Intelligent decision-making model in preventive maintenance of asphalt pavement based on PSO-GRU neural network

Abstract

The milage of asphalt pavement growth explosively around the world in the past decades resulted in a tremendous maintenance workload. Preventive maintenance (PM) is an effective strategy in saving budget, keeping the pavement in good condition, and extending pavement life. A particle swarm optimization (PSO) algorithm enhanced gated recurrent unit (GRU) neural network is developed in this research to predict five pavement performance parameters. The model is trained based on a dataset containing seven-year distress measurement data in 100-m intervals, traffic load data, climatic records, and maintenance records of a chosen highway in China. The random forest (RF) algorithm is used to analyze the influence of the factors on pavement performances for different lanes. The result shows the PSO-GRU model could increase the prediction accuracy by 21% on average compared with traditional ANN and 17% on average compared with the AdaBoost model. The validation case study shows a significant consistency between the predicted pavement quality index and the whole-year measurement data with a 0.67 coefficient of determination. This study demonstrates the potential of using the PSO-GRU neural network to provide the most effective treatment at a given location on a highway.