Year: 2025
Type: conference-paper
Abstract: To enhance the precision of electric vehicle charging load forecasts, this study introduces a novel approach known as the CNN-BiLSTM-Attention model. The initial step involves creating an input feature matrix that encompasses historical electric vehicle charging loads, temperature data, and information on weekdays. This matrix serves as the foundation for applying a bidirectional long short-term memory (BiLSTM) network, which is adept at uncovering the temporal relationships within the feature data. To refine the predictive capabilities further, an attention mechanism is incorporated, allowing for the assignment of varying weights to the output states of the BiLSTM. This integration results in the development of the CNN-BiLSTM-Attention model. Subsequent experimentation employs the Dundee dataset, comparing the proposed model against three established models: BiLSTM, CNN-BiLSTM, and CNN-LSTM-Attention. The findings from these experiments demonstrate that the proposed method achieves the highest level of prediction accuracy. It exhibits a more focused error distribution and superior prediction stability, alongside a pronounced linear correlation between the predicted and actual values.
Author: Xunxing Liu
Institution: Xuancheng Vocational and Technical College
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4413985835