Deep learning predicts real-world electric vehicle direct current charging profiles and durations

Area of Science:

• Electrical Engineering
• Computer Science
• Sustainable Energy

Background:

• Optimizing electric vehicle (EV) infrastructure requires accurate charging predictions.
• Direct current (DC) fast charging behavior is complex and influenced by numerous factors.
• Existing prediction methods may lack accuracy or real-time adaptability.

Purpose of the Study:

• To develop a deep learning framework for predicting EV charging profiles and durations.
• To enable real-time prediction updates using incremental data.
• To provide uncertainty estimates for the predictions.

Main Methods:

• A deep learning model was trained on 909,135 real-world EV charging sessions.
• The model predicts charging behavior from minimal input, refining predictions with new data.
• Generalization across diverse vehicle types and charging scenarios was tested.

Main Results:

• The model achieved 90% accuracy in predicting charging duration from a single data point.
• With six data points within five minutes, accuracy reached 95% with sub-minute error.
• The framework demonstrated generalization across various EV models and charging conditions.

Conclusions:

• Readily available charging data can enable highly accurate predictions of EV charging behavior.
• The proposed deep learning framework offers a practical and scalable solution.
• This approach supports efficient EV infrastructure planning, deployment, and reliability.