Density Clustering Analysis for Improved NILM Appliance Identification

In unseen scenarios where the composition of appliances is unknown, the existing unsupervised non-intrusive load monitoring (NILM) can determine the appliances present in the target scenario and identify their operating status by autonomously analyzing the aggregated load data. However, factors as improper setting of similarity and distance thresholds or poor algorithm performance may lead to serious misidentification results. To this end, a further data processing method is proposed to evaluate and realign the initial unsupervised NILM results based on density clustering analysis. First, density cluster analysis is performed on the data samples of each appliance resulted from unsupervised NILM, and the overlap degrees of sample distribution within each cluster and between different clusters are calculated to divide and merge clusters respectively. On this basis, data samples with high confidence from each reorganized cluster are extracted as typical samples. Then, the label propagation algorithm is used to update the labels of all appliance samples, thus reduce the erroneous recognition results. The experimental results in multiple real-world scenarios show that the proposed method can improve the consistency of samples in appliance clusters, which proves its effectiveness.