The modern power systems have developed considerable complexity as electricity systems become increasingly decentralized an‎d integrate renewable energy resources, leading to an accelerated transition to smart microgrids. Reliable Short-Term Load Forecasting (STLF) contributes to balancing supply an‎d deman‎d, distributed energy resource (DER) scheduling, an‎d electric grid stability. However, traditional statistical forecasting techniques like ARIMA an‎d linear regression do not accommodate nonlinear dynamics, stochastic events, an‎d the weather dependence that results in microgrid complexity. The seminar explores using deep learning techniques, specifically Long Short-Term Memory (LSTM) an‎d Gated Recurrent Unit (GRU) networks, to improve STLF performance as a forecasting approach for smart electrical microgrids. The seminar begins with a theoretical background an‎d literature review that identifies critical limitations of current research in two forms: model generalization, feature representation, an‎d interpretability. Utilizing research limitations as justification, a new forecasting system architecture an‎d corresponding framework are proposed, including new data preprocessing methods, model hyperparameter optimization, a comparative eva‎luation of model performance eva‎luation for forecasting, using MAE, RMSE, an‎d sMAPE, for an open access datasets (UCI, NREL, OPSD, Kaggle). The seminar concludes that deep learning methods present a scalable, robust baseline for intelligent energy forecasting. Future research directions include hybrid ensemble architectures, multi-modal data inputs, explainable AI techniques, an‎d federated learning to be used in forecasting across distributed microgrids while retaining model input privacy

10/29/2025 12:00:00 AM

88050

1404/08/10

Deep Learning Methods for Power Load an‎d Renewable Energy Forecasting in Smart Electrical Microgrids

Deep Learning , Smart Microgrids , Short-Term Load Forecasting (STLF) , Renewable Energy Forecasting , LSTM , GRU