This paper addresses the low efficiency problem of traditional geotechnical finite element calculation methods used in parameter design and optimization process of composite foundation, which is inherently nonlinear and with multiple constraints.A new method that integrates an improved surrogate model and an intelligent optimization algorithm is proposed.To achieve the fitting of multiple constraint conditions such as foundation settlement, stability, bearing capacity and the objective function of project cost, a backpropagation neural network (BPNN) surrogate model with weight initialization is established based on the parameter correlation analysis.Then the surrogate model is globally optimized by using the particle swarm optimization (PSO) algorithm to get the optimal parameter combination that satisfies the constraints.To comprehensively evaluate the performance of the method, two sets of comparative numerical simulations are implemented by taking the typical subgrade work point DK37+168 of the Ganzhou-Shenzhen High-Speed Railway as an example.In the fitting stage, the BPNN surrogate model is compared with the traditional surrogate models based on response surface method (RSM) and support vector machine (SVM).In the optimization stage, the method is compared with the genetic algorithm (GA) and the global traversal method.It is shown that the BPNN surrogate model has better nonlinear regression ability with all fitting determination coefficient values exceeding 0.99.Meanwhile, the method shows significant advantages in optimization efficiency and stability as well.Under the same termination conditions, the total time consumed by the method for one single calculation is 37.63 s, which is 48.8% lower than that of GA (73.54 s).Additionally, the method converges to the global optimal solution obtained by the global traversal method (117.9 s, lower calculation efficiency) with 80% probability, while GA has only 40% probability of achieving the same accuracy.The proposed method is expected to provide a reference for the design of optimization processes in similar engineering problems.
ChenX Y, DongS X, WangB, et al.Optimized design of CFG pile composite foundation based on genetic algorithm [J].Journal of Yan'an University (Natural Science Edition), 2023, 42(3): 105-109.
ChenC F, LiX, MaoF S.Settlement-based optimization design on composite foundation in region with deep soft soils by orthogonal numerical tests [J].Highway Engineering, 2018, 43(2): 49-53, 107.
ChenC F, LiX, GongX N, et al.Composite foundation's optimization design of settlement's surrogate models based on support vector machines [J].Journal of Railway Science and Engineering, 2018, 15(6): 1424-1429.
JiaoJ.The intelligent design of CFG piles composite foundation based on BP neural networks [D].Handan: Hebei University of Engineering, 2007.
[18]
焦健.基于BP神经网络的CFG桩复合地基智能设计[D].邯郸: 河北工程大学, 2007.
[19]
ChenY Z.Lightweight design of thin-wall structure based on machine learning and intelligent optimization [D].Changsha: Hunan University, 2022.
[20]
陈延展.基于机器学习与智能优化的车身薄壁结构轻量化设计[D].长沙: 湖南大学, 2022.
[21]
YuanY X.Evolutionary algorithm based on BP neural network proxy model and its application [D].Changchun: Jilin University, 2021.
[22]
袁翌昕.基于BP神经网络代理模型的进化算法及其应用[D].长春: 吉林大学, 2021.
[23]
HanY, YuM.A review on application of particle swarm optimization in path planning of mobile robot [J].Manufacturing Technology Machine Tool, 2024(7): 85-90.
ZhangX Q, ChengQ L, SunW, et al.Research on a TOPSIS energy efficiency evaluation system for crude oil gathering and transportation systems based on a GA-BP neural network [J].Petroleum Science, 2024, 21(1): 621-640.
CaoX Y, ManX Y, WangJ P, et al.Deformation prediction of treated landslides based on BP neural network optimized by bird swarm algorithm [J].Bulletin of Surveying and Mapping, 2023(5): 27-31.