This thesis aims at developing a method that makes use of advanced statistical models to analyze building consumption data and assess energy retrofit impact. The research is focused on tertiary buildings and the models are based on hourly and sub-hourly smart meters data

It is estimated that about 40% of worldwide energy use occurs in buildings [ 1 ]. Increasing energy efficiency in the building sector has become a priority worldwide and especially in the European Union. It is clear that an immense energy efficien cy potential lies in buildings and it is not properly harnessed. The energy efficiency increa se can be realized through energy retrofitting actions, optimization of the building c ontrol strategy, or through the timely reporting of abnormal energy performance. In this thesis, a framework for the evaluation of the impact of energy retrofitting measures, with a statistic al learning approach, is proposed. The model was developed as part of EDI-Net, a Horizon 2020 pro ject, with the main goal of facilitating energy consumption monitoring in buildings a nd allowing analysis and evaluation of applied energy efficiency measures (EEM). The baseline mod els for the impact evaluation are generated using Generalized Additive Models (GAM), enh anced with auto regressive terms. Three different pilot buildings (one in Spain and two i n the UK) are examined and their savings evaluated through the analysis of hourly smar t meter consumption data and weather data. The results show that it’s possible to evaluat e energy savings in tertiary buildings using a data-driven approach, although further w ork is needed, in order to validate and automatize the model.