| Due to the relatively tight land resources,the general direction of urban construction has been developed from the ground up and down.There are high-rise structures,light rail,underground market,parking lot,underground railway,etc.These developments are almost impossible without foundation pit construction.As their scale continues to expand,the deep foundation pit also gradually increases and even presents explosive growth,so the safety of the foundation pit has aroused people’s intense concern.The excavation of deep foundation pit not only causes the deformation of the foundation pit itself,but also causes the deformation and displacement of the surrounding earth objects(houses,roads,etc.).In order to ensure the safety of foundation pit construction,it is necessary to establish the appropriate deformation prediction model for foundation pit deformation prediction.At present,people have put forward many deformation prediction model,they all have certain advantages,such as regression analysis,ARMA model,grey system analysis model,kalman filtering model,artificial neural network model and hybrid model.Based on the actual engineering situation,the data is analyzed and processed based on the wavelet de-noising of the original measurement data,combined with the GM(1,1)model and the BP neural network model.Wavelet denoising can eliminate and interpolate the abnormal value of raw data,making the original data more real.The GM(1,1)model has a good effect in the process of incomplete signal processing and small sample,and the BP neural network has good computing ability and error correction ability.So according to the advantages of both gray BP network model was established by appropriate methods,first by using GM(1,1)model to forecast,forecast it is BP neural network input samples,by measuring the original value is expected output,after learning training,get the final forecasting result.In this paper,taking advantages of the three,based on wavelet denoising gray BP neural network model,using the MATLAB tools,according to the monitoring data of experiments and analysis,the GM(1,1)model,BP neural network model,the gray BP neural network model based on wavelet denoising and the gray BP neural network model to forecast the result contrast,It is concluded that the average relative error of the predicted data after wavelet de-noising is 4.67%,which is 1.9% higher than that before wavelet de-noising.On this basis,several other deformation monitoring points are selected for further verification,verify the gray BP neural network based on wavelet denoising model to predict deformation when dealing with noise signal has a good reliability and applicability. |
