为准确掌握寒区粗粒土路基冻胀变形特性，基于室内模型试验研究了初始含水率、冻结温度、细粒含量以及冻融循环次数对粗粒土路基冻胀位移和冻结速率的影响；提出了基于PSO-SVR优化的SVM支持向量机预测模型对粗粒土路基进行冻胀位移值预测。通过对单因素样本输入数据进行扰动，以输出变量值的波动情况作为各因素对路基冻胀敏感性大小的判断依据。结果表明：影响粗粒土路基冻胀位移各因素的敏感性依次为初始含水量> 细粒含量> 冻结温度> 冻融循环次数；粗粒土路基的最大冻胀位移与这4个因素呈一次线性函数关系，快速冻结期冻胀速率与4种因素呈正比关系。

In order to accurately understand the freezing and deformation characteristics of coarse-grained soil subgrade in cold regions based on indoor model tests, the effects of initial water content, freezing temperature, fine grains content, and number of freezing and thawing cycles on the freezing displacement and freezing rate were examined. Then, the SVM prediction model optimized by PSO-SVR was proposed to predict the freezing displacement value of coarse-grained soil subgrade. The sensitivity of each factor for the freezing expansion of the subgrade was determined by observing the variations of the output variable values through perturbing the input data of single-factor samples. The results indicate that the sensitivity of the studied factors influencing the freezing and expansion displacement of coarse-grained soil subgrade shows the following decreasing order: initial water content > freezing temperature > fine grains content, freezing and thawing cycles. The maximum freezing and expansion displacement of coarse-grained soil subgrade has a linear relationship with these four factors, where the freezing and expansion rate in rapid freezing period exhibit a positive proportional relationship with these four factors.