Prescribed Performance Fault-Tolerant Control for Distributed Multi-agent Systems with Improved LSTM Fault Compensation

Addressing the consensus control problem of distributed multi-agent systems under actuator faults, this paper proposes a prescribed performance fault-tolerant control method that integrates improved Long Short-Term Memory (LSTM) fault compensators. Traditional control can guarantee transient and steady-state performance of tracking errors, but performance severely deteriorates under fault conditions. This paper designs an improved prescribed performance function to ensure that the system can still maintain predetermined performance boundaries in preset time. Then, an improved LSTM fault compensator is proposed. The innovation features are as follows: (1) Bayesian uncertainty quantification mechanism for quantitative assessment of prediction credibility; (2) Multi-scale temporal feature extraction algorithm integrating short-term convolutional features, medium-term LSTM features, and long-term Gated Recurrent Unit (GRU) features; (3) Collaborative fault diagnosis mechanism based on information entropy confidence-weighted fusion; (4) Personalized compensation strategy based on the fusion of multiple types of compensators. Finally, the effectiveness of the proposed method under partial failure faults is verified through simulation. Results show that compared to uncompensated systems, the proposed method achieves significant improvement in overall performance and the performance during fault periods.