Prediction Methods for Time Series Data with Many Missing Data Based on RLS Method.
Kaname Seto, Kohei Arai · Journal of the Japan society of photogrammetry and remote sensing · 1999
There are two parameter tuning algorithms, time update and measurement update algorithms for parameter estimation of Kalman filter. Two learning methods for parameter estimation of Kalman filter are proposed based on RLS (Recursive Least Square) method. One is the method without measurement update algorithm (RLS1) . The other one is the method without both time and measurement update algorithms (RLS2) . The methods are applied to the time series data of DMSP/SSM/I data with a plenty of missing data. It is found that the proposed RLS2 method shows smooth and fast convergence in learing process in comparison to the RLS1.