Modeling and mechanistic analysis of drivers’ psychophysiological perception at tunnel threshold zones using a super-ensemble framework and spatiotemporal heterogeneity theory

Abstract The abrupt luminance transition in tunnel threshold zones induces psychophysiological stress, raising operational risks. Using multi-source field data from four highway tunnels, this study proposes a two-layer framework: a super-ensemble model (SEM) for high-accuracy prediction of drivers’ psychophysiological responses, and a spatiotemporal heterogeneity analysis for mechanism interpretation. SEM integrates tree-based learners with a multi-dimensional evaluation and adaptive reward-penalty weighting. Results show that luminance change rate dominates psychophysiological perception, exhibiting significant lagged and nonlinear effects. Compared with the best baseline model (GBM), SEM achieves RMSE reductions of approximately 8.5%-10.9% (p < 0.05, Diebold–Mariano test). Lagged effects are primarily observed at lags 1–3 sampling periods. Extreme scotopic transitions (luminance change rate absolute value exceeding approximately 6 cd m −2 s −1 ) trigger threshold-like regime shifts and long-term adaptation, with response dynamics varying across tunnels. This “predict–explain” scheme supports human-centered tunnel lighting control and risk-aware operations.