Leveraging Hybrid LSTM-Attention-CatBoost Model for Complex Energy Behavior Forecasting

We present an advanced energy prediction model that addresses energy imbalance by integrating Long Short-Term Memory (LSTM) with Attention Mechanism for time-series analysis and categorical boosting (CatBoost) for categorical feature optimization. This model accurately predicts the complex energy behaviors of prosumers who both generate and consume energy. Simulation results show that our model outperforms traditional methods with the lowest Mean Absolute Error (MAE) of 48.514. By achieving such a high level of accuracy, this model improves grid operation efficiency, offers a practical solution for smart grid management.