Hybrid deep learning model for improving the operational efficiency of microgrids

Tóm tắt

Forecasting renewable energy output is crucial for efficient power system operation as their penetration rises. Solar and wind power, highly weather-dependent, require accurate, flexible forecasting methods. This study proposes a hybrid model combining Long Short-Term Memory (LSTM) networks with Inductive Conformal Prediction (ICP) to estimate generation over short-term, medium-term, and long-term horizons. The model not only enhances accuracy but also provides confidence intervals for each forecast, supporting risk-aware decision-making. Simulations based on real-world data from solar and wind plants in Vietnam show that LSTM+ICP significantly outperforms traditional Random Forest Regression (RFR) and LSTM models. Forecast accuracy is evaluated using MAE and RMSE: 0.037/0.097 MW for short-term, 0.178/0.373 MW for week-ahead, and 0.524/0.962 MW for month-ahead forecasts. These results demonstrate the model's ability to reduce systematic error and improve forecast robustness, thereby enhancing operational efficiency and stability of microgrids under highly variable conditions.