Model Based Control of Refrigeration Systems

Today refrigeration control systems consist of a number of self-contained distributed control loops that during the past years has been optimized obtaining a high performance of the individual subsystems, thus disregarding cross-couplings as well dynamically as statically. The supervisory control of the supermarket refrigeration systems therefore greatly relies on a human operator to detect and accommodate failures, and to optimize system performance under varying operational condition. Today these functions are maintained by monitoring centres located all over the world. Initiated by the growing need for automation of these procedures, that is to incorporate some "intelligence" in the control system, this project was started up. The main emphasis of this work has been on model based methods for system optimizing control in supermarket refrigeration systems. The idea of implementing a system optimizing control is to let an optimization procedure take over the task of operating the refrigeration system and thereby replace the role of the operator in the traditional control structure. In the context of refrigeration systems, the idea is to divide the optimizing control structure into two parts: A part optimizing the steady state operation "set-point optimizing control" and a part optimizing dynamic behaviour of the system "dynamical optimizing control". A novel approach for set-point optimization will be presented. The general idea is to use a prediction of the steady state, for computation of the cost-function gradient and use it in an outer control loop for adjusting the setpoints. By introducing barrier functions the operational limitations of the system can be taken into account and furthermore by relaxing the overall control objective operational feasibility can be maintained under extreme conditions, i.e. by degrading the performance. The method has been successfully applied on a test frigeration system for minimization of the power consumption; the hereby gained experimental results will be presented. The present control structure in a supermarket refrigeration system is distributed, which means that the crosscouplings are disregarded. Though these cross couplings are relatively weak, it turns out that they pose a major problem when hysteresis controllers are used in the local loops. It can namely cause synchronization of the distributed hysteresis controllers, thus accumulating the effects of the cross-couplings resulting in large disturbances. In supermarkets refrigeration systems the temperature control in the refrigerated display cases are maintained by hysteresis controllers. Based on a model predictive hybrid framework a novel approach for desynchronization is presented. The approach is applied for desynchronization in a supermarket refrigeration system.