基于改进支持向量机的深基坑变形预测模型研究

发布时间：2018-06-13 19:40

本文选题：深基坑 + 变型预测　；参考：《江西理工大学》2013年硕士论文

【摘要】：随着城市化进程的快速发展，多层及高层建筑的建造及大量地下空间的开发，大量的深基坑工程不断涌现。深基坑变形监测与预测是深基坑设计施工中的一个重要的环节，也是基坑工程领域研究的热点问题之一。准确地预测深基坑未来的变形，是深基坑变形监测的最终目的。针对传统常用预测方法存在一定的局限性，结合支持向量机的研究现状，提出将能够有效地解决小样本、非线性、高维数、局部极小等问题的支持向量机模型应用于深基坑变形预测。首先，阐述了深基坑变形预测的意义，对深基坑变形预测研究现状作了全面的阐述，，分析了深基坑施工过程中的变形，提出了预测误差最小法来确定样本集的嵌入维数以及时间延迟，实现对样本数据的构造，针对传统支持向量机预测模型参数难以确定的问题，提出采用粒子群算法，通过种群随机初始化、适应度函数设置、粒子更新、终止条件设置，对支持向量机的相关参数进行寻优，得到基于粒子群算法的改进支持向量机预测模型。其次，结合深基坑围护桩桩体两个不同深度的实例测斜数据，根据预测误差最小法求出样本集的嵌入维数以及时间延迟，对变形数据序列进行坐标延迟相空间重构，利用相空间域的相点，通过建立的相空间结构得到学习样本；然后，在Matlab7.14平台上结合Microsoft Visual C++6.0编译器，利用libsvm工具箱进行扩展编程实现对传统支持向量机模型和改进支持向量机模型的训练和预测。最后，根据编制的Matlab程序，将改进支持向量机预测模型与传统的支持向量机模型以及Elman动态神经网络模型的预测结果，采用均方误差、平方和误差、平均相对误差对预测效果进行评价，得出改进支持向量机预测模型的均方误差、平方和误差、平均相对误差分别为0.0155和0.0164、0.1550和0.1639、1.2511%和4.2205%。实验结果表明，基于改进支持向量机预测模型的预测结果均方误差、平方和误差、平均相对误差均优于传统的支持向量机模型和Elman网络模型，通过粒子群算法优选支持向量机预测模型的相关参数，能够得到较好的改进支持向量机预测模型，且拟合效果、泛化性能、稳定性能均更好，具有较高的预测精度，证明了基于改进支持向量机预测模型能更好地反映深基坑系统的动态非线性特点，具有一定的优越性与工程应用推广价值。
[Abstract]:With the rapid development of urbanization, the construction of multi-storey and high-rise buildings and the development of a large number of underground space, a large number of deep foundation pit projects continue to emerge. The deformation monitoring and prediction of deep foundation pit is an important link in the design and construction of deep foundation pit, and it is also one of the hot issues in the field of foundation pit engineering. Accurate prediction of the future deformation of deep foundation pit is the ultimate purpose of deep foundation pit deformation monitoring. In view of the limitations of traditional prediction methods and the current research situation of support vector machine, it is proposed that it will be able to solve the problem of small sample, nonlinear and high dimension effectively. The support vector machine (SVM) model for local minima is applied to deep foundation pit deformation prediction. First of all, the significance of deep foundation pit deformation prediction is expounded, the research status of deep foundation pit deformation prediction is comprehensively expounded, and the deformation in deep foundation pit construction process is analyzed. The minimum prediction error method is proposed to determine the embedded dimension and time delay of the sample set, and the construction of the sample data is realized. Aiming at the difficulty of determining the parameters of the traditional SVM prediction model, the particle swarm optimization (PSO) algorithm is proposed. Through population random initialization, fitness function setting, particle update, termination condition setting, the parameters of support vector machine are optimized, and an improved support vector machine prediction model based on particle swarm optimization algorithm is obtained. Secondly, the embedding dimension and time delay of the sample set are calculated according to the prediction error minimization method, and the coordinate delay phase space reconstruction of the deformation data sequence is carried out. Using the phase points of the phase space domain, the learning samples are obtained through the phase space structure established, and then, the Microsoft Visual C 6.0 compiler is combined with the Matlab 7.14 platform. The traditional support vector machine model and the improved support vector machine model are trained and predicted by extended programming with libsvm toolbox. Finally, according to the Matlab program, the prediction results of the improved support vector machine model, the traditional support vector machine model and the Elman dynamic neural network model are presented, and the mean square error and square sum error are adopted. The mean relative error of the improved support vector machine prediction model is estimated. The mean square error and square sum error are 0.0155 and 0.01640.1550 and 0.16391.2511% and 4.2205 respectively. The experimental results show that the mean square error, square sum error and average relative error of the prediction results based on the improved support vector machine prediction model are better than those of the traditional support vector machine model and the Elman network model. The particle swarm optimization algorithm is used to optimize the parameters of the prediction model of support vector machine (SVM), and the prediction model of SVM can be improved, and the fitting effect, generalization performance and stability performance are better, and the prediction accuracy is higher. It is proved that the prediction model based on improved support vector machine can better reflect the dynamic nonlinear characteristics of deep foundation pit system, and has certain superiority and application value in engineering application.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TU196.1;TU753

【参考文献】