Multi-Factor Analysis Of Safety Situation Of Environment Around Deep Foundation Pit Based On Neural Network

With the continuous advancement of urbanization,the growing depth and scale of foundation pit engineering have a huge impact on the adjacent environment.It is often difficult for engineers to cope with complex foundation pit projects or projects with strict requirements for surrounding environment control only based on previous engineering experience and quantitative analysis.Therefore,it is of great significance to guide construction with the dynamic data provided by field monitoring.Based on the Hongqiao Road plot of Xujiahui Center in Shanghai,the pretreatment method of monitoring data is studied,and the future trend of the adjacent environment is predicted by neural network technology.Based on the predicted data,the safety situation of environment around deep foundation pit is analyzed from the perspective of multiple factors.The main research contents and conclusions are summarized as follows:i.The preprocessing of original monitoring data of foundation pit.Firstly,according to the original monitoring data of foundation pit with unequal time interval,Akima interpolation method is proposed for data filling,and the filling effect of cubic spline interpolation method is compared at the same time.The Akima interpolation method is applied to the monitoring data of Shanghai Xujiahui Center project.Secondly,the normality test of daily variation is carried out on the filled monitoring data,and the result shows that the daily variation does not obey the normal distribution.Therefore,the weighted moving range method is proposed to identify the outliers.Finally,in order to eliminate the noise influence of monitoring data in the acquisition process,and fully retain the trend information of original data,db N wavelet function is used to reduce the noise of the monitoring data.ii.Prediction of monitoring data of deep foundation pit based on neural network.The longshort-term memory neural network is used for short-term prediction of the pretreated monitoring data.According to the input feature number of the model,the neural network is divided into single feature model and multi-feature model for training.The training process mainly determines the key parameters such as network initializer,optimizer and activation function of the two models by comparing the loss function.The Pearson correlation coefficient between measuring points is calculated in the multi-feature model,and the combination of high correlation is selected for input to realize the effective prediction of multi-feature model for monitoring data.iii.Evaluation of safety situation of foundation pit based on fuzzy comprehensive evaluation method.Based on the prediction data obtained by neural network,the improved fuzzy comprehensive evaluation method is used to establish the safety situation assessment system for the adjacent environment of deep foundation pit.Firstly,the classification standard and system evaluation factor of environmental safety level are determined,and the parabolic membership function is constructed.Secondly,the evaluation weight is analyzed and calculated by analytic hierarchy process.Thirdly,the security situation assessment system is applied to the project of Shanghai Xujiahui Center.It is concluded that the environment around foundation pit is at the first environmental safety level on November 6,that is,the foundation pit and the adjacent environment are in a safe and stable state.