摘要
径流时间序列是时间、频率和流量等的函数。就目前对径流信息的认识来看 ,大体可将其分为确定性因素和随机性因素产生的信息。根据这些信息中隐藏的频率的不同 ,可利用小波变换原理将其分解到若干频率段上 ,即得到若干“近似”项和“细节”项。在运用小波原理对分解出来的各项重构之前 ,在各项上附加不同的阈1 值 ,即可达到对原始数据降噪的目的。本文将降噪后的径流数据用于BP网络的降雨径流模拟 ,网络的收敛性能和学习精度都得到了大大提高。通过实例验证表明 。
The runoff time series is the function of time, frequency and dischange. At the moment, the runoff information can be viewed as the product of determinate and stochastic factors. The information can be decomposed into some approximations and details by the wavelet analysis. Before reconstruction of the wavelet method, some thresholds were added on the every details so as to decrease the noise of the origin runoff data. BP network modeled with denoised data of runoff is presemted in this paper. It shows that the convergence ability and learning precision of the networks can be greatly improved, the generalization of BP networks can also be improved.
出处
《水力发电学报》
EI
CSCD
北大核心
2004年第4期5-10,共6页
Journal of Hydroelectric Engineering
关键词
洪水预报
BP网络
小波变换
降噪
flood forecast
BP networks
wavelet transform
denoising
