In this paper,the control system of industrial robots generates adversarial samples by using Jacobian-based significance graph attacks and explores classification behaviors,explores adversarial learning algorithms to optimize the tar-get supervision model,and explores whether the supervision model trained by adversarial machine learning algorithms in the process of industrial control network defense can help improve the robustness of artificial intelligence algorithms to network security defense systems.Based on the network environment of the constructed industrial robot,this paper adopts direc-tional vulnerability attack and collects traffic data generated in the process through the gateway to optimize and improve the anti-machine learning algorithm.In general,among adversarial machine learning algorithms,the classification perfor-mance of Random Forest and J48,two widely used classifiers,decreased by 16 and 20 percentage points respectively,in-dicating that the robustness of the protection strategy adopted by adversarial machine learning algorithm is significantly im-proved,and it can effectively protect related systems against network attacks.
machine learningindustrial robotprotection strategyrobustnessintrusion detection system
NETL
NSTL
万方数据
