Design of “Deep Learning Controller”

Koksal Erenturk

Deep learning allows computational models of multiple processing layers to learn and represent data with multiple levels of abstraction mimicking how the brain perceives and understands multimodal information, thus implicitly capturing intricate structures of large‐scale data. In the meantime, recent advances in deep learning, encompassing neural networks, hierarchical probabilistic models, and a variety of unsupervised and supervised feature learning algorithms, have brought about tremendous development to many areas of interest to the engineering community. In this work, an extended type of the current accomplishment of deep learning to chemical process control problem has been presented. As well-known, if one formulated the reward function properly, “deep learning” can be used for industrial process control purpose. 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 (DNN) serve as function approximators and are used to learn the control policies. Once the DNN trained, control actions can be achieved at the output of the learned network. Even though the policies are not explicitly specified for the DNN, the DNN has an ability to learn policies that are different from the traditional controllers. The designed “Deep Learning Controller” (DLC) for Single Input Single Output Systems (SISO) has been tested under various scenarios. Obtained results have been given in graphical illustrations for details and these results showed that DLC can be easily used for instead of any type of controller. Additionally, it can be concluded that DLC are very robust when compared with the other type of controllers in terms of noise and unknown disturbances.

https://dx.doi.org/10.31873/IJEAS.5.10.31