کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Attention-Based Seq2Seq Framework for Short-Term Electricity Load Forecasting with Meteorological and Calendar Features (Case Study: Iran)

نویسندگان :

Nasim Amani¹ Ghasem Alipoor² Alireza Kokabi³

1- دانشگاه صنعتی همدان 2- دانشگاه صنعتی همدان 3- دانشگاه صنعتی همدان

کلمات کلیدی :

Attention Mechanism،Deep Learning،Bayesian Optimization،Short-term load forecasting،Meteorological Variables،Sensitivity Analysis،Electricity demand forecasting،LSTM networks

چکیده :

Accurate electricity load forecasting is strongly influenced by meteorological and temporal factors. This study proposes an enhanced sequence-to-sequence (S2S) deep learning framework with Long Short-Term Memory (LSTM) layers, incorporating grouped meteorological variables, calendar features, and attention mechanisms to improve predictive performance. A sensitivity analysis identifies the most influential weather factors. Additive, global, and multi-head attention mechanisms are evaluated at one-hour and 24-hour (day-ahead) forecasting horizons. Hyperparameters are tuned using Bayesian Optimization. Using real-world data from Hamedan Province, Iran, the proposed model achieves significant accuracy improvements—reducing Root Mean Squared Error (RMSE) by up to 42.6% compared to baseline methods. These results highlight the effectiveness of combining structured meteorological features and attention-based modeling for robust short-term electricity demand forecasting.