Observer-Based Resilient Adaptive Neural Sliding Mode Control for DC-MGs Under Blended Attacks With Multidomain Attack-Aware Scheduling Protocol

This paper is concerned with the co-design problem of the adaptive neural sliding mode compensation controller and multidomain DoS-attack-aware scheduling protocol (MDSP) for the direct current microgrids (DC-MGs). It is considered that the process of transmitting the system data over wireless networks is vulnerable to intrusion by cyber-attacks, in which the aggressors might send denial-of-service (DoS) attacks to the output data and inject fake messages into the control signal. Due to the fact that not all data in DC-MGs can be acquired from direct measurements, an attack-parameter-dependent observer is introduced to estimate the unknown states under DoS attacks. Then, with the objective of enhancing the system responsiveness and bandwidth utilization, a novel MDSP is presented by introducing a multidomain structure, which enables to maintain the communication function of DC-MGs when some channels fail or are blocked. In the meantime, the neural network technique is employed to approximate unknown attack information based on the observer states, thus recovering normal data attacked by potential false data injection (FDI) attacks. Additionally, by introducing the partitioning strategy, the corresponding resilient finite-time boundedness for both the reaching phase and sliding motion phase is acquired, respectively. An adaptive sliding mode compensation controller is devised by introducing the additional compensation terms, which can improve the control accuracy and the adaptability to cyber-attacks, thus ensuring that the states of DC-MGs are rapidly entered and maintained on the sliding surface during a finite-time interval. Finally, the availability and validity of the presented approach is verified through a DC-MG with three constant power loads.