基于神经网络的基坑变形预测及安全控制研究  

Research on Prediction of Foundation Pit Deformation and Safety Control Based on Neural Networks

作　　者：郑知斌[1] 刘勇[1] 刘继尧[1] 赵智涛[1] 

机构地区：[1]北京市市政工程研究院,北京100037

出　　处：《铁道工程学报》2010年第9期64-68,共5页Journal of Railway Engineering Society

基　　金：国家自然科学基金资助项(2008BAJ06B01-3;2008BAJ06B03-3)

摘　　要：研究目的:基坑变形影响到自身安全及周边建(构)筑物的安全,通过神经网络分析预判基坑变形和预警方法的应用,提供一种新的风险控制措施。研究结论:神经网络可以有效地进行基坑变形预测,起到预判基坑变形稳定性的作用;本工程把时间序列作为输入层的预测结果更接近于实测值,说明把时间序列作为输入层元素更能体现出基坑围护结构变形与时间的内在非线性关系;从实际施工来看施工工序的转变,基坑围护结构的变形实测值出现突变,神经网络预测结果误差偏大,这可能与神经网络样本少而导致训练效果不佳有关。Research purposes： The foundation pit deformation threats the safety of the foundation pit and surrounding buildings and structures.Application of neural networks in prediction of the foundation pit deformation in advance and giving alarm is a new way to control the risk.Research conclusions：The neural networks can effectively predict the deformation and stability of the deep foundation pit in advance.and the stability based on which a judgement can be made on the stability of the deep excavation.The prediction value is much more closed to the measured value by using time series as an input layer.This shows the time series method can reflect the internal non-linear relation between the deformation of bracing structure of foundation pit and the time.In practice,the transfer of construction processing sequence can induce a sudden change of the measured value of the deformation.However,the big deviation of the predicted value from the mesaured value may result from ineffective training due to the small sample range of neural networks.

关 键 词：神经网络 动态预测 安全预警 安全控制 

分 类 号：TU753[建筑科学—建筑技术科学]