Aiming at the problem of insufficient consideration of incomplete data sets in current wind power prediction methods, this paper is committed to achieve accurate prediction of wind speed under incomplete data sets, and improve the robustness and reliability of wind power system. Therefore, a wind speed prediction method based on improved stochastic configuration networks (ISCNs) is proposed. The Differentiated Creative Search (DCS) algorithm is integrated with the stochastic configuration networks (SCNs) to enhance its nonlinear fitting and global optimization capabilities. Wind speed prediction experiments are conducted using the established model in Matlab. The proposed method demonstrates its effectiveness and superiority compared to the original SCNs, as verified by comparing the predicted results with actual wind speeds.