A Composite-Window-Based Load Event Detection Method in Non-Intrusive Load Monitoring

Abstract

Load event detection is the fundamental step in event-based Non-Intrusive Load Monitoring (NILM). Accurate detection of the complete transient process of events can effectively improve the subsequent event-based appliance state identification and load disaggregation. However, it is challenging for the existing methods to cope with the significant variances in the duration of events of different types. Moreover, load fluctuation and noise can also affect the detection accuracy. From this point, this paper proposes a load event detection method based on the composite-window strategy. The proposed method first utilizes the central sub-window to detect the edge of load events and tunes the window length by moving the other two sub-windows to cope with the various durations of events. Subsequently, the proposed method detects the start and end points of the events based on the principle of change point detection. Finally, the fluctuating events are identified and filtered out based on their waveform characteristics. Experiments conducted on the private and public BLUED datasets demonstrate that the proposed method outperforms the two benchmarks in general.