{"title":"Application of GA-BP in Displacement Force Inverse Analysis and Mechanical Parameter Inversion of Deep Foundation Pits","authors":"Guo Yunhong, Zhang Shihao","doi":"10.13052/ejcm2642-2085.3213","DOIUrl":null,"url":null,"abstract":"Aiming at the defects of various existing displacement inverse analysis methods, using the nonlinear mapping ability of neural network and the global random search ability of genetic algorithm, this paper proposes a displacement inverse analysis method based on optimized Genetic Algorithm- Back Propagation (GA-BP) for deep foundation pit support. The method changes the method that BP algorithm relies on the guidance of gradient information to adjust the network weights, but uses the characteristics of global search of genetic algorithm to find the most suitable network connection rights and network structure, etc. to achieve the purpose of optimization. Firstly, the deformation mechanism of deep foundation pit is analyzed, its failure mode is summarized, and the calculation method of lateral rock and soil pressure is sorted out according to the code. The theory and characteristics of BP neural network and genetic algorithm are discussed, and the method of using genetic algorithm to optimize BP neural network is proposed to improve the prediction accuracy. In view of the shortcomings of GA-BP neural network prediction model in training sample pretreatment and hidden layer structure design, the optimal normalization interval was determined by correlation coefficient regression analysis, and the analytical expression of the number of neurons in hidden layer was derived by statistical principle, and the value range of the optimal number of neurons in single hidden layer was proposed. Combined with the actual engineering, the mechanical parameters inversion and displacement force inverse analysis are performed using this method, and the results show that the optimized GA-BP has higher prediction accuracy compared with BP neural network and GA-BP, and the deviation of the displacement prediction value at each depth is kept within 0.2 mm, the absolute error interval width is 0.07 mm, and the maximum relative error is 1.35% at 4.0 m depth.","PeriodicalId":45463,"journal":{"name":"European Journal of Computational Mechanics","volume":null,"pages":null},"PeriodicalIF":1.5000,"publicationDate":"2023-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"European Journal of Computational Mechanics","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.13052/ejcm2642-2085.3213","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"MECHANICS","Score":null,"Total":0}
引用次数: 1
Abstract
Aiming at the defects of various existing displacement inverse analysis methods, using the nonlinear mapping ability of neural network and the global random search ability of genetic algorithm, this paper proposes a displacement inverse analysis method based on optimized Genetic Algorithm- Back Propagation (GA-BP) for deep foundation pit support. The method changes the method that BP algorithm relies on the guidance of gradient information to adjust the network weights, but uses the characteristics of global search of genetic algorithm to find the most suitable network connection rights and network structure, etc. to achieve the purpose of optimization. Firstly, the deformation mechanism of deep foundation pit is analyzed, its failure mode is summarized, and the calculation method of lateral rock and soil pressure is sorted out according to the code. The theory and characteristics of BP neural network and genetic algorithm are discussed, and the method of using genetic algorithm to optimize BP neural network is proposed to improve the prediction accuracy. In view of the shortcomings of GA-BP neural network prediction model in training sample pretreatment and hidden layer structure design, the optimal normalization interval was determined by correlation coefficient regression analysis, and the analytical expression of the number of neurons in hidden layer was derived by statistical principle, and the value range of the optimal number of neurons in single hidden layer was proposed. Combined with the actual engineering, the mechanical parameters inversion and displacement force inverse analysis are performed using this method, and the results show that the optimized GA-BP has higher prediction accuracy compared with BP neural network and GA-BP, and the deviation of the displacement prediction value at each depth is kept within 0.2 mm, the absolute error interval width is 0.07 mm, and the maximum relative error is 1.35% at 4.0 m depth.
针对现有各种位移逆分析方法的缺陷,利用神经网络的非线性映射能力和遗传算法的全局随机搜索能力,提出了一种基于优化遗传算法-反向传播(GA-BP)的深基坑支护位移逆分析方法。该方法改变了BP算法依赖梯度信息引导调整网络权值的方法,而是利用遗传算法全局搜索的特点,寻找最合适的网络连接权和网络结构等,达到优化的目的。首先,分析了深基坑的变形机理,总结了深基坑的破坏模式,并根据规范整理了深基坑侧向岩土压力的计算方法。讨论了BP神经网络和遗传算法的原理和特点,提出了利用遗传算法对BP神经网络进行优化以提高预测精度的方法。针对GA-BP神经网络预测模型在训练样本预处理和隐层结构设计方面的不足,通过相关系数回归分析确定最优归一化区间,利用统计原理推导出隐层神经元数的解析表达式,提出了单隐层最优神经元数的取值范围。结合工程实际,利用该方法进行了力学参数反演和位移力逆分析,结果表明:优化后的GA-BP与BP神经网络和GA-BP相比具有更高的预测精度,各深度位移预测值偏差控制在0.2 mm以内,绝对误差区间宽度为0.07 mm, 4.0 m深度最大相对误差为1.35%。