Recursive relaxation identification of linear multivariable systems with its parallel algorithm (Q1364496)

The identification problem of a linear discrete-time multivariable system, described by the state-space equation and disturbed by zero mean white Gaussian noise, is considered. A method for simultaneous least-squares recursive identification of system parameters and system state vector estimation by Kalman filtering from input-output measurements is proposed and discussed. The essence of the method lies in the fact that only linear problems are to be solved within the algorithm. The method comprises three interacting recursive schemes: for estimating state and output equation parameters and system state vector. The computational complexity and hardware requirements of the routine are analysed and its superiority over the extended and modified extended Kalman filtering is shown. Parallel processing ensures applicability of the method for fast real-time adaptive identification.