[关键词]
[摘要]
以湖北省宜昌市某一小流域为例, 详细介绍了研究区径流模拟的BP 神经网络模型建模方法, 并利用模型对 研究区的日径流进行了模拟预测研究。首先, 根据研究区降雨2径流的时间分布特征, 确定了对丰水期、枯水期分别 建模的建模方案; 接着分析了流域产流的主要影响因素, 确定了将前五日径流量、前三日降雨量、当前降雨量和蒸散 发量为作为模型的输入变量; 并在反复试验的基础上, 选取了合适的模型结构和学习效率参数; 最后, 利用确定性系 数对模型的径流预测精度进行了评定。结果表明, 针对丰、枯水期分类建立的BP 神经网络模型克服了以往模型对 极值事件模拟精度较差的不足, 对高流量和低流量的模拟精度高。
[Key word]
[Abstract]
The BP neural net wo rk model method for runo ff modeling was int roduced and the mo del was a pplied to simulat e the daily runo ff in a w atershed of Yichang in H ubei Prov ince. First, based on the tempo ral distributio n of rainfall2r uno ff in the study area, t he modeling approaches fo r the w et and dry seasons w ere developed separ ately . Second, the main facto rs affect ing the r un2 o ff wer e analyzed, and t he input v ariables of the model included the r unoff o f f ive previous days, rainfall of three previous days, cur rent rainfall, and cur rent evapotr anspir ation. Thir d, the appr opriat e model str ucture and learning eff iciency parameters wer e determ ined thro ug h trial2and2er ro r tests. Finally, det erminacy coefficient w as used to assess the accuracy o f simulation results. The results show ed t hat the BP neural netw o rk models of the wet and dry seasons over come the disadvantages of low accuracy in pr evio us models w hen simulat ing the ext reme event s, and the BP neur al netwo rk model can simulate the high and low r unoff co nditions wit h hig h accuracy .
[中图分类号]
[基金项目]
中国地质科学院水文地质环境地质研究所基本科研业务费专项( SK201312)