Identification of Hammerstein LSSVM-ARMAX systems and its application in Continuous Stirred Tank Reactor

摘要

A new Hammerstein structure is proposed to building the black-box model of continuous stirred tank reactor (CSTR). This approach uses least squares support vector machines (LSSVM) with a radial basis function (RBF) kernel to represent the memoryless static nonlinearity and uses auto-regressive moving average exogenous (ARMAX) models to represent the linear dynamical system. The twostage identification algorithm is used to estimate the parameters of Hammerstein LSSVM-ARMAX systems. The results of CSTR identification show that the proposed Hammerstein model is more accuracy than the traditional Hammerstein modej, and be able to accurately approximate the dynamic behavior of the actual CSTR.