Weather-Aware Energy Forecasting with NeuroFusion: A Hybrid Deep Learning and Gradient Boosting Framework

Accurate forecasting of energy consumption and generation is essential for the efficient operation of smart grids, particularly in weather-sensitive environments. This study proposes a hybrid ensemble framework, NeuroFusion, which integrates Long Short-Term Memory (LSTM) networks with extreme Gradient Boosting (XG-Boost) within a meta-learning structure. The LSTM model captures sequential and long-range temporal dependencies inherent in energy usage, while XGBoost captures non-linear interactions among meteorological variables through boosted decision trees. Combining these complementary models results in a robust and high-fidelity forecasting system that adapts to variable weather patterns. NeuroFusion was evaluated using real-world hourly datasets comprising both weather and energy data. The proposed model achieved prediction accuracies of up to 96.3% for energy generation and 91.3% for energy consumption, significantly outperforming individual base models. In terms of error reduction, NeuroFusion decreased the root mean squared error by more than 74% for consumption and over 77% for generation when compared to standalone implementations. These results confirm the efficacy of combining memory-based and gradient-based learning techniques for accurate and resilient forecasting in smart grid environments.