Subspace Projection Method Based Clustering Analysis in Load Profiling

Customers of different contract types have different shapes in daily load profiles in the manner of different characteristics. Therefore, maximally capture local and global shape variability is essential in load profiling which exhibits the customers´ different behaviors and characteristics. Existing approaches are focusing on the global property by considering all dimensions in the data set. However, the load shapes are determined by subspace of dimensions in most of the time. In this paper, we use subspace projection methods (subspace clustering and projected clustering) to capture such subspaces of load diagrams which maximize the difference between particular load shapes in different groups of customers. Also, we have treated clustering as classification to select most appropriate cluster numbers. The contribution of our study is that we have interpreted the strength and weakness of subspace projection method in load profiling. The result shows that subspace projection based method outperforms traditional clustering algorithms.