Fine-tuning non-intrusive load monitoring model through user interaction: A practical approach to appliance recognition with limited labeled data

A novel fine-tuning method is introduced for Non-Intrusive Load Monitoring (NILM), using transfer learning to adapt pre-trained deep learning models for deferrable appliances with distinct short cycles (such as washing machines and dishwashers). This approach enhances model generalization by using limited user-labeled data with readily available, low-frequency aggregate consumption data from smart meters. The method eliminates the need for high-frequency sampling or intrusive sub-metering, and high accuracy in deployable NILM applications. The results show that high accuracy in appliance state recognition is achieved with minimal user interaction, requiring only a small number of labeled appliance activations. The method achieves competitive results compared to state-of-the-art methods, providing a practical and effective NILM solution suitable for widespread adoption by consumers and utility companies.