Equipment Fault Diagnosis Under Sample Imbalance Based on NSGAII-RF
In the defect and fault diagnosis of intelligent manufacturing equipment,under-sampling is easy to lose important sample information,while over-sampling is easy to introduce redundant information.To solve these problems,non-dominated sorting genetic algorithm with elite strategy(NSGAII)is integrated into random forest(RF)algorithm.The multi-objective genetic algorithm is used to replace the bootstrap sampling method and feature selection method in the conventional RF algorithm to generate diversified feature subsets and sample subsets with good performance,so as to generate decision trees with strong diversity and high accuracy.At the same time,the optimal number of decision trees is determined to optimize the performance of RF algorithm and improve the classification accuracy.The fault diagnosis experiment is carried out based on the multi-class and highly unbalanced equipment fault dataset,and the proposed algorithm is compared with the integrated learning algorithm combined with synthetic minority oversampling technique(SMOTE)and other algorithms.The experimental results show that the fault diagnosis accuracy of the proposed algorithm is higher than that of the comparison algorithms.
NETL
NSTL
万方数据
