Abstract:

To investigate the variation patterns of asphalt pavement temperature in complex urban road environments, this study conducted wintertime (from November 2024 to February 2025) observations on six representative road sections across three typical road types in the urban area of Tianjin: overpasses, arterial roads, and cross-river bridges. The heterogeneous characteristics of asphalt pavement temperature and the underlying formation mechanisms for these three road types were analyzed. Based on enhanced feature variables, Lasso regression, polynomial regression, and support vector regression models were established for asphalt pavement temperature prediction. The findings reveal significant differences in the asphalt pavement temperature characteristics of the three road types, attributable to variations in their thermal properties of underlying surfaces and local energy exchange processes. Cross-river bridges exhibit the largest diurnal range, the highest rate of temperature change, and the most extreme temperatures due to their low thermal mass structures and efficient heat exchange between the bridge and air. Overpasses, characterized by massive structures with high thermal inertia, show the most gradual temperature changes. Arterial roads, regulated by heat storage in the soil subgrade and anthropogenic heat sources, display moderate diurnal variation and temperature change rates, with higher nighttime minimum temperatures than the bridges. In terms of modeling, integrating periodic and time-lag features effectively improves model prediction accuracy compared to using only original meteorological features. For the Lasso model, the RMSE is reduced by approximately 18% on average, ranging from about 1.04 to 1.61 ℃, with MAE ranging from about 0.74 to 1.33 ℃. The Lasso, polynomial, and support vector machine models all capture the variation patterns of asphalt pavement temperature well. Among them, the Lasso model maintains high prediction accuracy while offering greater interpretability and computational efficiency, thus presenting higher application value. Error analysis shows that the prediction error of the Lasso model is higher during the daytime than at night, indicating its limited ability to characterize high-frequency abrupt processes strongly influenced by shortwave radiation. Prediction errors for temperature extremes on cross-river bridges are relatively larger, exhibiting an obvious "peak suppression and trough filling" phenomenon. This is related to the structural thermal characteristics and the model training mechanism aimed at minimizing overall error. This study clarifies the heterogeneity of asphalt pavement temperatures in complex urban roads and its physical causes, providing a scientific basis for differentiated road-icing warnings and precise management.

Key words: asphalt pavement temperature, roads types, temperature heterogeneity, prediction models

摘要：

为揭示城市复杂道路环境下的路面温度变化规律，针对天津城区立交桥、主干道、跨河桥三种典型道路6个代表性路段，开展冬季（2024年11月至2025年2月）路面温度观测，厘清了三种道路路面温度异质性特征与形成机理，基于增强特征集构建了Lasso、多项式和支持向量机路面温度预报模型。结果表明：受下垫面热力性质和局地微环境条件影响，立交桥、主干道、跨河桥路面温度特征差异显著，跨河桥因低热容结构及强通风散热条件，呈现最大的温度日较差、温度变率，温度极端性最强；立交桥结构热惯性大，温度变化最平缓；主干道受土壤路基储热与人为热的共同调节，夜间最低温度高于桥梁路面。模型构建结果显示，相比仅采用原始气象特征，融合周期特征与时滞特征可显著提升模型预报精度，采用增强特征的Lasso模型RMSE平均降低约18%，RMSE为1.04~1.61 ℃，MAE为0.74~1.33 ℃。Lasso、多项式和支持向量机模型均能有效刻画路面温度演变规律，而Lasso模型在保持预报精度的同时，兼具更优的可解释性与计算效率，应用价值更为突出。误差分析表明，Lasso模型白天预报误差高于夜间，表明模型对短波辐射驱动下的温度高频突变过程刻画能力不足；跨河桥温度极值预报误差相对更大，存在明显的“削峰填谷”现象，其成因与桥体热力特性以及模型以全局误差最小化为目标的训练机制有关。本研究明确了复杂城市道路路面温度的异质性规律及其物理成因，可为城市道路结冰风险的差异化预警与精准化管控提供科学依据。

关键词: 路面温度, 道路类型, 温度异质性, 预报模型