【摘要】 为量化道路工程中寒区低温与冻融作用对沥青混凝土开裂的影响，建立寒区沥青路面服役性能模型，为未来低温环境下的路面开裂情况提供参照。建立不同类型沥青的劲度模量预测模型，并依据规范公式修正低温开裂指数预测模型，基于国内实际公路数据对比验证其精度；基于LTPP数据库，初选3种代表算法(神经网络、随机森林及支持向量回归算法),建立低温开裂裂缝长度预测模型，对比精度优选出最佳算法。结果表明：基于国内外数据建立的劲度模量预测模型的R~2大于0.99,修正后低温开裂指数预测模型MAPE为13.17%,具有较好预测低温开裂指数的能力。基于神经网络算法所建立的预测模型为最优算法，训练集及测试集的R~2为0.997和0.947,MAPE分别为3.16%和12.96%,精度大于80%,能较好预测低温开裂长度。更多还原

【Abstract】 In order to quantify the influence of low temperature and freeze-thaw action on asphalt concrete cracking in cold area of road engineering, a service performance model of asphalt pavement in cold region was established to provide a reference for the cracking of pavement under low temperature environment in the future. The stiffness modulus prediction model of different types of asphalt was established, and the prediction model of low temperature cracking index was modified according to the specification formula, and its accuracy was verified based on the comparison of domestic actual highway data. Then, based on the LTPP database, three representative algorithms(neural network, random forest and support vector regression) were selected to establish the long-term evolution model of pavement cold crack length, and the best algorithm was optimized by comparing the accuracy. The results show that theR~2 of the stiffness modulus prediction model based on domestic and foreign data is greater than 0.99. And the MAPE of the modified prediction model of low temperature cracking index is 13.17%, which has good ability to predict low temperature cracking index. The prediction model based on the neural network algorithm is the optimal algorithm. The R~2 of the training set and the test set are 0.997 and 0.947, and the MAPE of both are 3.16% and 12.96%, respectively. The accuracy is more than 80%, which can better predict the length of low temperature cracking.更多还原