Hybrid Feature Generation and Selection with a Focus on Novel Genetic-Based Generated Feature Method for Modeling Products in the Sulfur Recovery Unit

Nowadays, varied challenges in modeling variables in chemical industries have led to the implementation of designing intelligent systems to tackle these issues. In the sulfur recovery unit, online controlling of pollutants in different streams, which are extremely significant in the environmental context, is an arduous task and classic methods have numerous limitations. This paper aims to develop a reliable method in the field of feature engineering, meaning genetic-based generated features (GBGF); furthermore, it studies four feature selection methods and represents some innovations in the performance of the ant colony method (BACO), providing a powerful technique, the hybrid feature generation and selection (HF-G&S), for the prediction of different variables. Using the HF-G&S method, compared to the classic methods in feature generation and selection, the accuracy of the models increases significantly. The main reason for this achievement is introducing the GBGF technique and the simultaneous construction of compelling features, identifying essential features and repeating this cycle. Ultimately, by implementing this technique, the prediction precision of the SVR model for the H2S and SO2 concentration in the output flow reached over 98 and 97 percent, respectively.