資源描述：
《基于反向傳播神經(jīng)網(wǎng)絡的退化紅壤區(qū)杉木樹干液流模擬.pdf》由會員上傳分享，免費在線閱讀，更多相關內容在行業(yè)資料-天天文庫。
1、浙江大學學報(農業(yè)與生命科學版)41(2)：205~212，2015JournalofZhejiangUniversity(Agric．8LLifeSci．)http：／／www．journals．zju．edu．cn／agrE—mail：zdxbnsb@zju．edu．cnDOI：10．3785／j．issn．1008—9209．2014．05．191基于反向傳播神經(jīng)網(wǎng)絡的退化紅壤區(qū)杉木樹干液流模擬涂潔，劉琪壕，危駿，胡良(1．南昌工程學院生態(tài)與環(huán)境科學研究所，南昌330099；2．北京林業(yè)大學林學院，北京10
2、0083)摘要以江西退化紅壤區(qū)杉木人工林為研究對象，采用MATLAB工具箱中的log～sigmoid型函數(shù)(tansig)為神經(jīng)元作用函數(shù)，以空氣溫度、空氣相對濕度、平均凈輻射、水汽壓虧缺為輸入變量，液流速率為輸出變量，運用貝葉斯正則化算法和Levenberg-Marquardt算法對4000組氣象數(shù)據(jù)和液流數(shù)據(jù)進行網(wǎng)絡訓練和檢驗，構建拓撲結構為4—1O一1的杉木樹干液流反向傳播(backpropagation，BP)神經(jīng)網(wǎng)絡模型．結果表明：在2種算法下訓練樣本和檢驗樣本模型輸出值與實測值之間線性回歸的擬合程度均
3、較高，回歸方程的相關系數(shù)在0．93以上；訓練樣本的擬合精度分別為83．57和83．O6，檢驗樣本的仿真精度分別為82．87和82．15．說明該網(wǎng)絡模型能夠很好地反映液流速率與氣象因子之間的非線性函數(shù)關系，可為杉木人工林的可持續(xù)經(jīng)營和林地水資源的科學管理提供有效手段．關鍵詞杉木；樹干液流；貝葉斯正則化算法；Levenberg—Marquardt算法；反向傳播神經(jīng)網(wǎng)絡中圖分類號S718文獻標志碼ASapflowsimulationofCunninghamialanceolataindegradedredsoilreg
4、ionbasedonbackpropagationneuralnetwork．JournalofZh~iangUniversity(Agric．&LifeSei．)，2015，41(2)：205—212TuJie”，LiuQijing。，WeiJun，HuLiang(1．ResearchInstituteofEcology&EnvironmentalSciences。NanchangInstituteofTechnology，Nanchang330099，China2．DepartmentofForestScie
5、nces，BeijingForestryUniversity，Beijing100083，China)SummaryCunninghamialanceolataiscommonlyconsideredtobeoneofthemostimportanttreespeciesforforestrestorationandreconstructioninsubtropicalareaofChina，owingtoitsadvantagesofrapidgrowth，goodqualityandhighyieldperu
6、nitarea．However，theyalsoconsumecertainamountofwaterduringthecourseofgrowthandplayrolesofecologicalbenefits．Therefore，quantitativeresearchontreewaterconsumptioncharacteristicsbytranspirationhasbecomeahotissueinthefieldoftreephysiologicalecologyinrecentyears．Ta
7、kingtheClanceolataplantationindegradedredsoilofJiangxiProvinceastheresearchobject，thelog—sigmoidtypefunction(tansig)ofMATLABtoolboxwasselectedasthetransmissionfunctionfortheroleofneurons．Fourmainfactorsincludingairtemperature，relativeairhumidity，averagenetrad
8、iationandvaporpressuredeficitwerechosenastheinputvariables，andthesapflowvelocitywasselectedastheoutputvariable，totrainandexaminetheneuralnetworkmodelwithBayesianregularizationalgorithmand
