[关键词]
[摘要]
利用实验室清水条件下桥墩局部冲刷的试验数据,采用支持向量机(suppcrt vector machine,SVM)和BP(back propagation)神经网络的方法,基于量纲分析原理,对影响桥墩局部冲刷产生的相关因子进行分析。将试验数据的3/4作为预测模型的训练数据集、1/4作为预测模型的测试数据集。模型的输入因子有水流弗劳德数Fr、水深与墩径之比h/D、床沙中值粒径与墩径之比d50/D、冰盖下表面糙率与床面糙率之比ni/nb,输出因子为冲刷坑深度ds。采用相关系数(r)、均方根误差(δRMSE)、平均绝对百分比误差(δMAPE)、确定系数(R2)作为预测结果的评价指标,并将预测结果与试验结果做了比较。BP神经网络模型和SVM模型在预测明流条件下桥墩局部冲刷坑深度时,预测结果的r分别为0.89和0.88、MAPE分别为38.8%和31%;在预测冰盖条件下冲刷坑深度时,预测结果的r分别为0.78和0.73、MAPE分别为43%和46%。结果表明BP神经网络和SVM模型预测明流及冰盖条件下桥墩局部冲刷坑深度时具有较高的精度。
[Key word]
[Abstract]
The local scour of bridge piers is intensified by the existence of ice cover.Correct calculation of local scour pit depth of bridge piers under ice cover is very important for the safety design of bridge.Many experts and scholars studies the free flow under the condition of bridge pier local scour pit depth by experimental research and artificial intelligence method.The study of bridge pier local scour problem under the condition of ice compared to the free flow conditions in terms of local scour of bridge piers research started late,and the research is relatively small,so the use of artificial intelligence method is helpful to predict the laboratory bridge pier local scour pit depth. Based on the principle of dimensional analysis,the relevant factors affecting the local scour of bridge piers were analyzed by the support vector machine (SVM) and BP neural network based on the experimental data of local scour of bridge piers under the conditions of clear water scour in the laboratory.Three-quarters of the test data were taken as the training data set of the scour pit prediction model,and one-fourth as the test data set of the scour pit prediction model.When calculating the local scour pit depth of bridge pier under the condition of open flow,the input factors of the model are:flow Froude number Fr,the ratio of water depth to pier diameter h/D,the ratio of median particle size of bed sand to pier diameter d50/D.Output factor:scour pit depth ds.The local scour pit depth of bridge piers were calculated using the 65-1 and 65-2 revisions in the Chinese Highway Engineering Hydrological Survey and Design Code (2015) and the HEC-18 formula in the American Code,and the calculated results were compared with the predicted results of SVM and BP neural network model. When predicting the local scour pit depth of bridge pier under ice sheet conditions,the input factors of the model are:flow Froude number Fr,the ratio of water depth to pier diameter h/D,the ratio of median particle size of bed sand to pier diameter d50/D,the ratio of ice cover roughness to channel bed roughness ni/nb.Output factor:scour pit depth DS,and the predicted results are compared with the test results.The correlation coefficient (r),root mean square error (δRMSE),mean absolute percentage error (δMAPE),and determination coefficient (R2) was used as the evaluation indexes of the prediction results.When predicting the local scour pit depth of the bridge pier under the condition of open flow,the rof BP neural network model and SVM model are 0.89 and 0.88,and δMAPE is 38.8% and 31%,respectively.The rand δMAPEof local scour pit depth of piers are 0.83 and 0.53 cm,respectively,and 61.2% and 189%,respectively,according to Chinese code formula and American code formula.When predicting the scour pit depth under ice sheet conditions,the predicted r values are 0.78 and 0.73,and δMAPEvalues are 43% and 46%,respectively. By integrating the test data of pier local scour under the current ice sheet and open flow conditions,the depth of the pier local scour pit was predicted by BP neural network model,SVM model,Chinese code,and American code.It is found that the study of pier local scour under open flow is enlightening,and the relationship between pier local scour depth and water depth,velocity,and pier diameter under open flow and ice cover is significant,there is a nonlinear relationship between the influence factors.When the BP neural network model and SVM model were used to predict the local scour depth of bridge piers under open flow,the accuracy is generally higher than the calculation results of Chinese code and American code.BP neural network model and SVM model showed good performance in predicting the local scour pit depth of bridge piers under open flow and ice cap,and the prediction results have high accuracy,which can provide a certain reference for the safety design of bridges.
[中图分类号]
[基金项目]