Enhanced Temporal Attention based Hybrid Model for Non-Intrusive Load Monitoring

Kurzfassung

Non Intrusive Load Monitoring (NILM) enables appliance level power disaggregation using a single smart meter, avoiding the need for multiple sensors. However, accurate disaggregation remains challenging due to overlapping appliance signatures, temporal complexity and unknown loads. This paper presents an enhanced temporal attention based hybrid model that integrates Transformer architectures with appliance specific Temporal Convolutional Networks (TCN). The model combines Nyström attention, Grouped Query Attention and Rotary Positional Encoding to achieve efficient and scalable temporal modeling. A multi task learning framework is used for simultaneous power estimation and ON/OFF state detection. Experimental evaluation on the UKDALE dataset is conducted using evaluation metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and F1-score, together with cross-house validation to assess the accuracy, robustness, and generalization capability of the proposed model.