Adaptive Kalman Filter Based on Multiple Model Method

A kind of multiple model adaptive Kalman filter (MMAKF) was set up for the discrete time system without the statistic knowledge of measured noise. Multiple fixed Kalman filters according to the system with different fixed noise covariance matrices and a conventional adaptive Kalman filter were used to form a multiple model Kalman filter. An index switching function in the form of the function of output error of different Kalman filter was designed. At every sample time, the state value of the Kalman filter which gives the minimum value of index switching function will be switched as the estimated value of the state of system. From the simulation, it can be seen that the method proposed can improve the result of conventional Kalman filter greatly.