Tail Gas Quality Warning System in a Sulfur Recovery Unit based on H2S and SO2 Concentration Soft Sensor utilizing Multi-State-Dependent Modeling Method

The design and development of data-driven soft sensors is important to predict the concentration of perilous pollutants in industry effluents to protect environmental health. The aim of this research is to design a tail gas quality warning system in the sulfur recovery unit (SRU) based on H2S and SO2 concentration soft sensor utilizing multi-state-dependent modeling method. The SRU in the petrochemical plant of ERG PETROLI, located in Italy, is selected as the study region for implementation of the warning system. The generalized random walk- multi-state-dependent parameter method (GRW-MSDP) for soft sensor design is proposed. The GRW-MSDP estimation system is based on multi-state-dependent modeling method by utilizing the extension of the generalized random walk model. The method has been developed by utilizing the algorithms of extended Kalman filter (EKF) and fixed interval smoothing (FIS). The quality warning system of tail gas based on the estimated concentrations of SO2 and H2S sends instructions to adjust the ratio of air to feed flow in the reaction furnace of SRU by plant operators. The results indicate that the proposed estimation system can be efficient in dealing with process non-linearity, high-dimensional values, and random missing data. The comparative discussion of GRW-MSDP technique performance with different soft sensing methods shows that the designed soft sensor model is more reliable with fewer input variables, lower complexity and relatively higher prediction accuracy. Furthermore, the great efficiency of the designed quality warning system is obvious from the good accuracy and F1-score values of 99.4% and 0.8951, respectively.