Survey of machine learning-based modeling for process control systems

Abstract

This paper presents a survey on machine learning modeling for process control systems. Model-based control methods such as model predictive control are useful for automation of industrial plants because they can optimize their control policies efficiently with mathematical models of plants. In such methods, it is important to obtain models that can predict behaviors of plants accurately. Recently, machine learning-based modeling methods have attracted much attention to identify nonlinear plants. In this paper, we summarize standard machine learning-based approaches and introduce usage of prior knowledge such as scientific or common knowledge. Additionally, as an example in the real world, we introduce model predictive control with neural network modeling to deal with disturbance for a distillation plant. Finally, future work of machine learning-based modeling for industrial plants are analyzed and discussed.