Charging Load Prediction Method for Electric Vehicles using CNN-GRU-Attention model with SSA optimization strategy

This paper proposes a CNN-GRU-Attention model for short-term charging load prediction at electric vehicle (EV) charging stations. The model leverages the strengths of Convolutional Neural Networks (CNN) for feature extraction and dimensionality reduction, while Gated Recurrent Units (GRU) enhance temporal sequence learning. The attention mechanism assigns dynamic weights to the extracted features, improving the prediction accuracy. Hyperparameters are optimized using the Sparrow Search Algorithm (SSA), and an improved version, ISSA, further enhances model performance. Experimental results demonstrate that the CNN-GRU-Attention model outperforms baseline models, including CNN, GRU, LSTM (Long Short-Term Memory), and BiLSTM (Bidirectional Long Short-Term Memory), with higher prediction accuracy and fewer trainable parameters. The model's predictive capabilities enable optimization of power supply strategies in smart grids. However, it faces challenges in long-term load prediction and computational resource demands during hyperparameter optimization.