Método para descubrir patrones espacio-temporales de comportamiento de usuarios eléctricos utilizando herramientas de aprendizaje automático

This research presents a novel approach to define geographic areas with typical electricity consumption profiles from smart meter records, using machine learning and spatial analysis methods. Distribution system operators must guarantee the quality and reliability of the electric service. To achieve this objective, electricity distribution utilities need to know in detail and with an adequate periodicity the consumption profiles of their customers. Modern telemetering devices, such as smart meters, open the door to an immense amount of data and new analysis, due to a higher frequency and precision of consumer electrical consumption. However, conventional methods cannot deal with the voluminous and fast gathered data by smart meters. The objective of this research is to apply machine learning techniques combined with spatial analysis to generate more efficient and accurate load profiles in the areas of study. The study analyzes a voluminous database of measurements gathered during 4 years by 1000 georeferenced smart meters located in the city of Guayaquil in Ecuador. In the study an unsupervised learning methodology to group and classify the time series of energy measurements, using the dynamic time warping technique to discover typical load profiles according to their characteristic weekly consumption, is applied. Next, we perform a restricted space-time analysis to define geographic areas with constant and predictable behavior ​
​L'accés als continguts d'aquesta tesi queda condicionat a l'acceptació de les condicions d'ús establertes per la següent llicència Creative Commons: http://creativecommons.org/licenses/by/4.0/