摘要： 针对小波网络计算过程中出现大量冗余的特点，提出一种有偏小波网络模型。该模型在网络中添加一个偏倚层，通过自适应调整参数，增加自由度，减少计算冗余，同时采用基于多分辨率的初始化框架，加快了收敛速度，提高了计算精度。仿真计算表明，有偏小波网络能够反映非线性时间序列的内在特性，得到较好的径流预报结果，是一种有效的非线性建模方法。

关键词: 有偏小波网络, 偏倚函数, 非线性时间序列, 径流预报

Abstract: Biased wavelet network is proposed for the purpose of reducing the redundancy in general wavelet network. Including a biased layer, the presented network can adjust parameters adapt to a particular problem, which adds more degrees of freedom and reduces calculating redundancy efficiently. The network initialization based on a multiresolution scheme is employed to accelerate convergence and increase approximation accuracy. The simulation results show that, the biased wavelet network is an effective model for reflecting the inherent characteristics and obtaining better forecasts of nonlinear time series.

Key words: Biased wavelet network, Biased function, Nonlinear time series, Streamflow forecast