本文選題：徑流降尺度　切入點：貝葉斯神經(jīng)網(wǎng)絡(luò)　出處：《中國農(nóng)村水利水電》2016年01期 　論文類型：期刊論文

【摘要】：基于貝葉斯神經(jīng)網(wǎng)絡(luò),構(gòu)建了資料匱乏地區(qū)的徑流降尺度模型,模擬了葉爾羌河卡群站月平均徑流,與BP神經(jīng)網(wǎng)絡(luò)的結(jié)果進行了對比,驗證了BNN的優(yōu)越性,并結(jié)合CMIP5三種氣候模式GFDL_ESM2G,GFDL_ESM2M及MIROC5的RCP 4.5,RCP 6.0,RCP 8.5三種情景,對未來3個時段(2020年代,2050年代,2080年代)卡群站月平均徑流進行了預(yù)測,并定量計算了預(yù)測的不確定性區(qū)間,研究表明:貝葉斯神經(jīng)網(wǎng)絡(luò)降尺度模型可以較好地捕捉葉爾羌河的徑流特征,即相關(guān)系數(shù)達到0.9以上,效率系數(shù)達到0.8,且模擬效果比ANN較優(yōu);未來情景下,葉爾羌河流域受氣溫升高影響,3個時段年徑流均呈現(xiàn)增加的趨勢,增加幅度分別為75%~92%,83%~110%,88%~127%,其中RCP8.5情景下的徑流增加幅度比其他情景較明顯;不同月份徑流存在不同程度的增加趨勢,其中5-8月份變化趨勢相對較明顯。
[Abstract]:Based on Bayesian neural network, the downscaling model of runoff in the area of lack of data is constructed, and the average monthly runoff of Karn station in Yerqiang River is simulated. The results are compared with the results of BP neural network, and the superiority of BNN is verified. Combined with the three climate models of CMIP5, GFDL, ESM2G / GFDL2M and MIROC5's RCP 4.5RCP6.0 / RCP8.5.The monthly mean runoff of Cage stations in the next three periods is forecasted from 2020s to 2080's, and the uncertainty interval of prediction is calculated quantitatively. The results show that the downscaling model of Bayesian neural network can better capture the runoff characteristics of the Yerqiang River, that is, the correlation coefficient is more than 0.9, the efficiency coefficient is 0.8, and the simulation effect is better than that of ANN. The annual runoff of the Yerqiang River Basin was affected by the increase of temperature, and the annual runoff showed an increasing trend in the three periods. The increase range was 75 / 92and 110810810810108127respectively, in which the increase of runoff in RCP8.5 scenario was more obvious than that in other scenarios, and the increase trend of runoff in different months was different. Among them, the trend of change in May and August is relatively obvious.
【作者單位】： 河海大學(xué)水文水資源與水利工程科學(xué)國家重點實驗室;
【基金】：國家自然科學(xué)基金面上項目(41371051)
【分類號】：TV121
，

本文編號：1561807