A knowledge-based system for automated multivariate nonlinear process identification

A knowledge-based system for automated multivariable nonlinear process identification (ANPI) has been developed to provide a dynamic system modeling tool for chemical processes. ANPI is capable of detecting outliers in the process data, testing the system nonlinearity, identifying optimal linear/nonlinear dynamic models, and conducting model validation automatically. Several new nonparametric system identification algorithms, such as the generalized multivariable adaptive polynomial synthesis (GMAPS), the multivariable adaptive regression spline (MARS), the multivariable II method (MPIM), the multivariable projection pursuit regression (MPPR), the improved group method of data handling (IGMDH), and the generalized adaptive threshold models (GATM), give ANPI rich contents and vitality. The optimal adaptive robust M-estimator enables ANPI to estimate parameters for a board class of noise distributions, not just for normal distribution. The integration of two novel outliers detection algorithms (CMDMO criterion method and leave-k-out diagnostics), the nonparametric system identification algorithms, and the optimal adaptive M-estimator makes ANPI not only a powerful system identification tool, but also a new data analysis methodology.;ANPI has been implemented by using G2 real-time expert system shell. The architecture of the knowledge-based system is based on a hybrid paradigm that uses an item hierarchy to present the system with a rule based inference mechanism. ANPI's open system structure gives it the ability to modify and add any function. ANPI was validated by applying it to several industrial processes and a number of simulated systems.