Forecast based classification for power consumption data

In a smart power distribution system a crucial task is to categorize properly the different types of power consumers in order to optimize the transportation grid as well as the rates and contracts between the power suppliers and consumers. By using intelligent meters and analyzing the behavior of consumers relevant information can be obtained, which may be used for capacity distribution or to have more precise estimation for expected energy consumption for individual consumers or local regions. In this paper, we introduce new results on a recently proposed classification scheme based on the forecast of the consumption time series obtained from a smart meter using nonlinear methods. The new results include i) tests on measured power consumption data and performance evaluation in different cases; ii) comparison with other methods. The numerical results prove that our method is capable of distinguishing different consumers with different consumption patterns at lower error rate than the existing methods. As a result the forecast based method proved to be the most promising classification tool in real applications.