Abstract Many factors affect the stability of a tunnel cavern foundation; there is a complex nonlinear relationship
between these factors and the tunnel cavern foundation stability, and conventional methods have difficulty describing this complex relationship. Therefore, a dual-stage multi-strategy particle swarm optimization (DMPSO) BP neural network model is proposed for evaluating the stability of tunnel cavern foundations. The particle swarm algorithm is characterized by a strong capability for global optimization and a high search efficiency, with these characteristics being more prominent after improvement of the algorithm. The BP algorithm has a strong capacity for nonlinear mapping and generalization, with the disadvantages of slow convergence and local optimization. By using the dual-stage multi-strategy particle swarm optimization (DMPSO) to determine the weights and thresholds of the BP neural network, the shortcomings of the BP neural network are solved and the prediction accuracy is improved. Using the
Chongqing Xiaoshizi station cavern foundation as an example, the feasibility of the dual-stage multi-strategy particle swarm optimization BP neural network model (DMPSO-BP) is verified, with the prediction accuracy of the proposed model being better than that of the fuzzy neural network model and particle swarm optimization BP neural network (PSO-BP) model.
Abstract：
Many factors affect the stability of a tunnel cavern foundation; there is a complex nonlinear relationship
between these factors and the tunnel cavern foundation stability, and conventional methods have difficulty describing this complex relationship. Therefore, a dual-stage multi-strategy particle swarm optimization (DMPSO) BP neural network model is proposed for evaluating the stability of tunnel cavern foundations. The particle swarm algorithm is characterized by a strong capability for global optimization and a high search efficiency, with these characteristics being more prominent after improvement of the algorithm. The BP algorithm has a strong capacity for nonlinear mapping and generalization, with the disadvantages of slow convergence and local optimization. By using the dual-stage multi-strategy particle swarm optimization (DMPSO) to determine the weights and thresholds of the BP neural network, the shortcomings of the BP neural network are solved and the prediction accuracy is improved. Using the
Chongqing Xiaoshizi station cavern foundation as an example, the feasibility of the dual-stage multi-strategy particle swarm optimization BP neural network model (DMPSO-BP) is verified, with the prediction accuracy of the proposed model being better than that of the fuzzy neural network model and particle swarm optimization BP neural network (PSO-BP) model.
2017, Vol. 54 