A hybrid model for appliance classification based on time series features

Abstract

The non-intrusive load monitoring, which gets the energy usage information of each appliance by analyzing the aggregative signal in the centralized point is beneficial to electrical power savings of buildings. As an important step in the non-intrusive load monitoring, the appliance classification has gained great attentions. The feature extraction, the design of classification structure and the selection of classifiers are three important parts in the appliance classification. In this paper, a hybrid appliance classification model is proposed. In the hybrid model, the time series features are utilized to depict the characters of appliances, which is different from the conventional perspective of feature extraction. A secondary ensemble structure is designed to handle the class imbalance multiclass classification problems. The class imbalance is often the case in appliance classification but no effective solutions has been put forward yet. Besides, the random forest is selected as the binary classifiers in the secondary ensemble structure. The effectiveness of the proposed model is verified on the extended PLAID dataset. The experimental results indicate that the time series features provide better overall performance compared to the widely used features, the random forest performance better than the widely used classifiers and the overall performance of the proposed model is better than that of the state of the art model.