A novel method for dry density forecasting of high-speed railway graded aggregate fillers based on the PSO-ANN model

Abstract

To address the lack of intelligent prediction research on dry density, this paper proposes a machine learning-based prediction method for compaction quality of high-speed railway graded aggregate (HRGA) fillers. Firstly, to reveal the main control characteristics between HRGA fillers performance and dry density, 80 sets of vibration compaction tests were carried out. The Grey Relation Analysis (GRA) algorithm was used to quantify the degree of influence between the properties of particle gradation, particle shape and particle fragmentation, and dry density. Then, a stable database of dry density of HRGA fillers was established through 300 lab vibratory compaction tests. Finally, three Hybrid Machine Learning (ML) models (PSO-ANN, PSO-SVR, and PSO-RF) was used to fit the nonlinear relationship between the mian control characteristics and dry density. The results show that the gradation parameter b, maximum particle size dmax, and gradation parameter m are strongly correlated features, with a mean value of correlation reaching 0.77. These three features are considered the main control factors affecting the dry density and are used as input features for the subsequent ML prediction model. Moreover, the PSO-ANN model is shown to have the higher prediction accuracy with R2 = 0.9634 in the test set and lower uncertainty with U95= 0.1413, Tstat= 0.533, and SI = 0.2306 in the test set for dry density predictions. The finding can provide a significant reference for intelligent compaction.