Desagregación del consumo eléctrico e identificación de dispositivos en hogares en el contexto de monitorización no intrusiva (NILM) utilizando Deep Learning

Este trabajo propone una metodología innovadora dentro del campo de NILM – Non Intrusive Load Monitoring. Dicha metodología cubre todas las actividades necesarias desde la preparación y tratamiento de datos hasta el entrenamiento y validación de los modelos de Machine Learning, siendo especialmente innovadora la primera parte del mismo. Con esta preparación de datos, y enfocando el problema como uno de clasificación, se ha diseñado una CNN que ofrece buenos resultados en el reconocimiento de electrodomésticos en el ámbito NILM.

Research in the field of NILM (non-intrusive load monitoring) has seen an unprecedented surge in the past years mainly due to the increase in domestic household energy monitoring, finally making this alternative viable. Over these years, researchers have tried and implemented multiple techniques and algorithms that managed, with a certain degree of success, to perform energy disaggregation. However, these techniques are often very niche, do not generalize very well because of their complexity, or they simply do not perform all that great. This paper proposes the vizgram, a new approach on how to process and interpret energy data that serves as the model input, transforming it into an image that can later be fed to a Convolutional Neural Network, taking advantage of the already proven usefulness and remarkable performance of these types of architectures in image classification. The proposed approach has been applied with success to a real case consisting of five different appliances.