【摘要】 以粗集料的力学指标、分形维数以及BPN作为研究对象进行数据采集，基于前馈神经网络算法与支持向量机算法，提出了一种新型路面抗滑性能预测模型.采用加速加载试验，记录沥青混合料在轮载作用下表面抗滑性能的衰减过程，并利用指数模型对轮载次数与BPN的数值关系进行拟合.通过灰色关联度分析和相关性分析，评估各项因素对沥青路面抗滑性的影响程度.基于主成分分析结果，设计了6种方案，用于前馈神经网络算法和支持向量机算法的训练、验证和测试.结果表明，不同类型集料呈现出不同的抗滑性能，石英砂岩最佳，玄武岩次之，石灰岩较差.磨光值与沥青抗滑性能衰减模型的关联性最高.前馈神经网络算法构建的模型表现更为稳定，R~2值约为0.8,展现出良好的预测能力.更多还原

【Abstract】 Through comprehensive data collection, along with the coarse aggregate mechanical index, fractal dimension, and British pendulum number(BPN), a pavement friction prediction model was proposed on the basis of backpropagation neural networks(BPNNs) and support vector machine(SVM). An accelerated attenuation test was conducted to examine the antiskid performance of the asphalt mixture and aggregates at different wearing cycles. Subsequently, BPN was fitted using an exponential model. Gray relational and correlation analyses were performed to evaluate the factors influencing pavement skid resistance. According to the principal component analysis results, six schemes were prepared for the training, validation, and testing of BPNN and SVM algorithms. Test results indicate that different aggregates exhibit different antiskid properties. Quartz sandstone is the most suitable, followed by basalt and limestone. The polished stone value has the highest correlation with the attenuation model of asphalt antiskid performance. BPNN is more stable, with an R~2 value of approximately 0.8.更多还原