Feature Extraction and Energy Disaggregation of Commercial Loads Based on Discrete Wavelet Transform and Sequence-to-Point Learning

Commercial loads exhibit distinct energy consumption patterns and load characteristics compared to residential loads, necessitating the development of efficient non-intrusive load monitoring (NILM) methods for commercial buildings. Given this background, this paper presents a methodology that specifically targets the challenges associated with commercial loads. By considering the unique energy consumption and load behavior of commercial buildings, our approach focuses on extracting load features and performing load disaggregation. We leverage the discrete wavelet transform (DWT) for efficient load feature extraction and accurate detection of transient load event edges, and employ a sequence-to-point (seq2point) learning approach for load disaggregation. Experimental results conducted on the commercial building energy dataset (COMBED) validate the efficacy of the proposed methodology in accurately detecting and disaggregating commercial loads, specifically elevators and air handling units. This research contributes to the development of dedicated NILM approaches for commercial loads, enabling better energy management and efficiency in commercial buildings.