抄録
A limit of the RLS (Recursive Least Square) method for the parameter estimation of Kalman Filter is clarified based on the simulated time series of data from Time Variant AR (Auto-Regressive) model. Learning process is converged for the stational time series of data while the process is diverged for the non-stational time series of data. It is also found that the required time for the convergence depends on the degree of stationality of time series of data.
