Predicting the Usage of Energy in a Smart Home Using Improved Weighted K-Means Clustering ARIMA Model

Abstract

Modern development in making the whole world to act smartly has not left the home. The homes in modern era are becoming smart. The household appliances irrespective of their sizes are designed to behave smartly according to the needs of the human beings. By doing so, it helps the human beings in multiple ways. One of the ways to enjoy the benefits in a smart home is the efficiency to manage the electric energy. When electric energy is managed in a proper way, it helps the people in saving the amount that is spent on using electrical appliances and it also helps them to save the energy so that during the time of natural calamities there is no shortage of energy. To achieve this prediction of energy consumption plays a very important role. Based on the prediction done, the smart devices can be scheduled to operate so that pricing at peak hours can be reduced. Energy storage and production from renewable resources can be handled in a better way based on this forecasting of energy consumption during various seasons in a year. In this proposed new model Improved Weighted K-Means Clustering ARIMA (IWKMCA), prediction on energy consumption in a smart home is done based on the requirements during the various seasons in a year. This proposed model enhances the weighted k-means clustering algorithm to form clusters with better cluster points. Weights are added to the data points that belong to a particular cluster and based on the centroid of the clusters that are formed forecasting is done using the ARIMA model. This is done to increase the accuracy in forecasting and to reduce the various forecasting errors. The average amount of energy consumed from the smart grid by a smart home in the Pecan Project in Texas, USA is taken as the dataset for this proposed work and the average amount of energy consumed during various seasons by smart devices in the smart home is forecasted. This model shows better accuracy level when compared with the traditional ARIMA model and the Weighted K-Means Clustering ARIMA Model. The result of this proposed work shows lesser values than the ARIMA and Weighted K-Means Clustering ARIMA model when the forecasting errors like RMSE, MAPE, AIC, AICC are considered. This proposed work also shows a higher loglikelihood value than the ARIMA model proving that this model excels the standard ARIMA model in all aspects for time series forecasting.