Entropy‐Weighted Multi‐Factor Grey‐Markov Model for Pavement Performance Prediction in Subtropical Hilly Regions

ABSTRACT

Asphalt pavements in subtropical hilly regions are subjected to the long‐term coupled effects of high temperature, high humidity, intense rainfall, and traffic loading, while their performance datasets are often short and fluctuate considerably. To address the limitations of traditional Grey‐Markov models, including insufficient multi‐factor representation, subjective residual‐state partitioning, and inadequate use of multi‐step transition information, this study develops an entropy‐weighted multi‐factor Grey‐Markov model for pavement condition index (PCI) prediction. First, a GM(1,1) model is established using the PCI series from 2010 to 2018 to extract the overall deterioration trend of pavement performance. Second, five climate–traffic loading indicators are integrated through the entropy weight method to construct a composite pressure index for systematic residual correction. Finally, the remaining stochastic residuals are softly classified using the central‐point triangular whitenization weight function and corrected using weighted multi‐step Markov transition probabilities. The memory decay coefficient is further optimized by minimizing the mean absolute percentage error (MAPE) of rolling one‐step validation from 2017 to 2019. Using PCI data from a road section in Hunan Province from 2010 to 2019, the results show that the proposed model achieves the highest accuracy when the memory decay coefficient is 0.3. The corresponding one‐step, two‐step, and three‐step transition weights are 0.7194, 0.2158, and 0.0647, respectively. Under this setting, the relative error of the 2019 PCI prediction is 0.4875%, and the rolling‐validation MAPE is 0.5839%. Compared with the traditional Grey‐Markov model, the relative error and MAPE are reduced by 24.92% and 61.48%, respectively; compared with the equal‐weight multi‐step scheme, these two metrics are reduced by 18.43% and 18.20%, respectively. The results indicate that combining systematic residual correction with multi‐step Markov correction improves the accuracy and rolling‐validation stability of small‐sample pavement performance prediction.