Study on the Prediction Model of Building Vibration Response Induced by Super-large Diameter Shield Cutting Pile in Densely Urban Areas based on Machine Learning Algorithm
Accurate prediction of the vibration response of buildings induced by shield tunneling cutting piles can help reduce the impact of shield tunneling cutting piles on buildings and guide the safety progress of the project.In order to achieve reasonable prediction of the vi-bration response of buildings induced by shield tunneling cutting piles,based on the shield tunneling project in Haizhu Bay,Guangzhou,four neural network models based on convolutional neural network(CNN),long short-term memory neural network(LSTM),backpropaga-tion neural network(BP),and hybrid neural network(CNN-LSTM)were established to predict the vibration of buildings and compare the accuracy of each model.The factors affecting the prediction accuracy were analyzed.The results showed that the CNN-LSTM neural network had the best prediction performance,while the prediction accuracy of CNN,LSTM,and BP neural networks was similar.After comparison,it was found that using Bayesian optimization algorithm to adjust hyperparameters can effectively improve prediction accuracy.The selection of shield tunneling cutting piles method and the complex environment of on-site construction affect the vibration response of buildings,result-ing in a decrease in model prediction accuracy.In the future,indoor scale experiments can be conducted and environmental factors can be quantified to improve prediction accuracy.The research results can provide a reference for shield tunneling engineering.
NETL
NSTL
万方数据
