Threshold-Free Event Detection for Nonintrusive Load Monitoring Using Motif Difference Field Images and Convolutional Neural Networks
Yen-Kuang Lin, Men‐Shen Tsai · IEEE Access · 2025
Event detection is a critical process in non-intrusive load monitoring (NILM). Accurate detection enhances subsequent load identification and facilitates a prompt understanding of the system's real-time status. This paper proposes a novel method to determine event occurrence. First, this method transforms the root-mean-square (RMS) current into a Motif Difference Field(MDF) as a feature, generating images depicting geometric shapes rendered in different colors. These images provide clear differentiation between event and non-event occurrences. Subsequently, event and non-event classification can be achieved without threshold configuration, enhancing event detection accuracy by utilizing these images to train a convolutional neural network. The proposed method was validated on two public datasets, achieving anf1-score of 99%, which demonstrates its effectiveness and superior performance compared to existing methods.