A New Algorithm for On-Line Multivariate ARMA Identification Using Multimodel Partitioning Theory

In this study an adaptive algorithm for multivariate (MV) ARMA model order identification and parameter estimation is presented based on the multi-model partitioning theory (MMPT). The method proposed is based on the reformulation of the problem in the standard state space form and on implementing a bank of Kalman filters, each fitting a different order model. The first step will be to select the order of the MV ARMA model using the MPPT, for general (not necessarily Gaussian) data pdf´s. The assumption made is that the true model order is theta (lambda, lambda) where lambda = max (p, q), p is the order of the AR component and q the order of the MA component. The second step will be to estimate the AR and MA coefficients and the actual values of p and q.