Abstract: Standard Kalman filter strongly depends on the system mode and the statistical characteristic of noise，unfortunately，an accurate mathematical model of system is difficult to set up，so a new adaptive algorithm is presented based on federated felting and the adaptive estimation.The new method can make certain the value of error covariance by calculating the rate of the real value and theoretical value of residual covariance，and then it can change fading factor，and enhance the filter’s capability.By simulations and calculations，it is shown that the improved federated filter can effectively use observation information’s and adaptively adjust some parameters of the system.