Page 12 - 《水资源与水工程学报》2022年第4期
P. 12
!33 " ! 4 # & ' ( ) & * + , - Vol.33No.4
2022 $ 8 % JournalofWaterResources&WaterEngineering Aug.,2022
DOI:10.11705/j.issn.1672-643X.2022.04.02
' WD-COA-LSTMY7
()*, +,-, ./0
( E9|9èHI 9KLIX, ], /0 450046)
A B: 8oÝA¥dá9b(r1v, "Yï¯2¨Z(WD) ?34§«(COA) ßr5Ôã6ñ{
78(LSTM) á9b(rB0(WD-COA-LSTM)。9¨2¨Ze=:op(µF, ò=:°»;
ç, þÅ 1 l¢<=:? 3 l¥<=:; Ù×K34§«ßeñ{78(LSTM) B0op!§«; <×ÑØ
ì=:(r=4þÅdá9b(r。Ñ"B0'¯>?¶@jABCt?>ÄAýAt lÐbD
dá9b(r^, âM LSTM、COA-LSTM、WD-LSTMB0(ropeÁ。: " WD-COA-
LSTMB0(r1v<¥, M2¨Z?34§«ßDã34ÿ LSTMB0(r1v?Û«DE, 8dá
9b(rYÝt)}~。
CDE: dá9b(r; 2¨Z; 34§«ß; r5Ôã6ñ{78
FGHIJ:TV125 KLMNO:A KPQJ:1672643X(2022)04000806
MonthlyprecipitationpredictionbasedonWD-COA-LSTM model
WANGWenchuan,YANGJingxin,ZANGHongfei
(CollegeofWaterResources,NorthChinaUniversityofWaterResourcesandElectricPower,Zhengzhou450046,China)
Abstract:Inordertoimprovethepredictionprecisionofmonthlyprecipitation,theprecipitationpredic
tionmodelofWD-COA-LSTMisproposedbasedonwaveletdecomposition (WD),coyoteoptimization
algorithm(COA)andlongshorttermmemory(LSTM)neuralnetwork.Firstly,thetimeseriesispre
processedbyWDtoeliminateitsnonstationarity ,andalowfrequencysequenceandthreehighfrequen
cysequencesareobtainedastheresult.ThentheparametersoftheLSTMmodelareoptimizedbyCOA.
Finally ,thepredictedmonthlyprecipitationisobtainedbysuperimposingthepredictedvaluesofeach
subsequence.TheproposedmodelwasappliedtothemonthlyprecipitationpredictionofBaituTownin
LuanchuanCountyandGuxianTowninLuoningCounty ,LuoyangCity,andtheresultswerethencom
paredwiththoseoftheLSTM,COA-LSTM andWD-LSTM models.Itisfoundthattheproposed
WD-COA-LSTM modelproducedthehighestpredictionaccuracy ,indicatingthatWDandCOAcan
improvetheprecisionandgeneralizationabilityofLSTMmodel.Thismodelprovidesanewapproachfor
thepredictionofmonthlyprecipitation.
Keywords:monthlyprecipitationprediction;waveletdecomposition(WD);coyoteoptimizationalgo
rithm(COA);longshorttermmemory(LSTM)neuralnetwork
8 [6-8] s。È^r5Ôã6(longshortterm memo
1 ./01
ry,LSTM) ñ{78B0D1@(r=:, ã
á9(r±9Ä(·F<Ý'R§45$ 3ZQYÝzñ{78^"5RvSTÖ×, Â
ð, á9b?L±GfHI、 JKIËÀ·Íy 8mc£45vS。ô Kumar s [9] mUOñ
Ý, Í[, eÈ/(ræãR§bc。mc£&' {78(recurrentneuralnetworks,RNN) B0?r5
&QIveá9b(ropYHbLM, ¦§4 Ôã6(LSTM) ñ{78B0eVvR½<á9
56ßã¤NOB0 [1-2] 、 jë7bP [3-5] 、 ñ{7 bopY(r, LSTMB0Á RNNB0(
!"#$:20211109; %$:20220429
'()*: ],#RS4fMÚ&'(202102310259、202102310588); ],#¥'$%()*+(18IRTSTHN009)
+,-.: ®Äj(1976-), +, ],,-., /0, 23, /0., 456789ÄI*9KL。