Enhancing EV Charging Load Forecasting with Change-Point Aware Attention Mechanism

The accurate prediction of charging load for electric vehicles (EVs) is of vital importance for the efficient management of the power grid. However, the intermittent and sudden characteristics of EV charging pose difficulties for traditional deep learning models such as Convolutional Neural Network - Long Short-Term Memory Network (CNN-LSTM) with Attention Mechanism (AM). These models often struggle to fully capture the abrupt changes from zero to non-zero load in EV charging data, which is crucial for accurate prediction. To address this issue, we introduce the Change Point-Aware Attention (CPA-Attention) mechanism in the CNN-LSTM-AM framework. Our method can detect the load changes from zero to non-zero in the input sequence and generate a binary mask. Through an enhancement factor, it strengthens the attention mechanism's focus on these key charging start points. In the ACN-Data benchmark test, our CPA-CNN-LSTM-AM model has significant improvements in MAE, RMSE and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> metrics within the multi-step prediction range (4 to 96 steps) compared to the baseline CNN-LSTM-AM model. These findings indicate that explicitly modeling charging start events can enhance the accuracy of electric vehicle load forecasting.