Definition of a data analytics-based energy benchmarking process for prioritizing energy management strategies for stocks of buildings

The building sector is one of the world's top users of primary energy. According to the International Energy Agency (IEA), buildings account for around 21\% of total final energy consumption in EU member nations. As a result, the building sector is currently one of the most strategic targets for reducing total energy consumption and improving energy efficiency in order to meet stringent decarbonization objectives. In recent years, the development of Internet of Things (IoT) and Information and Communication Technologies (ICT) has made it possible to collect huge amount of building related-data de facto increasing valuable knowledge resources to guide experts and decision-makers in a multitude of energy management tasks. Among these, the analysis of energy consumption time series in large buildings is one of the application areas that is experiencing continuous innovation and refinements, especially for what concerns the birth of automatic tools for the development of energy benchmarking systems. While collecting and storing data is over time requiring less effort, the analysis of energy-related data in buildings still requires a high degree of expert knowledge, since their analysis poses several challenges related to the influence that factors such as building features, climatic conditions, occupant behaviour, comfort requirements, operating schedules and management have on the whole building performance. This thesis proposes a methodology for benchmarking energy consumption in stocks of non-residential buildings to allow a prompt and automatic recognition of abnormal or non-optimal performances of buildings and providing information for the identification of energy waste and for the prioritization of corrective interventions. The proposed approach makes use of different Energy Performance Indicators (EPIs), which have the purpose of identifying inefficiencies in buildings relatively to their load profile shapes, magnitude of energy consumption, operational schedules, energy use intensity and energy usage patterns. The main novelty introduced in this thesis is related to the concept of 'Regularity' of energy consumption, defining a new EPI that is based on the evaluation of the frequency and closeness of different building energy consumption patterns during specific load conditions. This new EPI is built upon one of the newest and most promising techniques for pattern recognition in time series that is the so-called Contextual Matrix Profile (CMP). CMP makes it possible to automatically identify the best matching subsequences in an energy consumption time series (i.e., motif discovery) and, on the other hand, also to detect infrequent energy patterns that can be associated to anomalous trends. The developed benchmarking system can be useful to recognize in a faster and automatic way buildings with poor performances compared to its peers, for which the application of targeted energy management strategies should be prioritized. The benchmarking tool, therefore, could be useful for energy mangers, stakeholders or building owners, to improve performance of their building portfolios, and for grid operators or governments to define specific energy management strategies and to target financial demand response programs (e.g., Time of Use tariff, Critical Peak Pricing, Real-Time Pricing) conceived for specific customer groups.