Predictive Functional Control of Superheat in a Refrigeration System using a Neural Network Model

This paper compares three methods for control of the superheat in a refrigeration system. A traditional gain scheduled PI-based controller, a predictive functional controller (PFC) and a predictive functional controller with a neural network model (PFCNN). The aim is to investigate the performance of the three controllers with respect to disturbance rejection measured both at the superheat deviation from the reference and the actuation of the expansion valve. The controllers are designed and tested on a laboratory set-up. The performance of the controllers turns out to be similar and distinguish between the concepts must be based on other parameters like tuning and demands for computational power.