An Incursion to Deep-Learning for Process Regulation

Abstract

A model which represents a physical process is usually composed by conservation equations, transfer mechanisms, and closure equations. These equations vary in the degree of certainty. This paper describes the incorporation of physical and empirical models. The empirical part is constructed by Deep Learning. This work describes the principles which have promoted Deep Learning as a complementary tool for the approximation of process engineering when is used for model-based control. In addition of the stability and accuracy to deal with unmeasured disturbances, a robust strategy is to use Reinforcement Learning thus the principles of this strategy are also described.

Keywords

• Model construction
• Model Validation
• Predictive Model
• Deep Learning
• Reinforcement Learning