Study On Predicting The Surface Settlement For Shield Tunneling Based On Intelligent Algorithm

Shield tunneling is becoming the mainstream of the subway tunnel construction methods. It has high degree of automation and less environmental impact. The surface settlement cause by shield tunneling has a great influence on underground pipelines and buildings surround. It may cause major security incidents without effective control.The prediction of surface settlement research status was introduced in this paper. The structure and working principle of Earth Pressure Balanced shield machine were elaborated in detail. The project overview of Beijing 6th subway line was introduced. The mechanism of surface settlement and its development process were studied.The BP neural network and wavelet neural network technique were elaborated in detail. In full consideration of the basic mechanism of settlement constructions on the ground shield and select the more sensitive on the surface settlement parameters to establish a neural network prediction model. A greater amount of monitoring data got from construction site was consolidated then chose the experimental and predicted samples at last. The neural network prediction model had been established was used to analysis the surface settlement.In order to get a good optimization, ant colony optimization was chosen at first, but it was affected by the slow convergence and prone to search stagnation. Differential evolution was introduced and combined it with ant colony optimization to improve the performance of algorithm. A differential evolution ant colony optimization neural network model was established and used it to optimize the original weight, bias(scaling parameters and translation parameters) values of neural network, compared with the ant colony optimization in the advantage of convergence speed and solving accuracy.The application of optimized BP neural network and wavelet neural network to predict the settlement respectively, compared to the differences between the two models in the convergence speed and prediction accuracy. The results revealed that the surface settlement prediction accuracy is great with both models. For each characteristic, it is chosen according to the specific circumstances.