AI-Based Energy Consumption Prediction and Scheduling for Smart Homes

The consumption of electricity in households shows growing levels of unpredictability and peaks in demand, which negatively affect the stability of power grids and decreases the efficiency of energy consumption. Many research works have been dedicated to short-term load forecasting and demand-side management; however, only a small number of studies pay attention to the combination of artificial intelligence algorithms that would allow for the integration of accurate load forecasting and optimal scheduling of the appliance in the household. The present work aims to provide a framework for AI-based residential energy management, which integrates hybrid SWT-LSTM load forecasting with Genetic Algorithm for optimal appliance scheduling. First, the forecasting model was developed and tested on the UK Domestic Appliance-Level Electricity (UK-DALE) data set, in which high levels of the model's forecasting accuracy were reached with Mean Absolute Error of 20.46 W, Root Mean Square Error of 44.25 W and R² coefficient of 0.9868. Then, the forecasted load profile was used in the Genetic Algorithm to determine the schedule for the washing machine, dishwasher, water pump, and iron in order to reach the reduction of the average peak demand level on 25.73%, PAR reduction on 25.73%, increase in the Load Factor on 38.62% and reduction of Peak Energy on 10.32%. For evaluating the robustness and the ability to be generalized to other types of operating conditions, the developed load forecasting model was validated on the solar powered energy consumption data set provided by the Centre for Embedded AI and Smart Energy Systems, and resulted in Mean Absolute Error of 83.09 W, Root Mean Square Error of 195.32 W and achieved the coefficient of determination R2 of 0.9823.