A Big Data toolkit to support systematic health assessment of energy systems in buildings utilising AI techniques

Buildings account for 40% of worldwide primary energy and 27% of the total CO2 emissions. It is essential to make energy consumption in smart buildings more effcient by trying to keep the building operating in optimal conditions for as long as possible. New health assessment technologies, such as Prognostics and Health Management (PHM) can help to meet this goal. With the increasing availability of sensor and meter data in smart buildings, as well as advances in artificial intelligence (AI) techniques, the search for a PHM-based system that integrates these techniques may be a solution to the problem of energy effciency in buildings. This thesis has achieved two objectives: to design a Health Management methodology for potential energy use optimisation in smart buildings, and to validate the methodology in a real case study. To fulfil the objectives, a number of tasks have been carried out which have led to several contributions. A new methodology has been proposed to monitor the health of the system using mainly data-driven AI techniques. The proposal defines a set of high-level tasks (such as consistency checking, state assessment, KPI monitoring, FDD, prognosis and predictive maintenance) and the concept of health state vector. The solution can be adapted depending on the KPIs and tasks needed for each smart building. It has been determined that one of the most relevant KPIs is the energy consumption of heating, ventilation, and air conditioning (HVAC) systems and that it would be important to be able to predict it in order to make decisions that lead to a reduction in consumption. For this purpose, an approach based on machine learning has been proposed: on the one hand, unsupervised techniques that allow estimating the operation mode of the system based on the meteorological conditions and, on the other hand, two prediction models have been developed using deep learning techniques that allow estimating the evolution of consumption in the short and medium term. The usual machine learning (ML) methodology has been applied to build two possible solutions for the fault detection and diagnosis (FDD) task of an HVAC system, which would allow early detection of errors in the system and thus correct them, avoiding consumption associated with failure situations.The approach has been tested in the Alice Perry Engineering building, where the KPIs have been defined, weather-dependent operating modes have been detected and consumption prediction models have been successfully built (according to ASHRAE requirements); finally the FDD module has been successfully tested with promising results.

Publisher

University of Galway

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International

cbn

Collections

University of Galway Theses (PhD Theses)