Abstract In view of the complexity and uncertainty of metro tunnel settlement, the main influence factors are analyzed in depth and the weight of each factor is obtained by using the super decisions (hereinafter referred to as the "SD") software solution of ANP. Using these weights as the initial weights, fine-tuning of weights is carried out by training the BP neutral network. Based on the modification of the weights and comprehensive consideration of the mutual feedbacks and effects among various factors and levels, an ANP-BP model is proposed. In the practical application of the Xi'an Metro project, the tunnel settlements are predicted with the ANP-BP model. Compared to results obtained by the GA-BP model and the PSO-BP model, the prediction results of the ANP-BP model are better in regard to the characteristics of sound adaptability, fast convergence, and high accuracy.
Abstract：
In view of the complexity and uncertainty of metro tunnel settlement, the main influence factors are analyzed in depth and the weight of each factor is obtained by using the super decisions (hereinafter referred to as the "SD") software solution of ANP. Using these weights as the initial weights, fine-tuning of weights is carried out by training the BP neutral network. Based on the modification of the weights and comprehensive consideration of the mutual feedbacks and effects among various factors and levels, an ANP-BP model is proposed. In the practical application of the Xi'an Metro project, the tunnel settlements are predicted with the ANP-BP model. Compared to results obtained by the GA-BP model and the PSO-BP model, the prediction results of the ANP-BP model are better in regard to the characteristics of sound adaptability, fast convergence, and high accuracy.
2013, Vol. 50 