Feature Selection for Improvement the Performance of an Electric Arc Furnace

Feature selection has as principal goal to find a representative space of minimal size from original set of larger size. Several research works have been developed on this problem. This paper presents Support Vector Machine-Recursive Feature Elimination (SVM-RFE), Genetic Algorithms (GA), and Differential Evolution (DE) algorithms for feature selection from a database of an Electric Arc Furnace (EAF) for locating variables related to energy consumption. The proposal suggests merging the coefficients generated by LDA and SVM, employing them in RFE to obtain the ranks for each discriminant variables. The measure of accuracy and error rate for each algorithm is presented like a decisive score for choosing the subset obtained by the algorithm with the best performance. The variables selected were adjusted for the EAF control system achieving the reduction of the energy consumption to 3.5% in a steel castings and 1 minute reduction of the connected EAF in a steel casting.