基于长期监测的大厚度柔性基层沥青路面温度作用下应变响应规律及智能预测模型研究

Strain Response Characteristics and Intelligent Prediction Modeling of Thick Flexible-Base Asphalt Pavement under Temperature Action Based on Long-Term Monitoring

摘要: 为揭示大厚度柔性基层沥青路面在温度作用下的应变响应规律，依托广西田新高速公路试验段长期监测数据，分析路面内部温度场、极值应变及平均横向应变预测性能。对监测数据进行缺失值补全、异常值剔除和平滑处理，并划分为0~10℃、10~20℃、20~30℃、30~45℃和45~60℃五类温度工况；采用温度波动幅值和衰减系数分析温度场的深度效应，选取典型层位研究温度–应变关系，同时引入多元二次多项式回归(QPR)作为统计基准，与CNN、PSO-CNN和PSO-BP模型进行比较。结果表明：路面内部温度波动随深度增加显著衰减并存在时间滞后，浅层沥青面层的应变温度敏感性明显高于深层基层；A1层最大横向应变与温度呈显著非线性关系，而B2层表现为近似线性弱响应。PSO-CNN在0~30℃范围内测试集R²为0.8410~0.8663，具有较好的综合预测性能；PSO-BP在30~45℃区间表现出局部优势，QPR在部分高温区间具有较好的趋势拟合能力，表明不同模型的适用性存在温区差异。研究结果可为大厚度柔性基层沥青路面的服役状态评估和智能养护提供参考。

Abstract: To investigate the temperature-induced strain response of thick flexible-base asphalt pavements, long-term monitoring data collected from a test section of the Tianlin-Xilin Expressway in Guangxi, China, were used to analyze the internal temperature field, extreme strain response, and prediction performance for average transverse strain. After missing-value imputation, outlier removal, and smoothing, the monitoring data were divided into five temperature ranges: 0~10˚C, 10~20˚C, 20~30˚C, 30~45˚C, and 45~60˚C. Temperature fluctuation amplitude and attenuation coefficients were adopted to quantify the depth-dependent characteristics of the pavement temperature field, and representative structural layers were selected to investigate the temperature-strain relationship. A multivariate quadratic polynomial regression (QPR) model was introduced as a statistical benchmark and compared with convolutional neural network (CNN), particle swarm optimization-based CNN (PSO-CNN), and particle swarm optimization-based backpropagation neural network (PSO-BP) models. The results show that temperature fluctuations within the pavement structure attenuate significantly with increasing depth and exhibit an evident time lag. The strain response of the shallow asphalt layers is more sensitive to temperature than that of the deep base layers. The maximum transverse strain in the A1 layer exhibits a pronounced nonlinear relationship with temperature, whereas the B2 layer shows an approximately linear and relatively weak response. Within the range of 0~30˚C, the PSO-CNN model achieves test-set R² values of 0.8410~0.8663 and demonstrates favorable overall prediction performance. The PSO-BP model shows a local advantage in the range of 30~45˚C, while the QPR model provides satisfactory trend fitting in some high-temperature ranges, indicating that model applicability varies with temperature conditions. These findings provide a reference for service-condition assessment and intelligent maintenance of thick flexible-base asphalt pavements.