This paper investigates the optimal formation control problem for multi-agent systems under actuator faults and proposes a predefined-time optimized fault-tolerant formation control method based on reinforcement learning. The method employs a predefined-time tuning function and combines with an actor-critic algorithm and significantly reduces algorithmic complexity. Based on this, an efficient fault-tolerant control mechanism is designed to ensure the system achieve the desired formation control objectives, even in the presence of actuator faults. Furthermore, the reinforcement learning-based adaptability further enhances the algorithm's robustness and adaptability in complex dynamic environments. Finally, simulation results validate the effectiveness and superiority of the proposed method across fault scenarios.