Fluctuating power holder and operating power rectification for energy disaggregation in non-intrusive load monitoring

Abstract

Non-intrusive load monitoring is a technique that separates the energy consumption of the individual loads contained in aggregated energy consumption data. However, current energy disaggregation methods suffer from significant bias, which primarily originates from the disaggregation of appliances in the ON states according to extensive experimental results reported in the literatures. The diversity of load types and operating modes leads to significant variations in actual operating power across different scenarios, which are the key factors contributing to this problem. To address this problem, we are motivated to propose a method for extracting power fluctuation information caused by load state transitions from the aggregated signal. And by preserving this fluctuation for a certain period, the energy disaggregation algorithm can more accurately capture the actual energy consumption of the target load from the input signal. Subsequently, a novel energy disaggregation model incorporating both temporal and spatial convolution pathway is proposed to extract spatial and temporal features from the raw aggregated signals and fluctuating power, with spatiotemporal fusion achieved through lateral connections. Further, we propose an operating power rectification unit to adjust the predicted operating power of the target appliance. The extensive experimental results on publicly available datasets demonstrate that the proposed method not only exhibits remarkable precision and robustness in energy disaggregation but also shows a distinct advantage in event detection compared to reference methods across multiple metrics.