Enhanced probabilistic prediction of pavement deterioration using Bayesian neural networks and cuckoo search optimization

Abstract

Abstract The predictive performance of probabilistic pavement condition deterioration is critical for effective maintenance and rehabilitation decisions. Currently, numerous improved models exist, but few rely on probabilistic models to improve pavement deterioration prediction. Therefore, this study proposed an improved probabilistic model for pavement deterioration prediction based on the coupling of Bayesian neural network (BNN) and cuckoo search (CS) algorithm. The model prediction performance is evaluated against two metrics: determination coefficient (R2) and standard deviation (stability). Finally, based on the data from the pavement management system in Shanxi Province, it was verified that the CS-BNN model outperforms the genetic algorithm-BNN, particle swarm optimization-BNN, and BNN models in terms of the two metrics. Sensitivity analysis further confirms the robustness of the CS-BNN model. The findings indicate that the CS-BNN model provides more reliable predictions with lower uncertainty, aiding road engineers in optimizing maintenance schedules and costs.