基于LM-BP神经网络的天然年径流一致性修正研究

Analysis on consistency correction of natural annual runoff based on LM-BP neural network

【目的】人类活动改变流域下垫面条件而导致河川径流发生变化,为保证径流系列的一致性,需进行天然径流系列一致性修正。【方法】基于武山水文站1956~2016年天然径流量,采用R/S分析、曼-肯德尔、斯波曼秩次相关等方法分析天然年径流变化趋势和突变点,并应用LM-BP神经网络预测模型对突变点前的序列值进行一致性修正,建立输入变量为多个影响因子的天然年径流预测模型。【结果】与传统降水径流法相比,通过一致性修正方法分析得出武山水文站1956~2016年天然年径流量值为4.422亿m^(3),结果基本合理。【结论】采用LM-BP神经网络一致性修正的方法,修正后的径流系列更具代表性、可靠性,可为水资源管理、防汛抗旱提供科学依据。

【Objective】The human activities change the underlying surface conditions of the basin,resulting in the change of river runoff.It is necessary to correct the consistency of natural runoff series in order to ensure the consistency of runoff series,.【Method】Based on the natural runoff data of Wushan Hydrological Station from 1956 to 2016,the variation trend and mutation point of natural annual runoff were analyzed by using R/S analysis,Mann Kendall,Spoeman rank correlation and other methods,and the sequence value before the mutation point was uniformly modified by using LM-BP neural network prediction model,and a natural annual runoff prediction model was established with multiple input variables.【Result】Compared with the traditional precipitation runoff method,the natural annual runoff value of Wushan Hydrological Station from 1956 to 2016 is 4.422×108 m3 through the analysis of consistency correction method.【Conclusion】The revised runoff series using the method proposed in this paper is more representative and reliable,which provides a scientific basis for water resources management,flood control and drought relief.