The application of energy storage on wind power prediction error compensation is becoming an ideal scheme now. However, limited by energy storage technology, the implement speed of is quite slow. The error distribution characteristics of wind power prediction are studied based on several improved typical wind power prediction error distribution for the reason that the demand for energy storage capacity depends largely on the cumulative error distribution characteristics. The relationship between the capacity allocation requirements and the error distribution characteristics of the energy storage system is obtained on the whole. Then, a wind/storage power grid-connected plan forecasting optimization strategy based on the average absolute error is proposed to improve the forecast precision of wind power grid-connected generation and reduce the capacity demand of energy storage configuration. The energy storage capacity allocation with different confidence levels is obtained by the kernel density estimation method, which provides a reference for the actual wind farm. Finally, an example is given to test and verify the economic feasibility of the proposed method.