Modeling energy usage in supermarkets using AI techniques.

Supermarkets contribute significantly to global energy consumption and greenhouse gas emissions. In this context, the H2020 Project Multipack monitored the operations of integrated supermarket refrigeration systems based on CO2 and compiled the data into a comprehensive database. In this context, this thesis, carried out in collaboration with the Istituto per le Tecnologie della Costruzione(ITC) and Consiglio Nazionale delle Ricerche (CNR) - Area territoriale di Ricerca di Padova, presents a data-driven analysis to model the power consumption of a supermarket located in Trento, Italy. Multiple models, including classical machine learning algorithms and deep learning solutions, are explored. SHAP (SHapley Additive exPlanations) is employed to interpret the models, providing insights into the factors influencing power consumption and enhancing the understanding of system dynamics. The study demonstrates the effectiveness of tree-based models and artificial neural networks in predicting key electrical power input values with high accuracy. It highlights that, in specific cases, adopting novel modeling approaches such as Neural Additive Models can yield interpretable results with acceptable performance. Moreover, the research emphasizes the benefits of incorporating models capable of learning temporal dependencies in the data, showing that such approaches, in some case, can enhance predictive performance. Finally, it discuss potential applications of the developed models and identifies future research directions.