This article proposes an adaptive input reconstruction based resilient model predictive control (MPC) strategy for continuous-time nonlinear cyber-physical systems (CPS) against deception attacks. The input reconstruction mechanism is first developed based on the self-triggered sampling mechanism which could not only relax the assumptions on the attack energy limitation utilized by the existing resilient MPC methods, but also significantly reduce the cyber resource consumption of the resultant CPS. Besides, the adaptive prediction horizon mechanism is incorporated into the proposed MPC method to reduce its computational complexity. Furthermore, the feasibility and closed-loop stability of the developed MPC algorithm under deception attacks are strictly proven. Finally, the effectiveness of the designed algorithm in defending against deception attacks and reducing resource consumption is tested through both simulation and robot experiments.
Self‐triggered control strategy has been widely applied in cyber–physical systems, which are unavoidable to suffer various malicious cyber‐attacks. This paper is aimed at solving the cyber security problem of self‐triggered model predictive control in cyber–physical systems under false data injection attack, and a resilient control strategy is proposed based on control signal reconstruction. Firstly, the continuous control signal is discretely sampled at the controller side, and the key data samples are determined and protected. Then, the discretised control samples are transmitted through the network channel, based on which the control signal is reconstructed at the controlled system side according to a preset mode. It is further theoretically proved that the control signals reconstructed from the critical control samples can guarantee the feasibility and stability of the control system against the FDI attacks. Finally, a simulation verification on a robot system is conducted to verify the effectiveness of the proposed algorithm.
31 August 2023 in IEEE Intelligent Transportation Systems Magazine
False data injection (FDI) attacks have significantly threatened the security of the cyberphysical system (CPS). To ensure the stability of the CPS under FDI attacks, resilient model predictive control (MPC) has been extensively researched. However, most of the existing resilient MPC methods fail to consider the network resource limitation of the CPS, which may not be applicable in some real systems. Therefore, in this article, a resilient self-triggered (ST) MPC strategy is developed for a discrete-time nonlinear CPS under FDI attacks, which can not only ensure the system’s stability but also reduce resource consumption via decreasing the update frequency of MPC and economizing the utilization of the protection resource. First, an input signal reconstruction mechanism is designed based on key control data selection, which could reconstruct a feasible control sequence for the CPS even if the original ST control sequence is tampered with by FDI attacks. Then, a resilient ST-MPC algorithm based on the input signal reconstruction mechanism is proposed to weaken the adverse effects of FDI attacks and reduce resource consumption simultaneously. Moreover, the recursive feasibility of the resilient ST-MPC mechanism and input-to-state stability of the controlled system are respectively proved. Finally, the performance of the resilient ST-MPC mechanism is shown through a cart–damper–string system and an intelligent vehicle system.
For solving the problem of building climate system uncertainty affected by spatio-temporal variables, an event-triggered multi-kernel learning-based stochastic model predictive control (EMSMPC) method is developed. Compared to the existing stochastic model predictive control (SMPC) methods, the developed method does not require the uncertainty to satisfy strict distributional conditions and can effectively handle the spatio-temporal coupling effects within the uncertainty. Firstly, the spatio-temporal uncertainty is learned via multi-kernel Gaussian process regression. The learning results are employed for constructing the cost function and designing the chance constraint tightening set, thereby ensuring that the chance constraints are satisfied while maintaining the robustness of the controlled system. Then, an event-triggering mechanism is introduced to reduce the frequency of solving optimal control problem (OCP) and online learning, further reducing the energy consumption of the controlled system. Moreover, the feasibility and closed-loop stability of stochastic predictive control method based on multi-kernel learning are critically analyzed. Finally, the effectiveness of the developed method is verified through simulation and experimentation.
10 July 2025 in IEEE Transactions on Industrial Electronics
This article develops a predictive-triggered model predictive control (MPC) approach based on parallel computing for continuous-time nonlinear system requiring fast response. Unlike the existing MPC approach, the proposed method advances the moment of solving MPC optimal control problem (OCP) by predictive-triggered mechanism and parallel computation method. Such a scheme ensures that optimal control input can be obtained timely by the controlled system, thereby removing the implicit assumption of existing MPC approach that OCP can be solved instantaneously. Besides, the input reconstruction mechanism is implemented to improve resource utilization of predictive-triggered MPC. Furthermore, the feasibility and closed-loop stability of the proposed predictive-triggered MPC are demonstrated, respectively. Finally, the proposed algorithm is tested to verify its resource utilization and real-time performance through both simulation and robot platform experiment.