Deep-Learning-Based Probabilistic Forecasting of Electric Vehicle Charging Load With a Novel Queuing Model

Area of Science:

• Electrical Engineering
• Computer Science
• Transportation Systems

Background:

• Electric vehicle (EV) adoption and fast charging necessitate accurate EV charging station (CS) load forecasting.
• Forecasting EV charging load is complex due to non-stationary traffic flow (TF) and unpredictable charging behaviors.

Purpose of the Study:

• To develop a robust method for forecasting EV charging load by addressing uncertainties in traffic flow and arrival rates.
• To create a probabilistic queuing model that integrates TF predictions, arrival rates, and driver behaviors for accurate load forecasting.

Main Methods:

• Traffic flow (TF) prediction using a deep-learning-based convolutional neural network (CNN) with uncertainty evaluation for prediction intervals (PIs).
• Calculation of EV arrival rates using historical data and a proposed mixture model.
• Development of a novel probabilistic queuing model to convert TF to charging load, considering service limitations and driver behaviors.

Main Results:

• The proposed models effectively learned the uncertainties associated with EV charging load forecasting.
• Accurate prediction intervals for traffic flow were formulated.
• The probabilistic queuing model successfully converted TF to charging load, demonstrating comprehensive uncertainty learning.

Conclusions:

• The integrated approach provides a significant advancement in understanding and managing EV charging load uncertainties.
• The methodology shows strong potential for practical applications in EV charging station planning and operation.
• Accurate EV load forecasting is crucial for grid stability and efficient CS management.