Abstract: In order to reduce the negative impact of data set's redundant features on classifier's training speed and detection accuracy, which is used for intrusion detection, the binary gray wolf optimization (BGWO) mutation probability is analyzed and its mutation related vector's expression is reconstructed, improving BGWO's mutation mechanism, speeding up feature dimensionality reduction, and reducing classifier's training time. In addition, the iterative decision-making form of PSO was integrated, enhancing BGWO's optimization capabilities. Hybrid BGWO was adopted for wrapped feature selection, making data set's feature structure more suitable for the decision tree classifier. The NSL-KDD data set tests show that this method has good detection accuracy for DoS, Probe attack traffic, and is suitable for data sets with balanced data distribution.