A hybrid soft computing approach based on feature selection for estimation of filtration combustion characteristics

Abstract

Filtration combustion is a type of combustion in which exothermic reactions occur within a porous matrix. In recent years, some researches were performed on the use of noncatalytic partial oxidation method for hydrogen production in reactors based on filtration combustion. In this paper, a new technique is presented for estimation of some important characteristics of filtration combustion process. Hydrogen production, peak combustion temperature, and wave velocity are three main variables which are estimated by suggested forecasting engine. The parameters which have direct effect on above-mentioned variables are: equivalence ratio (\(\varphi\)), reactant inlet velocity (V), bed porosity (\(\varepsilon\)), heat conductivity (K), and the heating value of fuel. The estimation process is realized though two steps: feature selection and regression. At first step, the features are ranked according to their importance degree. The well-known Gram–Schmidt orthogonalization feature selection is employed for ranking of effective input parameters. In the second step, extreme learning machine (ELM) and support vector machine (SVM) are utilized as regression cores of forecasting engine. For each feature subset, the estimation accuracy is calculated in order to find the most important feature subset. The obtained results show that the ELM yields the better performance for the prediction of peak temperature and wave velocity as compared to SVM.