为实现对路面结冰时间的准确预测，首先根据目前已有监测点路面传感器的历史数据，抽取结冰前一段时间的数据并剔除异常数据，形成数据标本。对数据进行特征选择和归一化处理，将与路面结冰强相关的数据进行BP神经网络建模。然后，让模型对数据标本进行自学习，并自动调整模型内各节点权值，进行模型优化。最后利用其他监测点数据对模型进行验证。结果表明：模型对2h之内的结冰预测结果较准确。并且预测模型对越接近结冰点的时间预测越准确，从统计结果可看出60min之内的预测数据和实际数据几乎完全一致。因此可见，使用BP神经网络算法可实现对道路结冰时间的有效预测。

In order to accurately predict the pavement icing time, according to the historical data of the road sensors at the existing monitoring points, the data for a period of time before icing was extracted and the abnormal data was eliminated to form a data sample. After data feature selection and normalization, the processed datasets that had strong correlation with pavement icing were selected to build the BP (back propagation) neural network model. Then, the parameters of the model were optimized via self-supervised learning, and the weight value of each node in the model was adjusted automatically. Additionally, this neural network model was also verified with data from other monitoring sites. The results show that the model is more accurate for icing prediction within 2 hours. Generally speaking, the closer to the point of freezing, the more accurate the result is. The statistical result shows an almost complete match between predicted data and actual data when the freezing occurs within 60 minutes. Therefore, the proposed BP neural network model can effectively predict the time of road icing.