摘要 |
PROBLEM TO BE SOLVED: To obtain a robust analysis result with accuracy which is approximately the same as a normal UFK(Unscented Kalman Filter) even when the statistic of a noise is unknown by using a non-linear model as a target.SOLUTION: A state space modeling section 31 is configured to define a state space model determined by a state update formula using a system noise and an observation equation using an observation noise. A state estimation section 32 is configured to, when performing UFK on the basis of the state space model and the time series data of an observation value, generate a sigma point sequence on the basis of a normalized covariance matrix ^P' normalized by a total sum tr(Q)+r of the diagonal components of a covariance matrix Q of the system noise and a dispersion value r of the observation noise and weight changed by using the total sum tr(Q)+r, and to perform state update and observation update by using the normalized covariance matrix ^P', a normalized covariance matrix P'(overline) normalized by the total sum tr(Q)+r, a normalized covariance matrix ^P', a normalized system noise Q' and a normalized observation noise r' to successively estimate a state vector X and a covariance matrix P. |