A data-driven soft sensing approach for quality prediction in sulfur recovery unit using state dependent parameter models
1Center for Process Integration and Control (CPIC), Department of Chemical Engineering, University of Sistan and Baluchestan, Zahedan
In recent years, using data-driven soft sensors for the purpose of both monitoring and control has gained much popularity in process industries. The goal of this paper is to present a data-driven soft sensor based on state dependent parameter models, which can effectively handle time-varying characteristics of nonlinear stochastic processes. The model identification process exploits the recursive fixed interval smoothing algorithms, combined with special data re-ordering and back-fitting procedures, to obtain estimates of any state dependent parameter variations. The proposed soft sensor is applied to an industrial sulfur recovery unit for prediction of H2S and SO2 concentrations. The results show that the proposed model can accurately predict the process qualities as well as their rapid changes, which make it suitable for online deployment. While the proposed model uses less parameters and input variables and has the simple structure, it is more computationally efficient and memory-saving. The result of the model has been compared with the result of other conventional soft sensing techniques such as PLS, MLP neural network, RBF neural network and NF systems for sulfur recovery unit, which indicate that the model has the smallest RMSE value.
soft sensor; data-driven models; state dependent parameter; Sulfur recovery unit; quality prediction