A Study on Improvement of EV Charging Occupancy Prediction Efficiency

In this paper, we propose a hybrid model structure combining CNN-LSTM and Fully Connected to improve the prediction of electric vehicle charging station occupancy efficiently. We conducted experiments on UK and Korean datasets to evaluate the proposed model's learning efficiency and prediction accuracy compared to existing models. As a result, no loss in prediction accuracy was observed in both datasets, and our model achieved an average learning speed of approximately 1.61 times faster than the existing model on the UK dataset, and approximately 1.46 times faster than the Korean dataset. Notably, for charging stations with the largest differences, the UK dataset demonstrated a 2.04 times improvement in the view of learning speed, whereas the Korean dataset demonstrated a 1.84 times improvement. This proposed process allows for dimensionality reduction and feature extraction, ensuring only critical data is utilized in the LSTM model.