In this work, we model with artificial intelligence techniques one of the most used solutions for energy saving in the field of refrigeration. This solution, called floating head pressure control, allows for the optimum pressure condensation depending on the working environment, thus increasing the overall efficiency of the plant. Usually, these mechanisms are controlled by algorithms that are stored in the control memory of the chillers. Under certain environmental conditions, the optimal temperature and condensation pressure fluctuate with the environmental temperature. Thus, the absorbed electricity depends on the working parameters fixed by the control system of the chiller. Results in energy savings in these terms could grow as the external temperature decreases. The problem addressed in this work regards standard floating head pressure control (FHPC) systems that do not combine working environmental factors with the conditions of the machine and historical data because they are implemented adopting static models. We provide evidence about the potential of machine learning models in predicting the velocity of fans in order to adjust their load based on environmental factors and the conditions of the machine. A good setting of fan velocity will result in lower energy consumption. To do that we analyzed and implemented machine learning algorithms to provide instruments that support the operation of the chiller, enabling the floating head pressure control mode in the control system of the machine, during its process. We performed an empirical evaluation on both synthetic and real data to assess the quality of our proposal. Synthetic data are produced by an industrial software that simulates the behavior of chillers, while real-world data is collected from a commercial chiller. The results show that machine learning approaches are able to approximate real data getting errors that are 5 times smaller than the errors committed by the system that is now adopted. All the source code as well as the datasets will be made available online after the review process.
Simulating Floating Head Pressure Control with Artificial Intelligence
Loreggia A.
;
2022-01-01
Abstract
In this work, we model with artificial intelligence techniques one of the most used solutions for energy saving in the field of refrigeration. This solution, called floating head pressure control, allows for the optimum pressure condensation depending on the working environment, thus increasing the overall efficiency of the plant. Usually, these mechanisms are controlled by algorithms that are stored in the control memory of the chillers. Under certain environmental conditions, the optimal temperature and condensation pressure fluctuate with the environmental temperature. Thus, the absorbed electricity depends on the working parameters fixed by the control system of the chiller. Results in energy savings in these terms could grow as the external temperature decreases. The problem addressed in this work regards standard floating head pressure control (FHPC) systems that do not combine working environmental factors with the conditions of the machine and historical data because they are implemented adopting static models. We provide evidence about the potential of machine learning models in predicting the velocity of fans in order to adjust their load based on environmental factors and the conditions of the machine. A good setting of fan velocity will result in lower energy consumption. To do that we analyzed and implemented machine learning algorithms to provide instruments that support the operation of the chiller, enabling the floating head pressure control mode in the control system of the machine, during its process. We performed an empirical evaluation on both synthetic and real data to assess the quality of our proposal. Synthetic data are produced by an industrial software that simulates the behavior of chillers, while real-world data is collected from a commercial chiller. The results show that machine learning approaches are able to approximate real data getting errors that are 5 times smaller than the errors committed by the system that is now adopted. All the source code as well as the datasets will be made available online after the review process.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
