Research on Intelligent Grid Information Attack Identification Method Based on MRMR-XGBoost

This paper proposes a power grid information attack identification method based on a combination of maximum relevance minimum redundancy and XGBoost to address the problem of accurately identifying attack types based on changes in measurement data after the smart grid is subjected to information attacks. Firstly, by using the K-means-smote oversampling method, the measurement data containing attack samples is balanced and preprocessed to solve the problem of imbalanced attack samples. Secondly, utilizing the Maximum Relevance Minimum Redundancy feature selection method, the optimal subset of representation features for information attack events is extracted, reducing data dimensionality and improving the recognition efficiency of information attacks. Then, establish an XGBoost classifier to classify and recognize three types of attack states and normal states, and evaluate the recognition performance of the model using indicators such as accuracy and recall. Finally, experimental validation demonstrates that the developed cyber-threat detection framework achieves marked enhancement in identification precision for adversarial interventions in smart grid networks, while maintaining robust generalization capabilities across varied attack scenarios.