The Development of a Novel Event Detection Approach for Nonintrusive Load Monitoring

Event detection is a crucial step in non-intrusive load monitoring (NILM). Adequate detection methods not only enhance the accuracy of subsequent NILM results but also facilitate rapid understanding of real-time system status, maintaining power supply quality and power dispatching security. In this paper, a novel feature-based approach is proposed. This approach uses patterns of root-mean-square (RMS) current mapped in the polar coordinate pattern as features. Regarding the event detection method that eliminates the need for threshold configuration, this paper uses patterns shown in a polar coordinate pattern as training data to train a convolution neural network (CNN), which is then employed to classify patterns. Accordingly, the proposed approach classifies event and non-event scenarios and detects load events without the need to configure thresholds. Experiments with 3 public datasets and 1 private dataset including DC-invertor-based appliances, revealed that the proposed approach achieved a recognition rate of 98%~99%, which verified its superior performance relative to other existing methods.