基于LSTM神经网络的沥青路面抗滑性能预测

Prediction of Anti-skid Performance of Asphalt Pavement Based on LSTM Neural Network

摘要:

为实现高速公路沥青路面抗滑性能衰减规律的研究和预测，收集了四川川南某高速公路2016—2021年路面抗滑性能指数SRI和交通量数据，分析了交通荷载与路龄对沥青路面抗滑性能影响；建立了RNN神经网络模型和LSTM神经网络模型，以历年的SRI值、累计轴载、路龄作为输入，最后1 a的SRI值作为输出，实现对高速公路SMA路面抗滑性能指数的预测，并用平均绝对百分比误差MAPE和均方根误差RMSE来对RNN神经网络和LSTM神经网络的预测精度进行评估。结果表明：LSTM神经网络训练集和训练集的平均MAPE值和平均RMSE值分别为0.296 7、0.349 9和3.914 3、3.621 6，相比于RNN神经网络，分别减少了1.002 8、1.021 4和0.441、0.201 1，LSTM模型能够对SMA路面抗滑性能指标进行有效预测。

Abstract:

In order to study and predict the attenuation law of the anti-skid performance of the asphalt pavement of the expressway, the pavement anti-skid performance index (SRI) and traffic volume data of an expressway in southern Sichuan from 2016 to 2021 are collected. The influence of traffic load and pavement age on the anti-skid performance of the asphalt pavement is analyzed. The RNN neural network model and the LSTM neural network model are established. Taking SRI values, cumulative axle loads and pavement ages over the years as inputs, and the SRI value of the last year as the output, the anti-skid performance indexes of the expressway SMA pavement are predicted. And the mean absolute percentage error (MAPE) and the root mean square error (RMSE) are used to evaluate the prediction accuracy of the RNN neural network and the LSTM neural network. The results show that the average MAPE and RMSE values of the training set and the test set of the LSTM neural network are 0.296 7, 0.349 9, 3.914 3, and 3.621 6, respectively. Compared with the RNN neural network, these values are reduced by 1.002 8, 1.021 4, 0.441, and 0.201 1, respectively, which indicates that the LSTM model can effectively predict the anti-skid performance index of the SMA pavement.