Classification for consumption data in smart grid based on forecasting time series

• A novel scheme for classification of consumption time series has been proposed.
• The scheme is based on forecasting and comparing consumption values to measurements.
• The method performs better than existing solutions both on artificial and real data.
• The proposed design of the proposed scheme implied clustering capabilities.

One of the most important tasks of present day smart grid implementations is to classify different types of consumers (households, office buildings and industrial plants) because they may be served by the power supplier with different parameters, rates, contracts.In this paper, we propose a novel classification scheme for smart grid systems where the collected data are processed in order to increase the efficiency of electricity transportation as well as demand-supply management. The new scheme is based on forecasting the consumption time series obtained from a smart meter. Class assignment is determined using the forecast error. Different linear and nonlinear methods were tested based on the corresponding assumptions on the statistical behavior of the underlying consumption time series.Performance tests were carried out with simulations in order to demonstrate the capabilities and to compare the achieved performance of the proposed scheme with existing solutions. The simulations have been executed using (i) artificially generated consumption data, which data came from a bottom-up semi-Markov model and (ii) real, measured power consumption data as well. The parameters of the model have been validated on real data. The numerical results have demonstrated that our method can better model and classify the consumption patterns of office-buildings than the existing methods. As a result, the proposed method may prove to be a promising classification tool.

Figure optionsDownload as PowerPoint slide