Non-Intrusive Load Monitoring in Smart Grid through Feature Extraction and Deep Learning

This paper presents a novel non-intrusive load monitoring (NILM) system for household appliance identification, addressing the critical need for granular electricity consumption data in smart grids. A custom-developed hardware platform enables high-sampling-rate current data acquisition from a single main supply point. Discriminative harmonic features are then meticulously extracted from these current signals, capturing unique appliance fingerprints for both linear and non-linear loads. A Feedforward Neural Network (FNN) is employed to learn the complex mappings between these features and 18 distinct appliance states/combinations. Experimental validation shows the system’s high performance, achieving a peak single appliance recognition accuracy of $98.7 \%$ and combined appliance recognition accuracy of $\mathbf{9 6 . 7 \%}$. The robust feature engineering and optimized FNN architecture prove effective in distinguishing diverse load patterns, underscoring the potential of high-fidelity current signature analysis combined with deep learning for detailed energy management insights and advanced smart grid applications.