A non-intrusive load monitoring algorithm based on real-time feature extraction and deep learning model

• A two-level real-time feature extraction method is proposed using signal windows and the SSSC-EMD signal processing technique. • IMFs are extracted during the signal window using SSSC-EMD, and features such as RMS, MEAN, DIFF, and MAX are calculated and stored for model training and testing. • The impact of the number of household appliances on the deep learning model architecture is analyzed, revealing that changes in appliance count require modifications to the model’s layers. • An adaptive deep learning model is developed based on BiLSTM layers to accommodate varying input complexities. • A custom objective function is introduced to automatically determine the optimal model architecture depending on the number of appliances under analysis. Researchers are increasingly interested in Non-intrusive Load Monitoring (NILM) as a cost-effective alternative to Intrusive Load Monitoring (ILM) for smart home energy monitoring. However, most NILM studies assume a fixed number of appliances and adjust classification models accordingly. This approach disregards the reality of varying appliance numbers in households, which can significantly impact model accuracy. To address this limitation, in this study, a comprehensive comparison between the proposed model and existing methods is conducted. The results show that the proposed model outperforms the existing methods on the UK-DALE and EMBED databases with an accuracy increase of 9.492% and 8.386%, respectively. Also, the accuracy of the proposed model has reached 99.74% compared to conventional methods, which shows its significant superiority over the lower accuracies of other models, including 95.11%, 92.09%, 88.70%, and 75.38%. This innovation combines a two-stage feature extraction method based on Soft Sifting Stopping Criterion Empirical Mode Decomposition (SSSC-EMD) and an adaptive deep learning model with architecture tuning capabilities, in order to improve the performance and accuracy of home appliance detection under different conditions.