Abstract: False data injection attack can bypass the traditional bad data detection. This paper studies an effective detection method for false data injection attack in smart grid. Aimed at the poor robustness of diffusion Kalman filter, an improved diffusion Kalman filter algorithm was proposed by adding exponential weighting function. On this basis, based on the obtained state estimates, combined with the traditional weighted least square estimation algorithm and Euclidean distance, a new false data injection attack detection method was proposed. This method reduced the impact of attacks on system state estimation, and could effectively detect false data injection attacks. A 3-node power system was simulated, and the results verified the effectiveness and applicability of this detection method.