Study on the asphalt pavement quality index evaluation using a dual probability-enhanced relative entropy information fusion method

• Proposes a dual-probability-enhanced relative entropy information fusion model for asphalt pavement quality index (PQI) evaluation. • Integrates truncated normal and Weibull distributions within a Kalman filtering framework to dynamically capture PQI attenuation. • Combines Newton-Raphson method and sequential least squares programming (SLSQP) for global optimization of distribution parameters and regression weights. • Reduces relative error range by 62.5–82.5% compared to single Kalman filter or least squares methods. • Provides tailored PQI evaluations: truncated normal suits ordinary sections; Weibull fits heavy-load sections with rapid decay. • Delivers interpretable, region-specific indicator weights to support targeted maintenance decision-making. Asphalt pavement quality is core to maintenance decision-making in pavement engineering, offering quantifiable parameters that shift decisions from experiential to quantitative. Existing methods rely on linear weighted summation of indicators, limited by fixed weights and neglect of nonlinear effects, failing to reflect complex engineering needs. This study proposes a dual-probability correction evaluation method for asphalt pavement condition based on relative entropy information fusion. First, the model employs a dual-path probability approach using truncated normal distribution and Weibull distribution, dynamically integrating the long-term impact of maintenance interval time on PQI attenuation through Kalman filtering. Second, it couples Newton-Raphson for maximum likelihood estimation of parameters in truncated normal distribution and Weibull distribution. Based on the integration of relative entropy and least squares method, the sequential least squares programming optimization algorithm is adopted to achieve global optimization of the dual objective function combining relative entropy and least squares. Applied to China's eastern coastal road network, results demonstrate robust performance: relative error range [-1.5%, 1.5%] reduced by 62.5–82.5% vs. single Kalman [-4%, 4%] or least squares [-8%, 8%], and 47% vs. random forest [-3%, 3%]. Truncated normal fusion boosts symmetry/centrality (skewness 0.05 vs. 0.3; variance 5 vs. 15) for ordinary sections, while Weibull captures rapid decay/low-value risks (skewness 0.5 vs. 0.3; P<threshold 0.55 vs. 0.20; kurtosis 5 vs. 3.5) for heavy-load sections. The fusion outperforms alternatives (11–59% error reduction), balancing dual distributions in heterogeneous scenarios, yielding targeted weights and accurate assessments to aid indicator prioritization and maintenance effectiveness.