Extração de características por transferência de aprendizado profundo no âmbito do monitoramento não Invasivo de cargas residenciais

Abstract

The development of techniques that allow the efficient identification of loads, ideally in a nonintrusive way, is a key factor for the practical implementation of home energy management systems. Recently, the use of techniques based on deep learning has gained attention in different domains, such as signal and image processing, highlighting the models based on convolutional neural networks. However, the efficient training of these models is strongly dependent on the amount and balance of data, i.e., characteristics that are not normally found in nonintrusive load monitoring datasets. To deal with these challenges, this thesis proposes an approach based on three stages, which are: (i) transformation of time series into 2D images; (ii) feature extraction using deep transfer learning; and (iii) classification/labeling of loads. In this sense, it was considered five loads present in the Reference Energy Disaggregation Dataset. The results indicate that the proposed approach was able to obtain an average f1-score of 84.2%. Moreover, from the analysis of the results, it was also possible to observe a greater capacity of the proposed approach to infer and generalize its responses compared to other evaluated approaches, presenting a consistent result even in the face of unbalanced data.