基于机器学习的信息物理系统安全控制  

Secure Control for Cyber-physical Systems Based on Machine Learning

作　　者：刘坤 马书鹤 马奥运 张淇瑞 夏元清[1] LIU Kun;MA Shu-He;MA Ao-Yun;ZHANG Qi-Rui;XIA Yuan-Qing(Key Laboratory of Intelligent Control and Decision of Com-plex Systems,School of Automation,Beijing Institute of Techno-logy,Beijing 100081)

机构地区：[1]北京理工大学自动化学院复杂系统智能控制与决策国家重点实验室,北京100081

出　　处：《自动化学报》2021年第6期1273-1283,共11页Acta Automatica Sinica

基　　金：国家自然科学基金(61873034,61503026,61836001);北京自然科学基金(4182057);国家自然科学基金重大国际(地区)合作项目(61720106010);北京市智能物流系统协同创新中心开放课题(BILSCIC-2019KF-13);北京理工大学研究生创新项目(2019C-X20031)资助。

摘　　要：研究了控制信号被恶意篡改的信息物理系统的安全控制问题.首先,提出一种改进果蝇优化核极限学习机算法(Kernel extreme learning machine with improved fruit fly optimization algorithm,IFOA-KELM)对攻击信号进行重构.然后,将所得重构信号作为系统扰动加以补偿,进而设计模型预测控制策略,并给出了使被控系统是输入到状态稳定的条件.另外,本文从攻击者角度建立优化模型得到最优攻击策略用以生成足够的受攻击数据,基于此数据,来训练改进果蝇优化核极限学习机算法.最后,使用弹簧−质量−阻尼系统进行仿真,验证了改进果蝇优化极限学习机算法和所提安全控制策略的有效性.This paper investigates the security control problem of cyber-physical systems whose control signals are maliciously tampered.Firstly,a kernel extreme learning machine with improved fruit fly optimization(IFOAKELM)algorithm is proposed to reconstruct the attack signal.Secondly,with the reconstructed signal treated as disturbance,a model predictive control strategy is designed to secure the system,and a condition that guarantees the input-to-state stability of the attacked system is given.In addition,to train the proposed algorithm,enough data of the system attacked with an optimal strategy is generated.This strategy is obtained by solving an optimization problem from the attacker's perspective.Finally,a numerical example of the spring-mass-damping system is illustrated to verify the effectiveness of the IFOA-KELM algorithm and the proposed control strategy.

关 键 词：信息物理系统 攻击信号重构 核极限学习机 果蝇优化算法 模型预测控制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP393.08[自动化与计算机技术—控制科学与工程]