A hybrid approach for intrusive appliance load monitoring in smart home

Appliance Load Monitoring (ALM) has become a crucial task in energy sector since the residential loads have been ever-increasing in the recent years. Several studies have been undertaken to monitor energy consumption of household appliances while also analyze the power data to obtain more useful insights of consumers' behaviors. The remaining challenge of the recent approaches is automatic appliance recognition. In this work, we propose a novel hybrid method which includes two main processes, namely the feature importance process and the appliance identification process. In the first phase, feature importance process extracts the temporal trends. We then replace the classification layer of Convolutional Neural Network (CNN) by the SVM classifier; thereby achieving a set of important features which is data input for the next phase. After that, we set the CNN's weights based on the analyzed feature importance of SVM, instead of initializing weights randomly. As a result, the proposed method of this study outperformed other approaches with more than 90% for both of accuracy and macro F1-score.