Investigating Learning Transferability and Deployment for Neural NILM Strategies

Abstract

This research focuses on Non-Intrusive Load Monitoring (NILM), a crucial component of energy management, enabling users to effectively monitor and reduce energy consumption. In a previous work, we developed a hybrid nerual network with combining 1D Convolutional Neural Networks (1D CNN) and Long Short-Term Memory networks (LSTM) for disaggregating 11 appliances using the AMPds2 dataset. Based on this work, we aim to create a generalized model capable of transferring knowledge gained from one building to others. Promising results have been achieved through fine-tuning techniques, indicating the model’s adaptability and effectiveness in diverse settings. Notably, our research breaks new ground by employing transfer learning for the disaggregation of 10 appliances, surpassing previous work while maintaining a lower complexity. This study underscores the potential of NILM techniques in energy conservation and establishes a foundation for scalable, transferable models that can contribute to sustainable energy.