基于MTLBO和ELM的沥青路面抗滑性能预测

Prediction of Skid Resistance of Asphalt Pavement Based on MTLBO and ELM

针对沥青路面抗滑性能预测在精度和可靠性方面的不足,本文提出了基于改进教学优化算法(Modified Teaching-Learning-Based Optimization,MTLBO)和极限学习机(Extreme Learning Machine,ELM)的沥青路面抗滑性能预测模型.首先,利用MMLS3载荷模拟器对6块路面试样进行模拟加载,达到每个加载周期后,利用高精度激光扫描仪和摆式仪分别对路面试样的宏观纹理和抗滑性能表征指标摆值(British Pendulum Number,BPN)进行采集,提取与抗滑性能相关的纹理指标.其次,鉴于ELM训练过程时间短且具有良好的模型泛化能力,本文利用ELM预测沥青路面BPN的变化趋势.为了提高ELM的全局搜索能力和预测结果的准确性,引入MTLBO对ELM输入层和隐藏层之间的权重(ω_(i))和偏置(b_(i))进行优化,从而更好地反映路面BPN与纹理指标之间的非线性映射关系.最后,以均方根误差(Root Mean Square Error,RMSE)和预测样本复相关系数(R^(2))作为评价指标,与现有的支持向量机(Support Vector Machines,SVM)、ELM、基于教学优化的极限学习机(TLBO-ELM)3种方法进行比较,本文所提出的基于改进教学优化的极限学习机(MTLBO-ELM)预测结果的均方根误差分别降低了1.80%,1.50%,1.07%.

To address the shortcomings of asphalt pavement skid resistance prediction in terms of accuracy and reliability,this paper proposed an asphalt pavement skid resistance prediction model based on Modified Teaching and Learning Optimization(MTLBO)and Extreme Learning Machine(ELM).Firstly,the MMLS3 load simulator was used to simulate the loading of six pavement specimens.After each loading cycle was reached,the macroscopic texture and the British Pendulum Number(BPN)of the pavement specimens were collected using a high-precision laser scanner and a pendulum meter respectively,and then the texture indicators related to the skid resistance were extracted.Secondly,given the short training process of ELM and its good model generalization capability,this paper used ELM to predict the change trend of asphalt pavement BPN.In order to improve the global search capability of ELM and the accuracy of the prediction results,MTLBO was introduced to optimize the weights(ω_(i))and biases(b_(i))between the input and hidden layers of ELM.It was aimed to construct a non-linear mapping relationship between the pavement BPN and the texture index.Finally,root mean square error(RMSE)and predicted sample complex correlation coefficient(R^(2))were used as evaluation indicators.Compared with existing Support Vector Machines(SVM),ELM and Extreme Learning Machines based on teaching and learning optimization(TLBO-ELM),the root mean square error of the prediction results of the proposed Extreme Learning Machine based on Modified Teaching Learning Optimization(MTLBO-ELM)is reduced by 1.80%,1.50%,and 1.07%respectively.