資源描述:
《基于遞歸神經(jīng)網(wǎng)絡的風暴潮增水預測》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在工程資料-天天文庫。
1、第12卷第5期智能系統(tǒng)學報Vol.12№.52017年10月CAAITransactionsonIntelligentSystemsOct.2017DOI:10.11992/tis.201706015網(wǎng)絡出版地址:http://kns.cnki.net/kcms/detail/23.1538.TP.20170831.1058.010.html基于遞歸神經(jīng)網(wǎng)絡的風暴潮增水預測111222雷森,史振威,石天陽,高松,李亞茹,鐘山(1.北京航空航天大學宇航學院圖像處理中心,北京100191;2.國家海洋局北海預
2、報中心,山東青島266000)摘要:風暴潮增水的準確預測能極大地減少人員傷害和經(jīng)濟損失,具有重要的實用價值。傳統(tǒng)的風暴潮預報方法主要包括經(jīng)驗和數(shù)值預報,很難建立起相對準確的模型?,F(xiàn)有的基于機器學習風暴潮預報方法大都只提取出靜態(tài)數(shù)據(jù)間的關(guān)系,并沒有充分挖掘出風暴潮數(shù)據(jù)背后的時序關(guān)聯(lián)特性。文中提出了一種基于遞歸神經(jīng)網(wǎng)絡的風暴潮增水預測方法。本文對風暴潮時序數(shù)據(jù)進行特定的處理,并設計合適結(jié)構(gòu)的遞歸神經(jīng)網(wǎng)絡,從而完成時序數(shù)據(jù)的預測。相較于傳統(tǒng)的BP神經(jīng)網(wǎng)絡,遞歸神經(jīng)網(wǎng)絡能更好地應對時序數(shù)據(jù)的預測問題。將該方法用于
3、濰坊水站的增水預測中,結(jié)果表明,相對于BP神經(jīng)網(wǎng)絡,遞歸神經(jīng)網(wǎng)絡能得到更好的預測結(jié)果,誤差更小。關(guān)鍵詞:風暴潮增水;預測;數(shù)值預報;機器學習;靜態(tài)數(shù)據(jù);時序特性;BP神經(jīng)網(wǎng)絡;遞歸神經(jīng)網(wǎng)絡中圖分類號:TP751文獻標志碼:A文章編號:1673-4785(2017)05-0640-05中文引用格式:雷森,史振威,石天陽,等.基于遞歸神經(jīng)網(wǎng)絡的風暴潮增水預測[J].智能系統(tǒng)學報,2017,12(5):640-644.英文引用格式:LEISen,SHIZhenwei,SHITianyang,etal.Predi
4、ctionofstormsurgebasedonrecurrentneuralnetwork[J].CAAItransactionsonintelligentsystems,2017,12(5):640-644.Predictionofstormsurgebasedonrecurrentneuralnetwork111222LEISen,SHIZhenwei,SHITianyang,GAOSong,LIYaru,ZHONGShan(1.ImageProcessingCenter,SchoolofAstro
5、nautics,BeihangUniversity,Beijing100191,China;2.BeihaiForecastCenterofStateOceanicAdministration,Qingdao266000,China)Abstract:Accuratelyforecastingstormsurgescangreatlyreducepersonnelinjuriesandeconomiclosses,andsohasgreatpracticalvalue.Traditionalmethods
6、forpredictingstormsurgemainlyinvolveexperienceandnumericalforecasting,whichmakesitveryhardtoestablishaccuratemodels.Mostoftoday’sstormsurgeforecastmethodsbasedonmachinelearningonlyextracttherelationshipsamongstaticdataandfailtoidentifytherelevanttimeserie
7、spropertiesofthesedata.Inthispaper,weproposeastormsurgeforecastmethodbasedontherecurrentneuralnetwork.Thestormsurgedataisrearrangedwithparticulartreatments,andanappropriaterecurrentneuralnetworkisdesignedtoperformthepredictionofthetimeseries.Comparedwitht
8、raditionalBPneuralnetworks,therecurrentneuralnetworkcanbetterforecasttimeseriesdata.Inthisstudy,weusedarecurrentneuralnetworktopredictsurgesattheWeifanggaugestation.Theresultsshowthattherecurrentneuralnetworkproduce