Identification of Armax Models With Time Dependent Coefficients

Abstract

A method that uses input/output measurements is developed for the estimation of the coefficients of stochastic Time-varying AutoRegressive Moving Average with eXogeneous imputs (TARMAX) models. The TARMAX coefficients are expressed as linear combinations of a set of pre-selected functions. The model coefficients estimation method is fully based on linear operations, does not require initial guess values and is suitable for micro-computer implementation. The good performance of the estimation method is verified through numerical examples. A TARMAX model is also used to capture the dynamics of a detailed highly nonlinear model of an automobile hydraulic active suspension system. The TARMAX model is used to relate a desired force provided by a corner processor to the actual force generated by the hydraulic actuator. The TARMAX model is shown to provide good signal prediction ability.