基于机器学习的改性沥青黏弹性预测

Viscoelastic Prediction of Modified Asphalt Based on Machine Learning

为利用机器学习方法预测改性沥青黏弹性,分析不同预测模型适用性和预测精度,测试了胶粉(CR)改性沥青、SBS改性沥青和废旧塑料(PE)改性沥青不同温度和频率下的复数模量.选择人工神经网络(ANN)、稳健线性回归(RLR)、线性支持向量回归(LSVR)、决策树回归(DTR)、高斯回归(GPR)和集成回归(ER)6种机器学习方法预测三种改性沥青复数模量.结果表明:预测结果散点图中,ANN和ER模型预测精度最高,DTR模型次之且存在数据聚类.6种预测模型预测结果的相关系数均大于0.9,纳什效率系数均大于0.85.不同预测模型在PE改性沥青中预测精度最高,SBS改性沥青次之.根据三种改性沥青复数模量预测结果的相关系数、相对均方根误差、分散指数、相对误差和纳什效率系数五个统计参数的平均值,6种预测模型预测精度从高到低依次为ER、ANN、DTR、GPR、LSVR和RLR.

In order to predict the viscoelasticity of modified asphalt by using machine learning method and analyze the applicability and prediction accuracy of different prediction models,the complex moduli of crumb rubber(CR)modified asphalt,SBS modified asphalt and waste plastic(PE)modified asphalt were tested at different temperatures and frequencies.Six machine learning methods including artificial neural network(ANN),robust linear regression(RLR),linear support vector regression(LSVR),decision tree regression(DTR),Gaussian process regression(GPR),ensemble regression(ER)were selected to predict the complex moduli of three kinds of modified asphalts.The results show that ANN and ER models have the highest prediction accuracy in the scatter plot of prediction results,followed by DTR model which has data clustering.The correlation coefficients of the prediction results of the six prediction models are greater than 0.9,and the Nash-Sutcliffe efficiency coefficients are greater than 0.85.Different prediction models have the highest prediction accuracy in PE modified asphalt,followed by SBS modified asphalt.According to the average values of the five statistical parameters including the correlation coefficient,relative root mean square error,scatter index,relative error and Nash-Sutcliffe efficiency coefficient of the prediction results of the complex moduli of the three modified asphalts,the prediction accuracy of the six prediction models from high to low is ER,ANN,DTR,GPR,LSVR and RLR.