Non-intrusive power waveform modeling and identification of air conditioning load

As a typical flexible load, air conditioner (AC) can play a crucial role in improving energy efficiency and optimizing power grid operation. However, due to its continuously variable load characteristics, AC faces difficulties in feature extraction and unsupervised modeling for non-intrusive load monitoring. In coping with these problems, a novel fully unsupervised non-intrusive AC monitoring scheme is designed. Firstly, an autonomous AC waveform modeling method is introduced. According to the general electrical characteristics, the candidate AC (start and stop, etc.) transient waveform templates are captured from the aggregated data. On this basis, the transient waveform samples similar to candidate template are extracted and verified based on the common usage habit characteristics. Then AC model consisting of waveform template and feature vector is subsequently established by multi-dimensional clustering of the waveform samples. Secondly, an online AC state identification and power disaggregation method is proposed. Based on the dynamic time warping algorithm and guided filtering algorithm, an AC transient waveform extraction method via template matching is presented, which can extract complete and pure transient AC waveforms from the multi-appliance mixed operation scenarios. According to the extracted AC waveforms, the state identification and energy consumption estimation can be realized. In addition, the incremental clustering is carried on the online identification results to further update the established AC model. Finally, the comparison experiments on the REDD dataset and the real-world data measured from multiple users in China show that, the proposed method can construct AC templates in unseen scenarios and update the established AC models automatically, thus outperform the benchmarks in both operating state identification and power disaggregation.