All Agents Connectivity-Preserving and Error-Based Cooperative Learning Control With Data-Filter Memory-Based Event-Triggered Strategy
Zeyi Liu, Huaguang Zhang, Jiayue Sun, Lei Wan · IEEE Transactions on Automation Science and Engineering · 2024
This paper proposes an all agents connectivity-preserving method and a data-filter memory-based event-triggered (ET) strategy to design cooperative learning control algorithm. Firstly, a type of error functions are proposed to achieve that all agents are within the communication boundary, which do not limit the initial values of agents. The designed method can dynamically adjust the boundary function based on the initial position of agents and gradually converge to the preset communication boundary, without abandoning any agents. Secondly, a data-filter memory-based ET strategy is proposed, which includes the designed error-based data filtering rules. The filtering rules avoid the problem that the ET mechanism stores abnormal historical data when the system has faults. Moreover, the presented error-based cooperative learning adaptive protocol does not need to presuppose that neighbor weights are bounded, reducing the conservatism. Finally, based on the above works, the constructed ET control method can achieve control objectives, and the effectiveness is demonstrated through theoretical analysis and simulation results. Note to Practitioners—In practice, multiagent systems (MASs) communicate through wireless communication mostly, which inevitably leads to an upper limit on the communication distance between agents. Exceeding the limitation of communication module will cause the problem that MASs cannot achieve signal transmission. Therefore, it is crucial to design appropriate constraint methods for different initial positions of agents to ensure that all agents can enter the predetermined communication boundary. In addition, the memory ET strategy can calculate the threshold of conditions based on the historical data. But in practice, it cannot guarantee the continuous normal operation of the system. If there are abnormal values in the stored signal data, this will lead to unreasonable calculation of the threshold. In response to this issue, this paper considers additional data filtering rules to avoid storing data when systems exist faults. Meanwhile, the cooperative learning algorithm proposed in this paper can adjust the learning information weights based on the control performance of neighbor agents, and remove the assumption of bounded neighbor learning laws.