Abstract: For nonstationary time series with outliers and strong noises, in order to increase the accuracy and robustness of wavelet threshold denoising in the low frequency signal decomposition, the arc-tangent function transformation is employed to transform the original nonstationary time series to a new more stationary series. The SNR （Signal-to-Noise Ratio） is improved and the negative effect of outliers is insulated. The suppressing capability of arc-tangent function transformation to outliers is explained analytically, and its parameters are optimized. Numerical experiments/simulations of a given signal prove that the arc-tangent function transformation can improve the resistance against outliers, and enhance the performance of wavelet threshold denoising. An example of wind speed time series denoising shows that the new method can provide a more real wind speed signal by eliminating the outliers with larger amplitudes.

Key words: nonstationary time series, wavelet threshold denoising, arc-tangent function transformation

摘要： 对存在野值和较强噪声的非平稳时间序列，为提高小波阈值去噪在低频信号分解中的准确性和稳健性，采用反正切变换将原始非平稳时间序列转换成新的更平稳的序列，以提高信噪比，并抑制野值的不利影响。通过解析分析，解释了反正切变换抑制野值的成因，并优化了其中的参数。采用已知信号的数值实验（仿真）证明，反正切变换能够提高对野值的抵抗能力，改善小波阈值去噪的效果。应用于风速时间序列的去噪表明，新方法能够有效抑制风速里的野值，获得更符合实际的风速信号。

关键词: 非平稳时间序列, 小波阈值去噪, 反正切变换