输入方式对BP模型年径流预测影响研究

Impact of Input Methods on Annual Runoff Forecast Based on BP Model

针对输入方式对BP模型的影响问题,基于岚河流域降雨、蒸发、径流等资料,设置了不同的资料情况及数据输入方式,分别构建了BP模型。研究结果表明,与未同时考虑降雨、蒸发、径流等影响因素的模型相比,同时考虑了三者影响的模型精度更高;以各资料前三年数据作为输入量较以前两年和前四年的数据作为输入量,紧邻下一年的径流数据作为输出量的模型的精度更高;既考虑了降水、蒸发、径流等影响因素又选用上述最佳的数据输入方式的模型精度最高。BP模型适用于干旱半干旱地区的岚河流域,其预报结果可用于该流域的水资源规划管理。

In view of the impact of input methods on BP model, a series of back-propagation （BP） artificial neural networks models for the Lanhe watershed are set up by using the data of precipitation, evaporation and discharge in different ways and input methods re- spectively. It is found that the BP model which takes precipitation, evaporation and discharge into consideration is more precise than the model which doesn＇t take; the model which takes every three years of the data as input and the runoff data of next year as output has higher accuracy than the model which takes every two or four years of the data as input and runoff data of next year as outputs the model which takes precipitation, evaporation and discharge into consideration and chooses the best data input mode has highest accuracy. The BP model is feasible to simulate the annual runoff and its result could be used for the water resources planning and management in the region.