Probabilistic Forecast of EV Charging Demand using Quantile Regression and LSTM with Attention Mechanism

The electrification of transportation has significantly increased electric vehicle (EV) charging demand on energy systems. Accurately capturing the uncertainty of future EV loads is essential for flexible system operation. This study proposes an LSTM-attention model for probabilistic EV load forecasting, featuring an encoder-decoder architecture with an intermediate attention layer. Using Quantile Regression (QR), the model predicts upper, median, and lower load quantiles. Evaluation is performed on parking station data from the SmoothEMS Met GridShield project in the Netherlands.