An Advanced Methodology to Enhance Energy Efficiency and Performance in a Hospital Cooling-Water System

Abstract

Hospitals are a type of building with especially high energy demands; and this is owing to the fact that they run life-saving services 24 hours a day, 365 days a year. Moreover, the healthcare services offered by hospitals are growing in number and complexity, which means that their energy demands increase every year. In order to cover the energy needs of all this activity, a vast amount of technical installations are required. In addition, supplying energy and liquids increasingly necessitates greater control, precision, and quality. Due to the critical role of cooling-water systems, this thesis focuses on these installations that are vital for both the comfort they provide through airconditioning and for healthcare activities. The objective of this research is to improve the performance of hospital refrigeration plants to increase energy efficiency, while also reducing inefficiencies in generator start-ups and maintenance, which are commonplace problems in this type of facility. By applying Machine Learning (ML) models to predict cooling demand, it has been possible to anticipate, adapt, and plan for actual thermal generation to meet, but not exceed, expected demand. To obtain said models, an already existing methodology based on genetic algorithms called GAparsimony was utilized. This methodology allows parsimonious models to be obtained in an automated fashion. The algorithms used include artificial neural networks (ANN), support vector machines for regression (SVR), and extreme gradient boosting machines (XGBoost). Prior to the modeling phase, an extensive general optimization of the cooling-water facilities was carried out; and during this process a methodology was developed to be applied in the following areas: the control system, the data acquisition system, and the physical systems. The optimization culminated with a demand prediction model being implemented in the BMS (Building Management Systems). This feature enabled the BMS to anticipate generator programming a day in advance, thus exercising predictive management. The research presented herein has been corroborated by the results obtained when the optimization methodology was applied, and by implementing the demand prediction model in the BMS as well.