干旱内陆河流域月径流的神经网络模拟

Artificial neural network model for monthly streamflow in an arid-inland basin of China

针对干旱内陆河流域地表径流形成、消耗的特点,采用1956-2003年的水文气象、灌溉用水等资料建立了用于模拟出山口月径流量、下游径流月径流量的分布式神经网络（ANN）模型。模型的输入为各子流域及中游本月降水、潜在蒸发蒸腾量、上月降水、潜在蒸发蒸腾量、中游灌溉面积、灌溉定额,输出为月径流量.检验结果表明,本文所建立的分布式ANN模型可以有效模拟干旱内陆河流域月径流,模型用于模拟子流域出山口径流的误差为0.18×10^7-0.42×10^7m^3,用于模拟下游径流的误差为0.52×10^7m^3.与单一ANN相比,尽管分布式ANN的输入不需实测出山口径流,但模型的精度没有明显减小.分布式ANN为研究干旱内陆河流域气候变化及农业活动对地表径流过程的影响提供了有效的工具.

According to the streamflow formation and depletion in arid-inland area, a distributed ANN model for monthly streamflow were trained and tested using hydrological and irrigation data with period of 1956-2003. Inputs of the model include monthly precipitation and evapotranspiration, last monthly precipitation and evapotranspiration, irrigation area, irrigation quantity in upper and middle reaches. The output of the model is monthly streamlow. Testing showed that the model can simulate effectively monthly streamflow in arid-inland area. Furthermore, errors of the model were 0.18×10^7-0.42×10^7m^3 ma and 0.52 × 10^7 ma respectively for upper and down reaches of the basin. At the same time, a single ANN model were trained to simulate monthly streamflow in the down reaches and input of the model include measured streamflow in the upper reaches. Compared to the single ANN model, precision of the distributed ANN didn＇ t decrease although inputs of the model didn＇t include measured streamflow in upper reaches. The distrib- utive ANN can be used to investigate response of streamflow on climate change and agricultural activities.