Time-Series Forecasting of Energy Consumption Using LSTM Networks for Optimized Microgrid Management

Accurate energy consumption forecasting is crucial for the efficient management of microgrids, especially as renewable energy sources become increasingly integrated into power systems. In this paper, we present a Long Short-Term Memory (LSTM) based model for time-series forecasting of energy consumption using smart meter data from Bareilly, India. The dataset, consisting of over six million samples recorded at threeminute intervals, was preprocessed and used to train an LSTM network that effectively captures temporal dependencies in energy usage. The model was evaluated using metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and $\mathbf{R}^{2}$ score, demonstrating its strong predictive performance with an $R^{2}$ value of 0.85984. While the model accurately predicts most energy usage patterns, it faces challenges in capturing extreme consumption spikes. These results highlight the potential of deep learning models in optimizing energy distribution.