Soft analyzers have been increasingly accepted as an alternative to physical ones in the chemical industry to infer and improve the product quality. In this study, an adaptive least-squares support vector regression (ALSSVR) algorithm is proposed for the issue of nonlinear multi-input−multi-output process modeling and applied to soft chemical analyzer development. The ALSSVR algorithm adopts the moving window scheme and a two-stage recursive learning framework to trace the time-varying dynamics of a process. The useless sample (i.e., a node of analyzer model), while not the oldest one, is selectively deleted from the model topology, using the fast leave-one-out cross-validation criterion. Consequently, the updated model can exhibit good generalization ability and trace the process characteristics effectively. Besides, a variable moving window is proposed, so its size can be adaptively adjusted, relative to process changes. The ALSSVR-based soft analyzer is then applied to an industrial fluidized catalytic cracking unit to predict its three key product yields. The obtained results show that the presented ALSSVR method is superior to conventional recursive least-squares support vector regression (RLSSVR) approaches. The maximal root-mean-square error (RMSE) of all product yields is <1.5 and the maximal relative prediction error (RE) is ∼7%, which can be acceptable in industrial practice for the intended objective of soft analyzers.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.