In order to solve the problem of swing angle suppression of overhead cranes, an intelligent anti-swing control method based on soft actor-critic（SAC）algorithm is proposed. The reward function is designed to reduce the swing angle and reach the desired position of the trolley as soon as possible, an action network is adopted, with the crane trolley displacement, trolley speed, payload angle and its angular acceleration as the input of the action network, and the driving force as the output of the action network. In order to improve the stability in the training process, four value networks are adopted, and entropy regularization term and entropy weighting coefficient are introduced into the networks to balance the relationship between exploration and utilization in the training process. The value networks obtain the target networks through soft update, which reduces the local optimization and divergence in the training process. Finally, the action network for control is obtained by using model training. The results show that the proposed intelligent anti-swing control system has a good suppression effect on the payload swing angle, and has good robustness to payload mass changes, rope length parameter variations and external disturbances. Compared with the generalized motion method based on the end effector, the proposed method shows certain advantages in the dynamic performance and anti-disturbance of the crane system.