Improving the Estimation of Smart Homes Energy Consumption by Clustering the Correlated Loads

In this paper, an estimation approach based on Deep Neural Networks (DNNs) is proposed for estimating the consumption of smart homes appliances. We consider the difference among the consumption of appliances to improve the accuracy of the load estimation. Therefore, we apply a clustering technique to detect the correlation among data. Then, we design a separate DNN corresponding to each cluster to estimate the load consumption. The proposed approach is compared to some prevalent machine learning based estimation approaches, such as, multilayer perceptron, support vector machine, random forests, gradient boosting machine, and linear regression. The comparing metrics for the accuracy of the estimations are linear correlation factor, root mean square error, and mean absolute error. The numerical results show that the proposed estimation approach outperforms the other approaches in all investigated criteria. Moreover, less complex DNNs are required when the dataset is clustered.