Abstract:Non-intrusive load monitoring (NILM) technology facilitates the analysis of aggregate electrical consumption data, employing energy disaggregation algorithms to estimate the power usage of individual appliances. To tackle the inherent challenges associated with non-intrusive energy disaggregation, this paper introduces an innovative energy disaggregation algorithm that leverages a Variational Autoencoder (VAE) framework. The proposed method capitalizes on the VAE encoder's superior feature extraction capabilities, which enable the precise identification of unique energy consumption patterns for specific devices, thereby enhancing the accuracy of the disaggregation process. Empirical evaluations reveal that, in comparison to the conventional Sequence to Sequence (S2S) approach, the newly developed model achieves a reduction of approximately 10% in mean absolute error across all tested appliances, alongside a more than 15% improvement in the F1 score. These findings not only underscore the enhanced precision of the VAE in energy prediction but also highlight its significant advancements in the recognition and isolation of target appliance power signatures. Consequently, the VAE-based energy disaggregation algorithm presented herein represents a promising development, effectively contributing to the decomposition of aggregate load energy and paving the way for broader practical applications of NILM technology.