Fuzzy C -means（FCM） algorithm is effective for the power load characteristics classification. Since FCM algorithm is prone to fall into a local optimum zone and sensitive to the initial values， an improved algorithm is proposed based on differential evolution（DE） adaptive optimization. The improved algorithm uses random parallel search of DE to replace steepest descent gradient of FCM， and integrates an adaptive adjusting strategy for control parameters and non-optimal individuals to achieve strong global searching ability and better robustness. The case study indicates that DE optimized adaptive FCM is capable of improving the validity of cluster results， reducing the dependence on the initial inputs and outperforming FCM under alternating cluster numbers， thus suitable for situations requiring higher accuracy of load characteristics classification.