Prediction Error Approach to Feedback Control Performance Assessment

Abstract

As discussed in previous chapters, performance assessment of multivariate control systems requires a priori knowledge about the process models, such as interactor matrices. In recent years, there has been growing interest in reducing the complexity of the a priori knowledge requirement, such as the work by Ko and Edgar (2001) [25], Kadali and Huang (2003) [26], and McNabb and Qin (2003) [27]. Although these attempts have reduced the complexity of the a priori knowledge requirement to some extent, they all require certain information that is computationally simpler but fundamentally equivalent to the interactor matrices; for example, the open-loop process Markov parameter matrices, the lower triangular Toeplitz matrix, or the multivariate time delay (MTD) matrix. That is, they all require a priori knowledge that is beyond the pure time delays between each pair of the inputs and outputs. Harris et al. (1996, 1999) [112, 21] introduced an extended horizon performance monitoring approach without using the interactor matrix. Kadali and Huang(1998) [137] and Shah et al. (2002) introduced curvature measures of multivariate performance without relying on the interactor matrix. Most recently, Huang et al. (2004) [115] proposed an algorithm for multivariate control performance assessment with a priori knowledge of the order of the interactor matrix (OIM) only.