Abnormal detection technology of industrial control system based on transfer 
learning

Wang, Weiping; Wang, Chunyang; Wang, Zhen; Yuan, Manman; Luo, Xiong; Kurths, Jürgen; Gao, Yang

doi:10.1016/j.amc.2021.126539

アイテム詳細

登録内容を編集ファイル形式で保存

一時保存へ追加

タグ情報を表示リリース履歴を表示詳細要約

公開

学術論文

Abnormal detection technology of industrial control system based on transfer learning

Authors

Wang, Weiping
External Organizations;

Wang, Chunyang
External Organizations;

Wang, Zhen
External Organizations;

Yuan, Manman
External Organizations;

Luo, Xiong
External Organizations;

/persons/resource/Juergen.Kurths

Kurths, Jürgen
Potsdam Institute for Climate Impact Research;

Gao, Yang
External Organizations;

URL

There are no locators available

フルテキスト (公開)

There are no public fulltexts stored in PIKpublic

付随資料 (公開)

There is no public supplementary material available

引用

Wang, W., Wang, C., Wang, Z., Yuan, M., Luo, X., Kurths, J., & Gao, Y. (2022). Abnormal detection technology of industrial control system based on transfer learning. Applied Mathematics and Computation, 412:. doi:10.1016/j.amc.2021.126539.

引用: https://publications.pik-potsdam.de/pubman/item/item_26397

要旨

In industrial control systems, industrial infrastructure is often attacked by hackers. Due to the serious sample imbalance in industrial control data, the traditional machine learning method has poor performance in anomaly detection. In this paper, TrAdaboost algorithm is applied to industrial control anomaly detection. The samples that are easy to classify are taken as the source domain data, and the samples with poor classification effect are taken as the target domain. The source domain data is used to guide the target domain data training. Then, we improve the traditional TrAdaboost algorithm from two aspects of initial weight and final classifier, and apply it to industrial control anomaly detection. Finally, the performance of the algorithm on two different industrial control data sets is verified. And the improved algorithm is compared with other traditional algorithms. The experimental results show that the improved TrAdaboost algorithm has a significant advantage in predicting categories with a small sample size. This algorithm can accurately identify a few abnormal samples. Moreover, the F1 value, recall and precision value of the improved TrAdaboost algorithm on the two data sets have been significantly improved. This indicates that the improved TrAdaboost algorithm greatly improves the overall prediction accuracy of the model.