<p>This paper focuses on dynamic marshalling cooperative optimization control of permanent magnetic maglev trains. The traditional distributed controller is difficult to adjust the control parameters adaptively for complex and changeable tracking scenarios, resulting in an unsafe, inefficient and uncomfortable marshalling process. This paper proposes a distributed active disturbance rejection resilient controller based on optimized twin delayed deterministic policy gradient (TD3) algorithm. First, an improved distributed active disturbance rejection controller (DADRC) based on the bidirectional leader communication topology is designed to realize the multi-train dynamic marshalling cooperative control and its stability is proved theoretically. Second, different dynamic marshalling processes are adaptively optimized by using the TD3 algorithm to train the DADRC. Third, an adaptive Mayfly (AMA) algorithm with the Steffensen mutation mechanism is proposed to optimize some sensitive hyperparameters of the TD3 algorithm. The simulation results show that compared with the traditional distributed controller, the proposed resilient controller can adaptively adjust its own control parameters and flexibly optimize the dynamic marshalling process according to different tracking scenarios. Compared with the Mayfly-TD3 (MA-TD3) and AMA-deep deterministic policy gradient algorithms, the proposed AMA-TD3 algorithm shows more stable and fast convergence, and can achieve a successful control rate as high as 99.8% and a generalization rate of up to 93.2%.</p>