Abstract：Accurate short-term electrical load forecasting is of great significance for the design and

optimization scheduling of integrated energy systems（IES）. However，the load data in real integrated energy

systems are low-quality and fluctuating，so the forecasting accuracy of existing prediction models is low. A shortterm electrical load forecasting method based on attention-based long short-term memory（AT-LSTM） and

Stacking learning was proposed. Under the framework of Stacking ensemble learning，AT-LSTM，random forest

and decision tree were ensembled to forecast short-term electrical load which can make up for the low prediction

accuracy of a single model. Based on the exploratory analysis results of data，the data feature engineering model

was constructed to input features，and this prediction method was used for short-term electricity load prediction.

The experimental results of the integrated energy system in Beijing show that compared to other algorithms，the

proposed method has a maximum prediction error reduction of 24.8%.

      Key words：short-term electrical load forecasting；attention-based long short-term memory（AT-LSTM）；

Stacking ensemble learning；integrated energy system（IES）；feature engineering