Feature Selection from Iron Direct Reduction Data Based on Binary Differential Evolution Optimization
Abstract
Nowadays increasing growth in the production of the steel products makes automatic identification of effective parameters in determining the quality of the output product is very important. In this regard, level II automation plays an important role. In this study, a novel method has been proposed to identify effective parameters in determining the purity of sponge iron in the process of Iron Direct Reduction. In the proposed method, differential evolution (DE) optimization algorithm with the binary approach has been used in order to identify the subset of effective parameters with the lowest estimation error in determining the purity of sponge iron. The binary differential evolution algorithm is combined to the Least Squares- Support Vector Machine (LS-SVM) regression method to candidate a subset of the effective parameters. Implementation of the proposed algorithm on data obtained from the practical project (Bardsir steel complex) confirms the effectiveness of the proposed method so that by choosing the effective parameters, the ability to estimate the sponge iron purity with 98.8% accuracy (1.2% estimation error) has been attained.