Data-Driven Recommendation Model Based on Meta-Learning for Non-Intrusive Load Monitoring

Smart meters with Non-Intrusive Load Monitoring (NILM) technology is explored as a means of reducing energy consumption and minimizing emissions in residential settings. Various NILM disaggregation algorithms have been proposed, but their effectiveness varies across different scenarios. To address this issue, a method called Data-driven Recommendation Model (DRM) based on meta-learning is proposed in this paper. DRM recommends the appropriate algorithms for various problems using the existing knowledge repository. Besides, a novel metafeature extractor named Meta-Learning Encoder-Decoder is designed to efficiently process the deep meta-features from a more significant number of training samples, which can be implemented to various meta-learners. Experimental results indicate that the proposed DRM exhibits strong learning and generalization capabilities in terms of the efficiency of load disaggregation for datasets with different characteristics.