Deep reinforcement learning approaches for process control

In this work, we have extended the current success of deep learning and reinforcement learning to process control problems. We have shown that if reward hypothesis functions are formulated properly, they can be used for industrial process control. The controller setup follows the typical reinforcement learning setup, whereby an agent (controller) interacts with an environment (process) through control actions and receives a reward in discrete time steps. Deep neural networks serve as function approximators and are used to learn the control policies. Once trained, the learned network acquires a policy that maps system output to control actions. Though the policies are not explicitly specified, the deep neural networks were able to learn policies that are different from the traditional controllers. We evaluated our approach on Single Input Single Output Systems (SISO), Multi-Input Multi-Output Systems (MIMO) and tested it under various scenarios.