Automated Anomaly Detection In Energy Consumption Time Series Of Buildings Through Pattern Recognition Techniques

Abstract

Commercial buildings are significant consumers of electrical and thermal energy, therefore energy savings, improving energy efficiency, and reducing greenhouse gas emission are the purposes for building owners, operators, and stakeholders. On the other hand, energy analysts have to understand the energy consumption behavior by looking for changes in energy patterns that may imply device failures or anomalous behavior. This master’s thesis deals with an energy data-mining approach that performs automated Anomaly Detection, through data analytics techniques called Matrix Profile (MP) and its extension Contextual Matrix Profile (CMP). This work aims to extract from large energy time-series data generated by sub-meters and smart sensors installed in Politecnico di Torino buildings, anomalous energy consumption patterns and to understand the root causes of the detected anomaly. The framework built up combines a hierarchical cluster algorithm, which helps to aggregate power consumption daily patterns, with MP and a final descriptive statistics outliers’ analysis.

References

• Simone Vitale. AUTOMATED ANOMALY DETECTION IN ENERGY CONSUMPTION TIME SERIES OF BUILDINGS THROUGH PATTERN RECOGNITION TECHNIQUES. Rel. Alfonso Capozzoli, Marco Savino Piscitelli, Roberto Chiosa. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Energetica E Nucleare, 2022