Page 74 - 《水资源与水工程学报》2022年第6期
P. 74
7 0 & ' ( ) & * + , - 2022 $
ageerror ≤15.0%,thepassrate≥73.3%,andthecoefficientofcertainty ≥0.99;however,itper
formedpoorlywhentheforecastperiodis ≥7months.Ithasanidealforecastingeffectforthedailyrunoff
withaforecastperiodof1-3d,withthemeanabsolutepercentageerror ≤1.23%,thepassrate=
100%,andthecoefficientofcertainty ≥0.999;italsoperformssatisfactorilywhenforecastperiodis4-
7d,withthemeanabsolutepercentageerror ≤ 15.3%,thepassrate ≥ 73.0%,andthecoefficientof
certainty ≥ 0.94;whereaswhentheforecastperiodis ≥ 8d,theforecasteffectispoor.TheWPT-
HPO-ELMmodelcangivefullplaytotheadvantagesofWPT ,HPOandELM,showinghighforecast
accuracyandstability ;however,theforecasterrorincreaseswiththeincreaseoftheforecastperiod.The
proposedmodelandmethodcanprovideanewapproachforthemultistepforecastingofrunofftimese
ries.
Keywords:runoffforecast;waveletpackettransform(WPT);hunter-preyoptimization(HPO);ex
tremelearningmachine(ELM);multistepforecast;simulationtest
WD-PSO-LSSVM`ç¥!"o}NNO
1 ./01
xó£"©ª«Äv, ¬®Ã`çâľ
£"MörÿÄÚ, (Ü£"Mörÿ°4 .ÄÚPO; =ÊÊf [9] }~«ò£Ð(sin
ÄÚKðÜNOÄÚ&Jº%3%。¸.£" gularspectrumanalysis ,SSA)- 8óoéG - Øé
Mörÿ°4ÄÚPO¥àNOÄÚ、 NQRa' LÅ2(supportvectorregression,SVR) ç, {
dufâÄ$$Z[。^à!ò£"í|hÅ] |õ!m[:ó£"Äv5l¥ç©ª«;,
íÁ,BP(backpropagation) ËÖÖh [1] 、 ØéL ¬®Ãç+>YôÇNO£"¯Ï
[2]
Å(supportvectormachines,SVM) 、 ,Å^8(ran ¼X, âÄ·ÄvY¬; ûH¶f [10] i>¡
[3]
domforest ,RF) 、 }ÏM`ËÖÖh(longshort k£(VMD)、 ¢£ÉÊ(detrendedfluc
[4]
termmemorynetworks,LSTM) fÑ\ÄvK3 tuationanalysis ,DFA) } Ï M ` Ë Ö Ö h
ÀÇCcZÄÚY¬。1×, 7à“ £ - Äv (LSTM) K, ¸¹ DFA-VMD-LSTMö£"ÄÚ
- û” bÄvÔK·*üuà£"M ç¥9:Njö£"©ª«ÄÚ, ÞC¾·ÄÚ
örÿÄv&J, ñìQf [5] 7à¡k£ Y¬; j#f [11] 7àMÉ£KØéLÅ
(variationalmodedecomposition,VMD) Kí* (SVM) }~«`ö£"ÄÚç¥ÕÆxö£
- }ÏÝ`ËÖÖh(convolutionalneuralnetwork "©ªÄÚ, ;«ÃçâÄ·ÄÚÎX;
andlongshortterm memorynetworks ,CNN-LSTM) j'*f [12] «MÉ¡.(ØéLÅR, ¸¹
}~ VMD-CNN-LSTMç, {|!"oÐ3 duMÉØéLÅ2祫!"oÐ3hN
hxÓ3îx 1960-2005 5vó£"±í¥ Oxö£"©ªÄÚ, ¬QÃç;ÄY¹
[13]
穪;, ¬® VMD-CNN-LSTMç ÄÚö£"; {éVf MÉK、 J
âÄoÄvPOÀPMÇ¿; â;ïf [6] èoéGØéLÅ, h}«MÉ - Jè
7àÖk£(empiricalmodedecomposition, - ØéL2¨ö£"ÄÚ礥8}@r
EMD) ( Elman ËÖÖh}~« EMD-Elman ó£ gùõNOxö£"©ª«ÄÚ, ¬®Ãç
"`Ävç, ¤«(uàj!ËgKùðN Ïö£"ÄÚrâľõX; ÐÏWf [14] }
Ox 1979-2009 ó£"MörÿÄvr; ñG ~«Ök£ (EMD)-/öe (forensic
tf [7] }~«SûÖk£(complemen basedinvestigation ,FBI) G - W6eÅ(extreme
taryensembleempiricalmodedecomposition,CEEMD) learningmachine ,ELM) `£"Ävç, ¤«Ã
-BP ËÖÖhç, ¤{|8!ó£"Äv5l¥ çuàQ"#÷Î!NOx£"Äv, Ç
穪«;, ;®«Ã`çÄvY¬¾·; C«¾·ÄvY¬。
sÝ>f [8] 7à/ÞûÖk£(complete 5È, ËçÃK÷ÏÀZE<Ã:j:
ensembleempiricalmodedecompositionwithadaptive (1) c°±£"ÄÚ¸`¥Ä¯ÝM 1óà 1d
noise ,CEEMDAN) MÉ£ (waveletdecomposi ø4ÄÚ, ÈÏ5Tur, ø4ÄÚÀc
tion ,WD) `¸Ô£K, ¸¹ CEEMDAN- jó£"Ãö£"ÄÚ5T÷K, {[í÷$ì