Page - 189 - in Short-Term Load Forecasting by Artificial Intelligent Technologies
Image of the Page - 189 -
Text of the Page - 189 -
Energies2018,11, 2008
(a) (b) DNN performs
betterANN
performs better
-5 -4 -3 -2 -1 0 1 2 3 4 5
ANN wmape - DNN wmape
0
5
10
15
20
25
30
Figure6. Thisfigure shows twohistograms: (a)Acomparisonof theperformanceofall 62models
between theDNNand the LR. Instances to the left of the center line are those forwhich the LR
performedbetter,while thoseontherightareareaswhere theDNNperformsbetter. Thedistance from
thecenter line is thedifference inWMAPE. (b)Thesameas (a)butcomparingtheANNtotheDNN.
Oneinstance (at10.1) in (b) is cutoff tomaintainconsistentaxes.
It isalso interestingtoconsider that insomeareas, theLRandANNforecastersperformbetter
than theDNN.This implies that in somecases, the simplermodel is thebetter forecaster. It is also
important topointout thatof the13areaswhere theLRoutperforms theDNN,only twohaveLR
WMAPEsgreater than5.5,whichmeans that thesimpleLRmodelsareperformingverywellwhen
comparedto industrystandards forshort-termloadforecastingofnaturalgasonthoseareas.
Figure7compares theperformanceof the twoDNNs.Aswith the twodistributions inFigure6,
a left-tailed t-testwasperformedon thehistograminFigure7 resulting inap-valueof 9.8×10−5.
Thismeans that theLargeDNNoffersastatisticallysignificantbetterperformanceover the62areas
than the smallDNN.However,much like in the comparisonbetween theDNNandothermodels,
thesmallDNNperformsbetter insomeareas,whichsupports theearlierclaimthatcomplexmodels
donotnecessarilyoutperformsimplerones.
Figure7.Acomparisonof theperformanceofall62modelsbetweentheDNNandtheLargeDNN.
Instances to the leftof thecenter lineare those forwhich theLargeDNNperformedbetter,while those
ontherightareareaswheretheDNNperformsbetter. Thedistancefromthecenter lineis thedifference
inWMAPE.
8.Conclusions
WeconcludethatDNNscanbebettershort-termloadforecasters thanLRandANNs.Onaverage,
over the62operatingareasexamined,aDNNoutperformedanotherwise identicalANNatshort-term
load forecasting of natural gas, and a larger DNNoffered even greater performance. However,
189
Short-Term Load Forecasting by Artificial Intelligent Technologies
- Title
- Short-Term Load Forecasting by Artificial Intelligent Technologies
- Authors
- Wei-Chiang Hong
- Ming-Wei Li
- Guo-Feng Fan
- Editor
- MDPI
- Location
- Basel
- Date
- 2019
- Language
- English
- License
- CC BY 4.0
- ISBN
- 978-3-03897-583-0
- Size
- 17.0 x 24.4 cm
- Pages
- 448
- Keywords
- Scheduling Problems in Logistics, Transport, Timetabling, Sports, Healthcare, Engineering, Energy Management
- Category
- Informatik