Stable recursive canonical variate state space modeling for time-varying processes

Abstract

An adaptive recursive process modeling approach is developed to improve the accuracy of modeling time-varying processes. We adopt the exponential weighted moving average approach to update the covariance and cross-covariance of past and future observation vectors. Forgetting factors are adjusted in the recursive modeling process based on the residual of model outputs. To ensure the stability of the identified model, we introduce a constrained nonlinear optimization approach and propose a stable recursive canonical variate state space modeling (SRCVSS) method. The performance of the proposed method is illustrated with an open-loop numerical example and simulation with the closed-loop data from a continuous stirred tank heater (CSTH) system. The results indicate that the accuracy of proposed SRCVSS modeling method is higher than that of state space modeling with traditional canonical variate analysis.