Using Machine learning to reduce energy usage in buildings

Buildings represent an important opportunity to reduce energy consumption, and help mitigate global warming. Most of the energy in buildings is used to maintain air quality and the right temperature inside the buildings. This is controlled by cooling, heating and ventilation systems, known as the technical building system. When this system is not efficiently operated, more energy is required, which incur higher energy and service costs and CO2 emissions. At the same time IoT and the development of new sensor technology has contributed to more buildings having sensors that measure energy consumption, temperature, pressure and humidity. In this ph.d machine learning and reactive programming is used to optimize operation of the technical building system (HVAC-system) and to develop a web-application that can deliver real-time predictive analytics. The research use predictive techniques to early detect errors in the HVAC-system. Furthermore, these methods are used to give precise measurements of energy savings, reduce maintenance cost and increase the lifetime of the technical building system.

Metodene som er utviklet er tatt i bruk i alle prosjekter hvor vi beregner energibesparelse. Videre er forskningen operasjonalisert inn i en webapplikasjon som er i bruk av våre energirådgivere til kontinuerlig overvåkning av prosjekter, til både energibesparelser, optimalisering og for å avdekke mulige tekniske feil i anlegg. Siden prosjektet har benyttet open-source teknologi til å operasjonalisere forskningen og knyttet en publikasjon opp mot dette vil dette være et bidrag som det er mulig for bransjen å anvende/replisere for å øke presisjonen i måling og verifisering (M&V) av energispareprosjekter.

Somewhere between 30 and 40% of the global energy consumption occurs in buildings (United Nations Environment Programme, 2007). Thus, buildings represent an important opportunity to reduce energy consumption, and further help mitigate global warming, one of the world's most important problems. Most of the energy in buildings is used to maintain air quality and the right temperature inside the buildings. This is controlled by cooling, heating and ventilation systems, known as the technical building system. When this system is not efficiently operated, more energy is required, which will incur higher energy cost, service costs and CO2 emissions. Additionally, a non-efficiently operated building system will have a lower expected lifetime. Using unique smart-meter energy data from more than 50 million hourly observations; the proposed project will demonstrate how automated statistical methods and big-data techniques can help building owners lower their energy consumption and reduce maintenance costs. To accomplish this the project will conduct research within three different areas. First, maintenance costs constitute a significant percentage of expenses in most buildings. A typical (reactive maintenance) strategy is "wait until something breaks". Using machine learning (predictive maintenance) we will demonstrate how to reduce maintenance cost, and increase lifetime of the technical building system. Second, peak load tariffs are steadily increasing. Using predictive analytics, we will show how forecasting energy usage can be used to shift energy consumption from the main power line to locally produced solar and battery storage during peaks hours. These methods can reduce total energy costs, and peak shaving / phase shift can increase the lifetime of the power grid. Third, and last, the project will demonstrate how to benchmark energy consumption using machine learning and mathematical programming.