Leveling control of multi-cylinder hydraulic press based on multi-agent reinforcement learning

In this paper, a multi-agent–based reinforcement learning (RL) algorithm is proposed to solve the leveling control problem of a multi-cylinder hydraulic press with coupling phenomena. This algorithm is a model-free control algorithm, which can avoid the modeling difficulties and low efficiency caused by the complexity of the model. The control algorithm of the hydraulic press adopts Multi-Agent Soft Actor–Critic (MASAC). The concept of multi-agent is introduced to control each coupling input separately. The distributed updating method is used to realize accurate and stable control of the hydraulic press. At the same time, a reward function of the piecewise function type is proposed in this paper. Compared with common algorithms such as the quadratic reward function, this algorithm has a faster and more stable convergence effect in the whole process. Experiments show that the proposed algorithm has better convergence speed and leveling accuracy than the traditional single-agent algorithm.