Application Of Improved Neural Network Algorithm In The Reliability Analysis Of Deep Excavation

With the rapid development of China’s economy,there are more and more deep excavation engineering projects,the size and the difficulty of them are also increasing.But the deep excavation may cause the support structure deformation and ground subsidence,which affect the safety of the surrounding buildings.Therefore,the monitoring and prediction of deformation caused by deep excavation are still an important part of these engineering projects.Proper prediction of deep excavation engineering projects can essentially help in setting scientific protection measures.With the analysis of the monitoring data,it can find out the possible potential dangers and formulate the countermeasures in time,which ensures the safety of the deep excavation construction.This is very important for the foundation pit construction.It has been proved that using BP neural network to build a prediction model is a feasible way.However,because the BP neural network model is unstable and easy to fall into local optimum,which affect the model’s further application in engineering projects,so the way to improve neural network model has been attracting many researchers,interests.Based on the existing research,this thesis studies the reliability forecast method of deep excavation.The particle swarm optimization algorithm(PSO)is applied into the existing BP neural network model,and the prediction results show that the proposed algorithm is more suitable for engineering application.First,this thesis gives an improved method of PSO.The example shows that this method has a stronger global search ability comparing with the existing methods.After using the improved PSO to obtain the initial parameters,this thesis then gives an improved BP neural network algorithms(MPSO-BP)which has a better performance when dealing with deep excavation prediction problems.Finally,using the method in this thesis,the practical data of the support structure deformation and supporting axial force is predicted.The prediction result shows that the improved method is more suitable for engineering application in precision and stability.Finally the main work of the thesis is summarized,and the future research ideas are provided.