A Novel Energy Consumption Prediction Model Combining Long Short-Term Memory (LSTM) and Fractional Differential Equations (FDLE)

Optimized energy generation and smart distribution in a sustainable manner requires accurate prediction of its consumption. However, the prediction of energy demands of households remains a tedious task due to variations in patterns of energy usage. Mathematical models and artificial intelligence (AI), such as smart energy-efficient designs, strategic planning for smart grids, and Internet of Things (IoT)-enabled smart homes, have recently been considered as solutions to these issues. A major issue encountered in energy consumption prediction systems is their restricted prediction horizons, as well as their dependence on one-step predictions. This study, therefore, suggests an innovative model for the prediction of energy demand that uses a long short-term memory (LSTM) and fractional differential equations (FDLE)-based model. The proposed LSTM-FDLE model was trained to predict the collective active power generated by household devices. LSTM’s memory and sequential learning capabilities were also explored in the proposed model for comprehending the complex temporal dependencies and trends in energy consumption data. The performance of the proposed model was evaluated on real-world household energy usage data and found to achieve good prediction accuracy; the performance of the model was also better than that of some conventional one-step prediction models. Therefore, better energy generation planning, and optimal distribution systems can be achieved by the longer forecasting period provided by the proposed “LSTM” model.