A machine learning approach to develop building energy retrofit strategies

edició 2019

Benedetto Grillone

It is estimated that about 40 % of worldwide energy use occurs in buildings. Increasing energy efficiency in the building sector has become a priority worldwide and especially in the European Union, although it is clear that energy efficiency potential that lies in buildings is not being properly harnessed. Given the relatively low turn-over rate of the building stock, energy efficiency retrofit appears to be a fundamental step in reducing the energy consumption and CO2 emissions in existent buildings.

Identifying the combinations of retrofit measures that provide the highest energy savings, for a particular building, is a major technical challenge that can require a long and expensive process of energy auditing. The scope of this PhD research is to find innovative solutions to this problem through IoT and big data analysis, two hot topics of industry 4.0. Our group is working on the development of a set of machine learning algorithms that, through the sole analysis of smart meter consumption data, weather data and building characteristics, can be able to develop efficient energy retrofit strategies according to different criteria (highest savings, lowest investment costs, highest cost/benefit ratio etc.).

A preliminary study on methods to calculate the impact of energy conservation measures (ECM) on real buildings using data-driven algorithms, such as Generalized Additive Models (GAM), was realized and gave encouraging results. The method was tested in several existing buildings in the framework of European projects SHERPA and EDI-Net, through the analysis of hourly smart meter consumption data and weather data. The results showed the viability of this quick and cost-effective approach to evaluate the impact of applied EEM and opened to further research to verify the method’s scalability to a district, city or national level when applied in a big data environment. The PhD research is now oriented towards the improvement of the introduced data-driven models and the development of new decision-making mechanisms, based on artificial intelligence, to develop energy retrofit strategies.