Path informed adaptive trend analyzer using Hilbert Huang transform for electric vehicle driving range prediction.
Belqasem Aljafari1, Thanikanti Sudhakar Babu2, Shitharth Selvarajan3,4
1Electrical Engineering Department, College of Engineering, Najran University, Najran, 11001, Saudi Arabia. bhaljafari@nu.edu.sa.
Scientific Reports
|May 7, 2026
Summary
Accurate electric vehicle (EV) range prediction is crucial for energy management and safety. A new hybrid deep learning model, combining HART, PAITHRA, and CASH, significantly improves prediction accuracy for diverse driving conditions.
Area of Science:
- Engineering
- Computer Science
- Energy Systems
Background:
- Accurate electric vehicle (EV) range prediction is vital for energy management, battery utilization, and safe operation.
- Existing prediction methods struggle with the nonlinear, dynamic, and non-stationary nature of EV data, limiting accuracy and applicability.
- There is a need for advanced models that can capture complex driving characteristics across various scenarios and vehicle types.
Purpose of the Study:
- To propose a novel hybrid deep learning architecture for accurate EV range prediction.
- To address the limitations of existing methods in handling dynamic and non-stationary EV operational data.
- To enhance energy management and battery utilization through improved range forecasting.
Main Methods:
- Developed a hybrid deep learning framework integrating Hilbert-Huang Auto Recurrence Transform (HART) for feature extraction, Path-Informed Adaptive Inverted Trend Hub Range Analyser (PAITHRA) for predictive modeling, and Chaotic Algae Sparrow Hyper-tuner (CASH) for attention optimization.
- Utilized a combination of high-level feature extraction, adaptive sequence modeling, and attention weight optimization.
- Tested the model on the EV Energy Consumption Dataset and the Full Electric Vehicle Dataset 2024, encompassing varied driving styles, battery states, and weather conditions.
Main Results:
- Achieved a Prediction Error of 2.5, MAE of 0.8 kW, RMSE of 1.3 kW, and an R² Score of 0.991.
- Demonstrated high Validation Accuracy of 99%.
- The model converged within 35 minutes for all epochs, with an inference time of 8 ms per sample, outperforming conventional deep learning models.
Conclusions:
- The proposed hybrid deep learning model effectively addresses the challenges of EV range prediction.
- The model exhibits superior accuracy and applicability across diverse driving conditions and vehicle types compared to existing methods.
- This advancement offers significant potential for optimizing EV energy management and enhancing user experience.
