基于机器学习的圆柱桥墩局部冲刷深度预测方法对比研究  

Comparative Study on Machine Learning-based Prediction Methods of Local Scour Depth for Cylindrical Pier

作　　者：魏凯[1] 裘放 向琪芪 Wei Kai;Qiu Fang;Xiang Qiqi(Key Laboratory of Railway Industry on Disaster Prevention and Mitigation for Railways in Special and Complex Mountains,Southwest Jiaotong University,Sichuan,Chengdu 610031,China;Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]西南交通大学复杂艰险山区铁路防灾减灾铁路行业重点实验室,四川成都610031 [2]中国铁道科学研究院集团有限公司铁道建筑研究所,北京100081

出　　处：《铁道技术标准（中英文）》2023年第6期42-50,共9页Railway Technical Standard(Chinese & English)

基　　金：国家自然科学基金区域创新发展联合基金(U21A20154);国家自然科学基金优秀青年科学基金(52222804)。

摘　　要：准确预测桥墩局部冲刷深度对保障桥梁基础稳定具有十分重要的意义,然而常用的局部冲刷经验公式在预测桥墩局部冲刷深度的准确性和稳定性方面仍存在较多问题。随着人工智能算法的发展,机器学习在预测桥墩冲刷方面表现良好。为评估众多基础局部冲刷预测方法的可靠性与适用性,本文对比研究基于机器学习的圆柱桥墩局部冲刷深度预测方法。首先,介绍BP神经网络、RBF神经网络和支持向量机3种机器学习算法的理论基础和算法特点;其次基于已发表文献中的140组圆柱桥墩局部冲刷深度试验数据,分别构建基于上述机器学习算法的圆柱桥墩局部冲刷深度预测模型;最后应用上述构建的3种机器学习预测算法和6种常用的桥墩局部冲刷经验计算公式分别预测其余60组试验冲刷深度,并采用±25%误差线、均方误差和皮尔逊相关系数评估上述方法预测结果的准确性和稳定性。通过对比试验冲刷深度与预测冲刷深度发现,机器学习算法能够较好根据已有试验数据进行圆柱桥墩局部冲刷深度预测;相比经验公式,机器学习算法在局部冲刷深度预测方面具有更高的准确性和稳定性;BP神经网络的局部冲刷深度预测精度较高,支持向量机的局部冲刷深度预测稳定性较优。Accurate prediction of the local scour depth around the bridge piers is of great significance to ensuring the stability of bridge foundations.However,there are still many problems with the accuracy and stability of predicting the local scour depth around the piers by the commonly used empirical formula.With the development of artificial intelligence algorithms,machine learning has performed well in predicting pier scour.To evaluate the reliability and applicability of many local scour prediction methods for foundations,a comparative study of local scour depth prediction methods for cylindrical piers based on machine learning was carried out.First,the theoretical basis and algorithm characteristics of the BP neural network,RBF neural network and support vector machine(SVM) were introduced;Second,based on the 140 groups′ experimental data of local scour depth around cylindrical piers in the literature,the prediction models of local scour depth around the cylindrical pier based on the above machine learning algorithm were constructed;Finally,three machine learning prediction algorithms and six commonly used empirical formulas for local pier scour were used to predict the scour depth in the other 60 groups of tests.The accuracy and stability of the results obtained through the above prediction methods were evaluated by using the ±25% error line,mean square error,and Pearson correlation coefficient.It is found from the comparisons between test scour depth and predicted scour depth that the machine learning algorithm can well predict the local scour depth around the cylindrical pier based on the existing experimental data.The machine learning algorithm has higher accuracy and stability in local scour depth prediction than that of empirical formula.The prediction accuracy of the BP neural network is higher,and the prediction stability of SVM is better than the other algorithms.

关 键 词：圆柱桥墩 局部冲刷 冲刷预测 机器学习 经验公式 

分 类 号：TV143[水利工程—水力学及河流动力学]