BP 网络日径流序列预测的NLP 适用条件研究

对非线性预处理在人工神经网络日径流预测中的适应过程进行了仿真和模拟．提出了非线性预处理（NLP）适用条件的解算思路，通过实测数据和模拟数据，研究了NLP的适用条件。推导出NLP在神经网络SISO系统中适合于日径流预测，不适用于周平均流量序列、旬平均流量序列和月平均流量序列的预测，提出了判断NLP神经网络SISO系统进行日径流预测的有效性标准——多年日径流拐点14百分位．并通过广西平乐水文站和四川宝珠寺水文站1973～2001年的日径流量进行对比预测，验证了该标准是合理的。

Adaptivity of NLP in daily runoff prediction with BP ANN

This paper simulated the adaptation process of non-linear prediction ( NLP) in Back Propagation( BP) Artificial Neutral Network ( ANN) for daily runoff prediction, and proposed the calculation approach of adapting condition of NLP. It was deduced that the NLP can well run in Single Input Single Output ( SISO) mode for daily runoff predict ion, but would become void for predicting weekly-mean runoff, 10-day mean runoff andmonthly mean runoff through measured and simulated data. The authors proposed that the forecasting validity standard for prediction of other rivers by NLP in SISO ANN would be the inflexion position of 14 percent of mult-i year mean daily runoff. This paper examined the standard by predicting and comparing of daily runoff during 1973 through 2001 at Pingle station in Guangxi province and Baozhusi station in Sichuan province. The results show that it is rat ional.