Typical reinforcement learning (RL) methods show limited applicability for real-world industrial control problems because industrial systems involve various constraints and simultaneously require continuous and discrete control. To overcome these challenges, we devise a novel RL algorithm that enables an agent to handle a highly constrained action...

Reinforcement learning with distance-based incentive/penalty (DIP) updates for highly constrained industrial control systems