[Purposes] This study aims to explore the spatial evolution patterns of driving workload for different driver groups in extra-long tunnel sections.[Method] Through a naturalistic vehicle experiment, physiological, psychological, and driving behavior data were collected from two types of drivers: those familiar and unfamiliar with the tunnel road conditions. Based on the spatial division of the tunnel section, driving workload features were extracted, and an indicator system was constructed focusing on gaze duration, heart rate, coefficient of speed variation, and maximum acceleration. Principal component analysis (PCA) and fuzzy C-means clustering (FCM) were employed to classify driving workload levels, while the LightGBM model was used for evaluation and parameter optimization.[Findings] The results show that driving workload increases significantly at the entrance and exit of the tunnel, gradually accumulates inside the tunnel, peaks at approximately 1 km, and stabilizes after about 2.5 km. Unfamiliar drivers exhibited significantly higher workload levels than familiar drivers. The constructed machine learning model achieved an identification accuracy of 93.9%, indicating that the selected indicators effectively explain variations in driving workload.[Conclusion] This study reveals the dynamic evolution of driving workload in extra-long tunnels and the differences among driver groups, providing an important theoretical basis and reference for tunnel safety assessment and the optimal design of traffic safety facilities.