Image Segmentation-based Event Detection for Non-Intrusive Load Monitoring using Gramian Angular Summation Field

A Non-intrusive Load Monitoring approach extracts the operation time of individual appliances from an aggregated load measured at a single entry-point using their energy consumption characteristics. Event detection represents an important step for load segregation where energy state change on aggregated load and duration are obtained. This paper proposes two event detection algorithms using image segmentation based on two diverse methodologies namely, k-means clustering and thresholding technique. The proposed algorithms are applied to an image generated by encoded Gramian Angular Summation Field of time series data. The method is simple to implement and efficient in computation. The proposed approach is tested and validated using real-world load measurements: Almanac of Minutely Power dataset, and for said purposes, comprehensive simulation studies have been carried out on a low-cost Raspberry Pi 3B+ platform. The corresponding results are promising in terms of event detection and indicate that the proposed approach has a strong potential towards more robust and accurate event-based NILM systems.