Stochastic output-only state space modeling based on stable recursive canonical variate analysis

Abstract

An adaptive recursive stochastic output-only state space modeling approach is developed to improve the accuracy of modeling time-varying processes. The exponential weighted moving average approach is adopted to update the covariance and cross-covariance of past and future observation vectors. A novel method for adjusting forgetting factors based on the concept of angle between subspaces is proposed. To ensure stability of the identified model, we propose a constrained weighted recursive least square approach and propose a stable recursive canonical variate analysis (SRCVA) method. The performance of the proposed method is illustrated with simulation of the Tennessee Eastman (TE) process. Simulation results indicate that the accuracy of proposed SRCVA modeling method is superior to that of stochastic output-only state space modeling with canonical variate analysis.