Gray box dynamic modeling of vapor compression systems for control optimization

Abstract

Buildings account for 75% of electricity use in the U.S. and more than 24% of building electrical energy is consumed by vapor compression equipment, including air-conditioners, refrigerators/freezers and heat pumps. Dynamic modeling of vapor compression systems (VCS) is particularly important for developing and validating optimal control strategies to maximize the system efficiency and reliability. However, existing modeling techniques are rarely used in control practices because of the significant model development effort and requirement of high computational resources. This dissertation presents an efficient and robust gray-box dynamic modeling approach for VCS to support control optimization. The presented methodology allows automated construction of data-driven VCS models with minimum training data and human inputs. The overall approach incorporates a multi-stage training procedure with separate estimation of the steady-state and dynamic model parameters along with a finite control volume scheme to achieve good model identifiability while ensuring adequate prediction accuracy. To improve model reliability, the modeling approach incorporates sensitivity analysis and de-correlating steps in a pre-conditioning procedure to avoid over-parameterization. The system-level training identifies the refrigerant charge that minimizes the steady-state simulation errors while the dynamic modeling stage transforms the established steady-state system model into a dynamic counterpart, in which the optimal thermal capacitances of the heat exchanger walls are identified to best reproduce system transient responses.