Identification of Multicomponent Nonstationary Chemical Processes Using Dynamic Neural Networks

摘要

The identification problem for multicomponent nonstationary ozonization processes with incomplete observable states is addressed. The corresponding mathematical model containing unknown parameters is used to simplify the initial nonlinear modeland to derive its observability conditions. To estimate the current concentration of each component, a dynamic neuro observer is suggested. Theorems concerning the observation error bound are presented. Based on the obtained neuro observer outputs, thecontinuous time version of LS-algorithm, supplied by special projection procedure, is applied to construct the estimates of unknown chemical reaction constants. Simulation results related to the identification of ozonization process illustrate theeffectiveness of the suggested approach.