基于改进聚类和DTW的非侵入式负荷识别

Non-intrusive Load Identification Based on Improved Clustering and DTW Matching Algorithm

利用事件检测提取的功率特征数据进行K-medoids聚类分析,聚类结果将同DTW匹配距离结合识别电器设备,最终能够有效地识别出事件设备。通过改进K-medoids方法的初始点选择方法以及距离判别方法有效地改善了聚类结果;改进了聚类DB评判方法,既能有效评判聚类效果,也能简化计算。良好的聚类结果将为进一步匹配分类做铺垫,最后详细分析了DTW模板匹配与聚类结合识别负荷的过程,并且对多种匹配模式的DTW方法进行了实验,经过实验选择了合适的匹配模式,并且实验验证其能够有效减小匹配距离。

The K-medoids cluster analysis is performed using the power signature data extracted from the event detection. The clustering results will be combined with the DTW matching distance to identify the electrical equipment,and eventually the event equipment can be effectively identified. By improving the initial point selection method of the K-medoids method and improving the distance discriminating method, the clustering result is effectively improved;and the cluster DB evaluation method is improved,which not only can effectively evaluate the clustering effect, but also can simplify the calculation. Good clustering results will pave the way for further matching and classification. Finally, the process of load identification combined with DTW template matching and clustering is analyzed in detail.Experiments are conducted on the multiple methods of matching modes of DTW, and suitable matching mode is selected through experiments.Models, and experiments verify that it can effectively reduce the matching distance.