基于神经网络的基坑变形动态预测研究

Research on Dynamic Prediction of Foundation Pit Deformation Based on Neural Networks

基坑开挖过程中,需要根据实际情况及建筑安全等级进行严格的变形控制,有效的变形预测能更好地指导施工。根据某基坑工程现场监测数据,应用MATLAB 7神经网络工具箱,建立了基于BP网络的基坑变形多步预测模型A和动态预测模型B,并与广义回归网络建立的模型对比。结果表明,动态预测的精度明显高于多步预测模型;在变形数据随时间递增的情况下,BP神经网络比广义回归网络动态预测精度高,泛化能力强,平均预测误差约3.3%,能满足实际工程要求。

In the process of geotechnical excavation ,the deformation must be controlled strictly according to the actual situation and building safety grade ,and the good deformation prediction can effectively guide the construction .According to the in-site deformation monitoring data of a foundation pit ,the multi-step prediction model A and dynamic prediction model B were built based on BP network by using MATLAB 7 neural network toolbox ,then compared with the models es-tablished by generalized regression neural network （GRNN） .The results show that the dynamic prediction’s accuracy is much higher than that of the general multi-step prediction model ;when the deformation data is increased with time ,the BP network has better generalization ability and higher accuracy than GRNN , the average prediction errors are around 3 .3% ,which can meet the engineering requirements .