Path informed adaptive trend analyzer using Hilbert Huang transform for electric vehicle driving range prediction

The adequate forecast of the electric vehicle (EV) range is the key to the success of the improved energy management and the improved battery utilization, not to mention ensuring the safe operation of this product in the actual conditions of driving in the real life. Despite the significant advancements, the existing methods tend to be unable to possess the nonlinear, dynamic, and non-stationary characteristics of EV operation and driving data, leading to a low prediction accuracy and applicability to a large range of vehicle classes and driving scenarios. We propose a novel hybrid deep learning architecture that will address all these challenges in this paper, the combination of high-level feature extraction, adaptive sequence modeling and attention weight optimization. The framework consists of three basic components, i.e., Hilbert-Huang Auto Recurrence Transform (HART) of rich temporal and recurrent features extractions, Path-Informed Adaptive Inverted Trend Hub Range Analyser (PAITHRA) of a combination of predictive model using pathformer and adaptive routing of features, and Chaotic Algae Sparrow Hyper-tuner (CASH) of a dynamic attention weight optimizer. An extensive testing of the EV Energy Consumption Dataset and the Full Electric Vehicle Dataset 2024 containing different driving styles, battery condition and weather conditions is conducted. The Proposed model is characterized by a Prediction Error (2.5), MAE (0.8 kW), RMSE (1.3 kW), Validation Accuracy (99), R 2 Score (0.991), and it required 35 min totally for all epochs up to convergence and the inference time was just 8 ms per sample, being superior to the conventional deep learning models.