Automated multivariable system identification and industrial applications

A technology for automatic identification or modeling of multivariable systems from observational input/output data has been applied to a number of difficult industrial applications resulting in major improvements. A tutorial presentation is made of the primary elements of this technology, followed by a discussion of significant applications. This technology involves the identification of linear, time invariant dynamical processes with noise disturbances and possible feedback and includes determination of the system state order. The basic method involves a canonical variate analysis (CVA) that, for each potential state order, gives an optimal statistical selection of the system states. The computation involves primarily a singular value decomposition. The accuracy of the method is close to the optimal lower bound achieved by maximum likelihood (ML) for large samples. The CVA method has been widely applied in both academic and industrial settings to a variety of problems involving high-order multivariable systems that are possibly unstable, nonminimum phase, and/or involve nonstationary noise, stiff dynamics, unknown feedback and delays. This paper describes applications to spectral analysis, system monitoring, detection of abrupt system changes, and adaptive modeling and online adaptive control. Automated multivariable system identification is a technology critical to enabling widescale industrial automation and adaptation of control systems.