Abstract
By adding a function to fade memories for past observation data to the Kalman filter which has often been used as a time marching identification algorithm we developed an adaptive Kalman filter scheme. The rate of memory fading was defined by a forgetting factor multiplying to past data at each time step. In order to track fast variation in the system parameters the value of the forgetting factor should be small. On the other hand, to remove the random noise from the signal, the number of sample points used at any time should be large enough, that is, the large value of the forgetting factor should be used. The Akaike-Base Information Criterion was applied to determine the optimal forgetting factor.
