Renewable energy generation forecast is important for power grid operations. As for modeling this random and volatility time series, this paper studies three representative multi-scale decomposition algorithms: wavelet transform, empirical mode decomposition, and mathematical morphology. Taking the wind power time series as an example, their decomposition performances are evaluated. Based on information entropy theory and Hilbert transform, an evaluating indices system for algorithm performance is constructed from three aspects, the complexity, amplitude characteristics and frequency characteristics of decomposition results. The proposed evaluating method is verified by the forecast models based on artificial neural network and decomposition algorithms, and the results show that the proposed evaluation system is effective, and provides guidelines for selecting decomposition algorithms.