Dual-Layer Fusion Model Using Bayesian Optimization for Asphalt Pavement Condition Index Prediction.

Area of Science:

• Civil Engineering
• Data Science
• Infrastructure Monitoring

Background:

• Traditional pavement performance models struggle with temporal dynamics and multi-factor interactions.
• Integrating heterogeneous data streams (sensors, distress, maintenance) is challenging.
• Need for advanced models to capture complex pavement deterioration mechanisms.

Purpose of the Study:

• To develop an advanced pavement performance prediction model.
• To integrate heterogeneous data from IoT sensors and empirical records.
• To improve the accuracy and mechanistic understanding of pavement degradation.

Main Methods:

• Proposed a Bayesian Optimization Dual-Layer Feature Fusion Model (BO-DLFF).
• Utilized a dual-stage feature selection (BP-MIV/RF-RFECV) to identify 12 critical predictors.
• Employed Local Cascade Ensemble (LCE) and TCN-Transformer networks with a Stacking framework.

Main Results:

• Achieved a prediction accuracy of R² = 0.9292 on an 8-year dataset.
• Effectively revealed non-linear relationships between Pavement Condition Index (PCI) and multi-source features.
• Demonstrated strong correlation between strain gauge data and long-term pavement degradation.

Conclusions:

• The BO-DLFF model offers superior pavement performance prediction capabilities.
• The framework provides mechanistic insights into pavement deterioration processes.
• Advances infrastructure monitoring by synthesizing IoT sensing and empirical data.