H-KTT Clustering Method And Its Applications In Analyzing Large-Scale AMI Data

With the development of society,big data problem is becoming more pervasive.As one of the big data problems in the field of smart grid,the clustering of the load patterns from Advanced Metering Infrastructure(AMI)is of vital importance for a set of applications.However,several popular clustering algorithms are inefficient or even infeasible due to the growing amount of data.K-means,a simple and effective clustering algorithm,has been widely used in the solving of a range of big data problems.But the clustering result of classical K-means algorithm is sensitive to the initial centroids,and can only obtain a single local optimal solution.Both two problems become more serious with the increasing scale of dataset,which leads to unsatisfactory clustering results.To address this issue,two major improvements for enhancing classical K-means algorithm are presented in this paper.Firstly,a Hierarchical K-means(H-K-means)method is presented for enhancing classical K-means algorithm with better initial points by simplifying the structure of the original dataset.Secondly,a Hierarchical K-means enhanced by TRUST-TECH(H-KTT)method is presented for further enhancing H-K-means method by converting the original clustering problem into a nonlinear optimization problem,and solved by TRUST-TECH methodology,a high-performance methodology developed for solving nonlinear constraints optimization problems,which can systematically compute multiple,local optimal solutions in a tier-by-tier manner without getting trapped in anyone of them.The proposed H-K-means method and the proposed H-KTT method have been tested based on a large-scale AMI dataset from USA,several common clustering methods have also been applied for comparison purposes,so as to strengthen the persuasiveness of the results.As expected,the H-K-means method has achieved outstanding performance in terms of adequacy measures,a practical application and computation efficiency,and the H-KTT method can make the clustering results of the H-K-means method further improved.