jaue2021-071: A Short-term Electric Load Forecasting Method Based on Dual-stage Attention Mechanism and Long Short-Term Memory Neural Network

Abstract

Short-term electric load forecasting is the basis for reasonable dispatch and smooth operation of power grids. In
order to improve the accuracy of load forecasting, a dual-stage attention mechanism was introduced to long shortterm
memory (LSTM) as the predictive model, to extract association rules between input features and time
dependencies between history time points. The feature attention mechanism was used for mining the relationship
between the target parameter and other state information automatically to correct prediction result appropriately,
which also overcomes the limitation by the preset threshold in traditional association rule mining algorithm. The
temporal attention mechanism independently selects the key time points of historical information, and further
enhances the information expression of key time points on the basis of the LSTM time series model, and improves
the stability of the prediction effect for a longer period of time. Taking the load data of a school as an actual
calculation example, the results show that this model is more accurate than other models. Furthermore, by using
feature attention mechanism, the model can track how input features influence outputs at each step of inference.