A Nearest Neighbors Gaussian Process Approach to Modeling Customer Electricity Consumption Profiles

This thesis addresses the modeling of customer energy consumption profiles within an energetic community by leveraging data analysis techniques and Nearest Neighbors Gaussian Processes (NNGP). A comprehensive dataset of energy consumption was initially analyzed to extract insights and identify inherent patterns, including short-term correlations, 24-hour cyclic behaviors, and weekly periodicities.\\ A model was developed that incorporates these multi-scale repetitive patterns and whose formulation is designed to describe the temporal dynamics of the consumption data, by accounting for both immediate sample dependencies and longer cyclic trends.\\ To estimate the model parameters, an NNGP framework was employed, offering a computationally efficient approach for simulation and inference. The performance of the proposed model was validated by comparing its outputs with synthetic data specifically generated to mimic the observed patterns, as well as against the real dataset.\\ The results show that the NNGP-based model effectively captures the essential features of customer energy consumption.