Electric Vehicle Charging Load Prediction in Malaysia Using LSTM With Direct and Recursive Forecasting Methods

The rapid growth of Electric Vehicles (EVs) introduces highly variable electricity demand, posing challenges to grid stability and charging infrastructure deployment. Accurate load forecasting is therefore essential for reliable grid operation and optimal energy management. This study forecast EV charging demand using a Long Short-Term Memory (LSTM) model trained on historical data that reflects actual Malaysian charging behaviour. The dataset is preprocessed to remove outliers, normalize continuous features, and generate sequential input-output pairs suitable for LSTM training. The predictive performance is examined through Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). For the Direct Multi-Step strategy, the model achieves a normalized RMSE (NRMSE) of 56.02% and normalized MAE (NMAE) of 33.02% for a 1-day horizon, with errors increasing to 70.40% and 34.33% for NRMSE and NMAE for a 1-month horizon, respectively. The Recursive Multi-Output strategy performs best at the 1-week horizon with an NRMSE of 47.92% and NMAE of 28.22%, but shows significant error accumulation for the 1-year horizon with NRMSE 126.61%. Comparative analysis shows that the Recursive Multi-Output approach is more accurate for short and medium-term horizons, whereas the Direct Multi-Step method provides greater stability for long-term predictions. These findings demonstrate the potential of LSTM-based forecasting to support infrastructure planning and sustainable EV integration.