A hybrid iTransformer–KAN framework for high-precision tire lateral force estimation using intelligent tire signals

Tire lateral force is a critical state for vehicle dynamics control, yet conventional sensor-based methods suffer from limited accuracy due to nonlinear tire characteristics and signal noise contamination. This paper presents a novel iTransformer-KAN architecture that integrates advanced signal preprocessing with hybrid neural networks for enhanced force estimation. The approach employs a three-stage denoising method combining Improved Complete Ensemble Empirical Mode Decomposition (ICEEMDAN), Permutation Entropy (PE), and Wavelet Threshold Denoising (WTD) to extract clean features from three-axis acceleration signals, followed by an innovative iTransformer-KAN network where Kolmogorov-Arnold Networks replace traditional MLPs to capture complex nonlinear tire-road interactions through spline-based mappings. Experimental validation demonstrates that the proposed method achieves superior performance with an NRMSE of 4.75%, representing a 12.5% improvement over GRU networks and a 21.2% improvement over MLP approaches, while maintaining computational efficiency with a 77.82-s training time. The method enables real-time lateral force monitoring for vehicle dynamics control systems, advancing the development of intelligent tire technologies for enhanced automotive safety and performance.