Novel Technique for Feature Computation and Clustering of Smart Meter Data

摘要

Application of data analytics and machine learning techniques on high frequency smart meter data is giving interesting insights about consumption pattern of electricity by residential customers. In this paper one such technique is proposed for clustering the smart meter data based upon the computed feature set. A feature set (with five features) consisting of three novel features is proposed. Effect due to temperature variations is also taken into consideration. These features are clustered using two unsupervised machine learning techniques i.e. K means and K medoids. MATLAB and R programming software are used for carrying out feature computation and clustering. Evaluation of model is done by computing silhouette coefficients. Very few negative silhouettes are found with average silhouette coefficient value ranging from 0.25 to 0.28 which shows that clusters are well separated from each other.