lifted. Figure14represents theoverallaccuracyofbothmodels as thenumberof lags is increased. It isworthnoticinghowtheARmodel improvesspeciallyat lags7, 14and21. Table8.Forecastingerror (RMSE)withdifferent laggedfeedback. LAG Overall Regular Special Hot Cold AR NN AR NN AR NN AR NN AR NN 0 1.80% 1.73% 1.76% 1.64% 2.08% 2.31% 1.94% 1.89% 2.00% 1.57% 1 1.73% 1.65% 1.68% 1.55% 2.04% 2.34% 1.86% 1.81% 1.85% 1.47% 3 1.64% 1.61% 1.59% 1.49% 1.96% 2.36% 1.81% 1.82% 1.87% 1.53% 5 1.62% 1.57% 1.57% 1.48% 1.95% 2.16% 1.81% 1.91% 1.86% 1.44% 6 1.56% 1.56% 1.51% 1.47% 1.89% 2.11% 1.75% 1.91% 1.81% 1.41% 7 1.45% 1.56% 1.40% 1.47% 1.81% 2.14% 1.55% 1.89% 1.72% 1.47% 8 1.45% 1.55% 1.39% 1.46% 1.81% 2.15% 1.55% 1.90% 1.72% 1.45% 13 1.45% 1.57% 1.39% 1.47% 1.84% 2.16% 1.52% 1.97% 1.72% 1.45% 14 1.43% 1.55% 1.37% 1.46% 1.83% 2.10% 1.51% 1.93% 1.71% 1.45% 15 1.43% 1.57% 1.37% 1.48% 1.83% 2.15% 1.51% 1.92% 1.71% 1.50% 20 1.43% 1.58% 1.37% 1.48% 1.83% 2.21% 1.52% 1.95% 1.70% 1.48% 21 1.42% 1.56% 1.35% 1.48% 1.83% 2.08% 1.49% 1.91% 1.68% 1.48% 22 1.42% 1.54% 1.35% 1.46% 1.83% 2.03% 1.50% 1.92% 1.68% 1.45% 24 1.42% 1.58% 1.36% 1.48% 1.84% 2.17% 1.50% 1.93% 1.68% 1.47% 25 1.42% 1.59% 1.35% 1.49% 1.84% 2.25% 1.50% 1.92% 1.68% 1.49% Testconditions: 7YT,4N,10RN,5TL,12MF. 1.40% 1.45% 1.50% 1.55% 1.60% 1.65% 1.70% 1.75% 1.80% 1.85% 0 1 3 5 6 7 8 13 14 15 20 21 22 24 25 LAGS Overall accuracy vs lags in feedback AR NN Figure14.Overall forecastingerror (RMSE)withdifferent laggedfeedback. 4.8.OverallResults Theprevioussubsectionsshowhowthere isnotasinglesolutionfor the load-forecastingproblem. Theconditionsunderwhichthe forecast isdoneduetoavailabilityordataor timeconstraintsaffect 154