Machine Learning for Energy Consumption Prediction in Smart Homes

Predictions of energy forecasting in smart homes is very necessary for the energy management and proper directions of development in future. Nevertheless, standard assumptive frameworks of the energy consumption prediction have the problems of inimitability of long range dependence and highly complicated temporal structures. We propose a new method in this paper that increases prediction accuracy of energy consumption by using TCNs with the help of Attention Mechanisms. The model can pay specific attention to time intervals that impact the computation of energy consumption patterns with the help of attention mechanism that the model introduces, used in TCNs to solve problems that deal with data with long range dependencies. This work utilizes real world smart home energy data to benchmark our approach to how these types of data are usually forecast using standard time series models, through our model forecasts are both better in short term and long term. The proposed model is relatively reliable, adaptive, which can be effectively exploited in terms of energy distribution and savings, cost cut downs and energy efficient measures