Deep Learning Calculation and Application of Axle Loads in Highway Sensor Data

Abstract

Axle load data and traffic survey data are both important outputs of highway sensors. This study targets highways and ordinary national and provincial highways, seeking to calculate axle load spectrum and equivalent axle times across the network. There is often an association in the spatial extent of traffic survey data and axle load detection data in highway networks. Initially, using the Highway Asphalt Pavement Design Specification, it analyzes the demand for these calculations in road sections. Considering the current axle load detection coverage, a method supported by highway traffic data is proposed. For integrating multi-source data, a generalized regression neural network model is established, enabling deep learning calculations. The method is validated and applied to Xuzhou’s highway network. Results show consistency between the calculated average axle load spectrum and actual data. Among validation samples, 3-axle vehicles exhibit the smallest deviation, while 6-axle vehicles show the largest. Calculating equivalent axle numbers reveals the distribution and grading of heavily loaded road sections, aiding maintenance decisions.