A two-stage pattern recognition method for electric customer classification in smart grid

摘要

Identifying the consumption patterns of electric customers and grouping them to classes according to their load characteristics can be very meaningful for power supply and demand side management in smart grid. Previously, tariff structures are mainly based on the type of activity. However, the type of activity and electrical behavior of the customer have poor relationship. Using clustering techniques to classify customer according to load curves is more meaningful. This paper proposes a two-stage clustering algorithm combining supervised learning methods to classify electric customer. Firstly, clustering results are obtained based unsupervised learning method. Clustering method and number to get the result of first-stage are selected via the clustering evaluation index. Secondly, customers are reclassified using supervised learning algorithm. Different supervised learning algorithms for second-step reclassification are compared in the case studies. Case studies show that second-step reclassification can make up for the weakness of first-step clustering in load shape similarity.