IEEE Access (Jan 2023)

Decentralized H<sub>&#x221E;</sub> Observer-Based Attack-Tolerant Formation Tracking Network Control of Large-Scale LEO Satellites via HJIE-Reinforced Deep Learning Approach

  • Bor-Sen Chen,
  • Hung-Yu Lin

DOI
https://doi.org/10.1109/ACCESS.2023.3246182
Journal volume & issue
Vol. 11
pp. 17165 – 17196

Abstract

Read online

In this study, an H∞ attack-tolerant decentralized observer-based formation tracking control strategy is designed for the network control system (NCS) of large-scale LEO satellite team under external disturbance and attack signal. First, smoothing signal models of attack signals are embedded in each satellite to avoid their corruption on state estimation of Luenberger observer and to compensate their effect on the formation tracking of large-scale satellites. In addition, the observer-based formation tracking NCS of each satellite must efficiently attenuate the external disturbance, measurement noise and coupling effect from adjacent satellites. For the proposed decentralized H∞ attack-tolerant observer-based team formation NCS of large-scale satellites, each satellite needs to solve a very complicated but decoupled Hamilton Jacobi Isaacs equation (HJIE). Therefore, a proposed HJIE-reinforcement learning-based deep neural network (DNN) is employed for each satellite to directly solve a corresponding nonlinear partial differential control-observer-coupled $HJIE_{i}$ of decentralized H∞ attack-tolerant control problem. When trained by the proposed HJIE-reinforcement Adam deep learning algorithm, DNN can be reinforced to solve $HJIE_{i}$ for H∞ control gain and observer gain as well as the worst-case external disturbance, measurement noise, and coupling of each satellite of the team formation in the off-line training phase. That is, the proposed HJIE-reinforcement learning algorithm-based DNN scheme in each satellite NCS can achieve robust decentralized H∞ attack-tolerant observer-based team formation control strategy. When the HJIE-reinforcement-based Adam learning algorithm converges, we can show that the proposed reinforcement learning-based DNN formation tracking control scheme of each satellite can approach the theoretical robust decentralized H∞ attack-tolerant observer-based formation tracking strategy of large-scale satellites NCS. In the simulation example, a satellite team with external disturbance, measurement noise and wireless communication malicious attack are given to validate the effectiveness of proposed method separately.

Keywords