Aiming at the poor operation of doubly fed induction generator under unbalanced grid voltage, neural network sliding mode controller, which combined neural network control with secondorder sliding mode control, was applied to direct power control of doubly fed wind generator. A secondorder sliding mode controller was designed. The secondorder sliding mode could effectively weaken the chattering of traditional sliding mode control. At the same time, a radial basis function neural network was designed to approximate the uncertain part of the system. Finally, based on Lyapunov stability theory, the adaptive law of the neural network weight was deduced, and the stability of the control system was proved. The simulation results showed that the proposed control strategy could effectively control the active power, reactive power and stator current, and weaken the chattering in traditional sliding mode control.