An adaptive LASSO-GARCH framework for robust EV parking occupancy forecasting and uncertainty management

Area of Science:

• * Artificial Intelligence and Machine Learning
• * Transportation Systems Engineering
• * Energy Infrastructure Management

Background:

• * Accurate prediction of electric vehicle (EV) parking occupancy is crucial for efficient charging infrastructure management.
• * Existing methods struggle with data scarcity during initial operational phases (cold start) and dynamic demand fluctuations.
• * Dynamic volatility modeling is needed to capture real-time changes in parking lot usage.

Purpose of the Study:

• * To develop a hybrid predictive framework for electric vehicle (EV) parking lot occupancy capacity.
• * To address the cold start problem using historical data from similar days.
• * To dynamically track parking demand volatility using a GARCH model.

Main Methods:

• * A hybrid framework combining adaptive regression (LASSO) and dynamic volatility modeling (GARCH).
• * Input dimension reduction via similarity-based neighbor extraction.
• * A two-stage prediction mechanism: LASSO for cold start, GARCH for ongoing volatility, with a physical processing step for valid outputs.

Main Results:

• * Achieved a Root Mean Square Error (RMSE) of 2.71 EVs, outperforming ARIMA, LSTM, and GRU models.
• * 50% of predictions had zero absolute error; 95% had errors below 6 EVs during working hours.
• * The integrated LASSO-GARCH model optimized uncertainty bounds (reduced MPIW, maintained 95.5% PICP).

Conclusions:

• * The proposed LASSO-GARCH hybrid model significantly enhances the accuracy and stability of EV parking occupancy predictions.
• * The synergistic integration of LASSO and GARCH components is essential for optimal performance.
• * The framework provides reliable predictions and robust uncertainty quantification for EV charging management.