Study On Monitoring And Prediction Of Deep Foundation Pit Excavation In Central Business District Of Lanzhou City

Due to the complicated geological conditions,construction technology and construction site conditions,there are various accidents often occur in deep foundation pit construction,resulting in a certain degree of economic loss and negative social impact.In order to avoid a series of foundation pit safety accidents,the deformation monitoring of deep foundation pit must be carried out in the whole process of its construction,and the establishment of appropriate prediction model for foundation pit deformation prediction has important guiding significance for the safety construction of foundation pit.This paper based on the monitoring project of provincial government station and central business district of the first phase of the Lanzhou Rail Transit Line 1,and the monitoring data collected on the spot,established the time series model and NAR(Nonlinear Auto Regressive models)artificial neural network model according to the actual monitoring data,and the deformation trend of foundation pit is predicted in the future.Since the data acquisition process is subject to many external interference factors,Kalman filter is introduced to filter the original time series.On the basis of combining the advantages of the two models,a time series-NAR artificial neural network model based on Kalman filter is proposed.The model is applied to predict and analyze the actual data.Finally,the prediction results of the three models are compared and analyzed.The main contents of this paper are as follows:(1)Through monitoring the foundation pit monitoring project of the provincial government station and the central business district of the first phase of Lanzhou Rail Transit Line 1,the actual monitoring data of each monitoring project are analyzed.It is found that there are no monitoring alarm for each monitoring project during the monitoring period,and the foundation pit excavation process is basically in a safe and stable state.(2)Through IDW(Inverse Distance Weighted)interpolation processing for multi-stage settlement monitoring data of foundation pit retaining piles,extraction and analysis of settlement slope of foundation pit support piles based on R software platform using least squares method,it is determined that the maximum settlement deformation trend of foundation pit supporting pile is near ZJ52 monitoring point,and the later prediction and analysis will focus on the monitoring data of this point.(3)By comparing the prediction results of the three models,it can be seen that the prediction accuracy of the ARIMA-NAR combination model based on Kalman filter is the highest.The accuracy indexes MAE,RMSE,MAPE and R~2 are 0.2787,0.5279,3.9150 and 0.99972,respectively,which are better than those of the single ARIMA and NAR neural network models.The prediction accuracy of the NAR artificial neural network model is second,and the time series ARIMA model has the lowest prediction accuracy.(4)The practicability of the proposed time series-NAR combined model based on Kalman filter in the field of foundation pit monitoring of prediction is verified,and it is applied to the monitoring points such as the horizontal displacement and the depth(oblique)displacement of the foundation pit support pile for deformation prediction and analysis.According to the prediction results,both the time series model and the NAR artificial neural network model are able to obtain accurate prediction results.Based on the Kalman filter time series-NAR artificial neural network combination model,the results obtained by Kalman filter denoising have higher precision and more applicability,which can provide more reliable data support for the safe construction of foundation pits.