摘要
精确的径流模拟在水资源的规划和管理中起着非常重要的作用。但是,传统方法在径流峰值及突变点附近的模拟存在一定的局限性。首先利用Mann-Kendall突变检验法识别汉江流域安康水文站控制断面以上流域降雨径流突变成分,并采用R/S分析法和小波分析对其进行趋势和周期分析。然后利用偏最小二乘回归(PLSR)和BP神经网络-偏最小二乘回归(BP-PLSR)对径流序列进行模拟,分析峰值处和突变点附近的模拟效果。结果表明:降雨突变点出现在1973年、1984年、2002年,径流突变点出现在1977年、1985年;降雨径流Hurst指数均接近0,未来均存在反持续趋势;BP-PLSR对径流的模拟效果(RMSE=92.863,NSE=0.797)优于传统PLSR(RMSE=152.182,NSE=0.456),对原始数据进行BP预处理能更好地避免径流峰值处过拟合现象及突变点附近的局部最优现象。
Accurate runoff simulation plays a very important role in the planning and management of water resources.However,traditional methods have some limitations in the simulation of runoff near peaks and abrupt points.This paper identifies the abrupt change components of rainfall and runoff in the basins above the controlled section of Ankang Hydrological Station in the Hanjiang River Basin through the Mann-Kendall test.analyzes the trend and cycle of the rainfall and runoff by R/S analysis and wavelet analysis,simulates the runoff series with the partial least squares regression(PLSR)and BP neural network-partial least squares regression(BP-PLSR),and analyzes the simulation effect of runoff near peaks and abrupt points.The results show that:The abrupt points of rainfall appear in 1973,1984 and 2002;and those of runoff appear in 1977 and 1985.The Hurst index of rainfall and runoff is close to 0,indicating that there will be an anti-continuous trend in the future.The simulation effect of BP-PLSR on runoff(RMSE=92.863,NSE=0.797)is better than PLSR(RMSE=152.182,NSE=0.456),and preprocessing the original data by BP can better avoid the over-fitting and local optimization near the abrupt points.
作者
刘易文
李家科
丁强
郝改瑞
LIU Yiwen;LI Jiake;DING Qiang;HAO Gairui(State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China,Xi’an University of Technology,Xi’an 710048,China;Shaanxi Environmental Monitoring Centre,Xi’an 710054,China)
出处
《人民珠江》
2021年第6期59-69,共11页
Pearl River
基金
陕西省重点研发计划(2019ZDLSF06-01)。
关键词
突变识别
趋势分析
偏最小二乘回归
BP神经网络
汉江流域
abrupt recognition
trend analysis
partial least squares regression
BP neural network
Hanjiang River Basin
