ABSTRACT
In recent years, the hotspot problem in the tunnel has been shifted from how to better construct the tunnel to how to effectively maintain and ensure the safety of the tunnel, among which the long-term settlement prediction of tunnel is of great interest. Through the existing monitoring data to quantitatively predict the long-term settlement of the tunnel in the future will help us to make effective judgments and take corresponding active control measures in advance to guarantee the safety of the tunnel. As a time-series data, the monitored long-term settlement of the tunnel can be highly fitted by Genetic Algorithm optimizing Back Propagation neural network. However, as a simple mathematical algorithm, the neural network does not consider the longitudinal deformation characteristics and mechanical mechanism of the tunnel when performing the fitting prediction. Based on the traditional Genetic Algorithm optimizing Back Propagation neural network, this paper proposes a classification prediction strategy that can overcome the shortcomings of disregarding the structure. In order to verify the superiority of the proposed strategy, the model is trained and predicted using the monitoring data from the Oriental Sports Center to the Lingzhao Xincun in Shanghai Rail Transit Line 8. Compared with the traditional Genetic Algorithm optimizing Back Propagation model, the improved strategy can reduce the prediction error of tunnel settlement by 20%. In addition, the predicted settlement using the proposed strategy is closer to the monitoring value. The Mean Square Error between them reaches 0.457 and the THEIL Inequality Coefficient is only 0.028, proving the validity of the method.
