Adaptive Integration of Steady-State Change Features for Enhanced Multi-Label Energy Disaggregation

Energy disaggregation plays a vital role in analyzing household and building energy consumption by identifying the operating states of multiple appliances from aggregate power measurements. However, achieving accurate multi-label energy disaggregation remains challenging due to overlapping appliance signatures, variations in power consumption patterns, and complex steady-state transitions. In this work, we present an adaptive framework titled. The proposed methodology utilizes Adaptive Feature-Temporal NILM (AFT-NILM), Lightweight Hybrid Interpretable NILM (LHI-NILM), and Hybrid Ensemble NILM (HE-NILM) to improve appliance-level identification and energy disaggregation performance. Initially, steady-state change features are adaptively integrated to capture significant variations in appliance operating behavior. The extracted features are processed through AFT-NILM to learn temporal consumption characteristics, while LHI-NILM provides efficient and interpretable appliance-level disaggregation. Furthermore, HE-NILM combines the strengths of multiple prediction mechanisms to enhance multi-label classification accuracy and robustness. Experimental analysis demonstrates that the proposed framework effectively identifies multiple appliance states from aggregate energy signals and improves overall disaggregation performance. The adaptive integration of steady-state change features, together with the complementary capabilities of AFT-NILM, LHI-NILM, and HE-NILM, contributes to more accurate, efficient, and reliable non-intrusive load monitoring for modern energy management systems.

Keywords: Non-Intrusive Load Monitoring (NILM), Energy Disaggregation, Multi-Label Classification, Steady-State Change Features, Adaptive Feature-Temporal NILM (AFT-NILM), Lightweight Hybrid Interpretable NILM (LHI-NILM), Hybrid Ensemble NILM (HE-NILM), Appliance Identification, Energy Monitoring, Smart Energy Management.