Técnica de seleção de variáveis em problemas de classificação de falhas aplicada em processo industrial usando o algoritmo genético MOEADD

Abstract

In this work we propose a method of variable selection called MOEADD-KNN-M, which is based on the genetic algorithm MOEADD (Evolutionary Many-Objective Optimization Algorithm Based on Dominance and Decomposition), on the classification algorithm KNN (K-nearest neighbors), and in adapted genetic operators. The approach adopted in the proposed algorithm is bi-objective, where one objective is to minimize the amount of solution variables and another objective is to minimize the failure classification error rate. Experiments were performed with the proposed method using data from a real petrochemical industrial process, called Tennessee Eastman for failure classification, and the results were compared with other algorithms. The results showed that the proposed method leads to solutions with low classification error and low number of sensors, which are the quantities sought to be minimized. Thus, this approach has shown promise for application in the selection of variables in fault classification problems in industrial processes.