Enhancing Adversarial Robustness of Deep Learning-based Soft Sensors:A Defense Method Integrating Adversarial Training and Attack Detection
摘要
深度学习软测量(DLSS)已成为解决复杂工业过程关键变量测量难题的有效方法,但其极易遭受肉眼难以察觉的对抗攻击而输出虚假预测结果,进而危害生产安全.现有对抗攻击检测方法对微小扰动对抗样本(SPAS)普遍难以检出,且对抗训练防御方法仍广泛存在对抗鲁棒过拟合现象.针对上述问题,提出一种对抗训练与攻击检测融合(ATADI)的对抗防御方法:首先,理论证明仅采用 SPAS 进行对抗训练更有助于缓解对抗鲁棒过拟合.其次,在此基础上提出一种基于 SPAS的特征锚定式对抗训练(FAAT)方案,即 SPAS-FAAT.ATADI 中的攻击检测器在防御过程中仅检测较大扰动对抗样本(LPAS)以规避其难以检出 SPAS 的不足,SPAS 则被输入 SPAS-FAAT 后的 DLSS 以获得最终软测量结果.最后,在转子热变形软测量对抗攻防案例上进行了实验.结构消融实验结果显示,将对抗训练与攻击检测相融合可作为 DLSS 对抗防御的有效手段,各类攻击产生的 LPAS 均能以不低于 98.42%的准确率被检出,且 SPAS-FAAT 后的 DLSS 在 SPAS 与正常样本上的软测量结果均满足精度要求.ATADI方法显著增强了DLSS的对抗鲁棒性,为对抗防御方法的研究提供了新思路.
Deep learning-based soft sensors(DLSS)effectively measure critical variables in complex industrial pro-cesses.However,DLSS models are highly vulnerable to imperceptible adversarial attacks that generate misleading outputs and endanger production safety.Existing adversarial detectors poorly identify small perturbation adversari-al samples(SPAS),and adversarial training defense methods still widely suffer from adversarial robustness overfit-ting.To tackle these issues,an adversarial training and attack detection integration(ATADI)defense method is proposed.First,it theoretically proves that SPAS-only adversarial training better mitigates such overfitting.Second,on this basis,a SPAS-based feature-anchored adversarial training(FAAT)scheme,namely SPAS-FAAT,is proposed.In ATADI,the detector only targets large perturbation adversarial samples(LPAS)to avoid SPAS detec-tion difficulties during defense,while SPAS are directly fed into SPAS-FAAT-enhanced DLSS to obtain the final soft sensing results.Finally,experiments are conducted on an adversarial attack-defense case of rotor thermal de-formation soft sensing:Structural ablation results verify that the integration of adversarial training and attack de-tection serves as an effective defense means for DLSS.LPAS from various adversarial attacks are detected with over 98.42%accuracy,and SPAS-FAAT-enhanced DLSS achieves satisfactory accuracy in the soft sensing results for both SPAS and normal samples.The ATADI method significantly improves DLSS adversarial robustness,provid-ing a new avenue for adversarial defense method research.