摘要
针对河套灌区地下水位预测问题,结合小波变换的时频局部特性和神经网络的逼近功能,构建了两种不同耦合方式下小波和BP神经网络相结合的小波网络模型,比较了不同耦合方式下小波网络模型与单纯神经网络模型的预测效果。两种耦合方式下的小波网络模型模拟结果均比单纯使用人工神经网络模型更接近实测值,对低频信号序列及高频信号序列分别进行神经网络模型预测后再进行重构的预测方式比直接将小波分解的多级信号与神经网络结合的预测方式具有更好的预测效果。
In view of forecasting groundwater level of irrigation distinct,two coupling wavelet and BP network models are established by combining time-frequency local property and approximate function of neural networks.The prediction effect of wavelet network model and single neural network is compared under different coupling mode.The comparison indicates that the error of simulation with two-coupled models is smaller than that with single artificial neural network model.The results also show that the coupling mode,in which signals are reconstructed after detailed stationary time series and smoothed non-stationary time series are forecasted with neural network model,is better than the coupling mode,in which the decomposed signals are forecasted with the neural network model.
出处
《水电能源科学》
北大核心
2012年第5期16-20,共5页
Water Resources and Power
基金
国家自然科学基金资助项目(41006046)
关键词
地下水位预测
小波分析
BP神经网络
耦合模型
groundwater level forecasting
wavelet analysis
BP neural network
coupling model
