Improved density prediction model based on global optimization algorithm for GPR system

Density prediction models are key to pavement compaction and density estimation using ground penetrating radar (GPR). However, current models do not account for the effects of dielectric constant errors and rely on limited data for parameter fitting. This article introduces an electromagnetic (EM) mixing theory and genetic algorithm (GA) based density prediction model named EM-GA. A comprehensive density database is established from literature, laboratory and field tests. After comparing Bayesian optimization, GA, simulated annealing (SA), and particle swarm optimization (PSO), the GA, which performs the best, is selected. On the testing set, the EM-GA model outperforms traditional models, with a notable reduction of 43.6 % in Mean Absolute Percentage Error (MAPE) and 39.8 % in Root Mean Square Error (RMSE). More importantly, the EM-GA predictions are accurate in the validation of four continuous survey lines. The proposed EM-GA model effectively enhances the accuracy of GPR-based density estimation for newly constructed asphalt pavement.