In order to smooth the fluctuating wind output, it is often optimized by energy storage devices. At the same time, it is especially important to determine the value of active wind power that is injected to power systems so that the capability of stored energy can be allocated reasonably. At present, the main ways to calculate the reference power of grid-connected power are low-pass filter and spectrum compensation methods which always have some drawbacks. In this paper, a method that combines empirical mode decomposition with wavelet analysis is proposed to get reference power of grid-connected active power, which can restore the original power. Then synthesizing some other situations such as load conditions of the battery, efficiency and so on, the minimum storage energy is determined. Compared the minimum capacity allocation by traditional low pass filter smooth method, the effectiveness of the proposed method can be verified. This work is supported by National Natural Science Foundation of China (No. 51422701).