Study On Construction Monitoring And Deformation Prediction For Deep Excavation In Railway Station North Square Of Changchun

With the development and utilization of underground space, all types of building foundations, subways, underground stations, underground warehouses, a large number of underground construction projects, which produce large amounts of excavation. And the scale of deep foundation expanded and the depth of the deepening trend. In recent years, pit accidents have occurred, but with the complexity and uncertainty of geotechnical engineering, it is difficult to determine problems that may occur in the construction process. So taking safety monitoring in the construction project, and analysis the safety condition by monitoring data, it becomes an important tool to ensure security of the current excavation pit. In the process of deep foundation monitoring, deformation prediction of deep foundation becoming importance, to grasp the deformation of the foundation, in order to arrange the construction methods to ensure that excavation safety.First, a three-dimensional model of item analysis is established by FLAC3D, and combined with Chang Chun Railway Station North Square Foundation field tested data, analysis of the variability characteristics of foundation. Mohr-Coulomb elastic model is used to simulate soil, pile element is used to simulate pile enclosure, the cable anchor unit is used to simulate the structure, respectively. The three-dimensional solid model of foundation is established to numerical analysis the soil settlement around foundation, retaining pile displacement and axial force of anchor. The results can reflect the characteristics of pit deformation preferably, the results are consistent with the final measured to verify the correctness of the model.Subsequently, the method of construction monitoring is determined according to the results of theoretical analysis. It includes vertical displacement of surrounding soil, horizontal displacement of enclosure structure and settlement and inclined of surrounding building. In this chapter, the method of monitor item is described in detail, and the alarm value have determined. In the early of excavation, deformation rate of soil is big as the speed of excavation is fast. When excavation reaches the design elevation, deformation rate of soil is reduction. In spring, deformation rate of soil increase because of temperature change violently, the soil between frozen and thaw. When the concrete in bottom have finished pouring, the deformation of soil and retaining pile is stability. The results can reflect the characteristics of pit deformation preferably, the results are consistent with the final measured to verify the correctness of the model. Simulation results and measured data show that in the early of excavation, the excavation speed, soil deformation rates were higher. At the end of excavation, the soil deformation rate slowed down. When entering the spring, due to drastic temperature changing and soil freezing and thawing, soil deformation phenomenon also appeared repeatedly. After poured concrete in the floor, the distortion of soil and retaining pile were basic stop.Finally, the PSO algorithm to optimize BP neural network is used to predict the horizontal displacement of the pile, the surrounding ground surface subsidence, cable axial force. As BP neural network with good nonlinear mapping ability, the prediction has been widely used. However, BP neural network model also exists on the initial weights, the larger the threshold dependence; slow convergence; easy to fall into local minimum and so inadequate. In order to improve the convergence speed and reliability of BP neural network, this paper attempts to use PSO algorithm to the initial weights of BP neural network and the threshold to optimize the BP neural network with three layers to predict deep horizontal displacement of soil around foundation, retaining pile horizontal displacement, axial cable. In order to improve the diversity of the sample, the dynamic prediction method, taking sample feedback, increasing training samples, and re-training of BP neural network constantly. Examples of actual projects by the predicted results agree well with the actual prediction which can prove PSO algorithm to optimize BP neural network is feasible.