A User-in-the-loop Digital Twin for Energy Consumption Prediction in Smart Homes

This paper describes a digital twin of smart homes able to predict future energy consumption and help the user make better decisions about the activation of smart appliances and the scheduling of automations comprising different appliances' activations. To deal with the problem of time series forecasting for energy consumption prediction, a deep learning approach based on Long-Short Term Memory has been adopted, and a grid search has been used to identify the values of hyperparameters with the best prediction accuracy. Proper information visualization and interaction features have been then implemented in the digital twin interface to explain to the user the predicted consumption data and the reasons underlying warnings and suggestions provided by the system. In this way, the digital twin becomes a system based on artificial intelligence that exhibits an explainable behavior, which allows the user to make decisions about smart home management in a more conscious and sustainable way.