Self-Adaptive Non-Intrusive Load Monitoring Using Deep Learning

To optimize energy utilization, efficient energy management practices are important. Non-intrusive load monitoring (NILM) has emerged as a promising solution, particularly with the advent of deep learning techniques. This paper introduces a novel approach to NILM: Self-Adaptive Non-Intrusive Load Monitoring using Deep Learning. Conventional NILM models often struggle to adapt to changes in power consumption patterns, especially with aging appliances. To address this challenge, we propose a Self-Adaptive NILM model that integrates deep learning techniques with transfer learning and pseudolabeling. Unlike conventional NILM models, our approach incorporates a unique self-adaptive feature, enabling the model to autonomously adapt to changing power patterns caused by aging appliances. We leverage synthetic data generation and advanced neural network architectures to train and validate our model, achieving exceptional accuracy rates in disaggregating power consumption. Experimental results demonstrate the effectiveness of our approach, with accuracy exceeding 97% over a six-year period for a three-phase refrigerator. This research not only fills a significant gap in existing NILM literature but also sets the stage for a more robust and resilient energy management system.