Clonal Selection Algorithm And Its Application To Some Questions Of Foundation Engineering

In foundation engineering, issues of backward analysis of displacements, measure and determination of ground modulus, modeling of the dynamic response and prediction of foundation settlement show complex features of high dimension, multi-peak value, nonlinear, discontinuousness and non-convexity and so on, and often contain all kinds of uncertain information types. According to the basic idea of evolutionary algorithm, solutions to typical issues can be transformed to solutions to optimization issues in foundation engineering. Traditional deterministic optimization method comes into contact with difficulty when it tries to solve such complex optimization issues. Intelligent computing technology with evolutionary algorithm as representative provides a new way for solutions to complex optimization issues, however, further application of evolutionary algorithm is restricted because the algorithm itself exists insufficiency. Based on above background and related research subjects, some research works are done on the Clonal Selection Algorithm (CSA) in this dissertation, and the author presents a solution method for backward analysis of displacements based on the combination of CSA and BP neural network. At the same time, the solution method for deformation prediction and modeling based on the combination of CSA and the Support Vector Machines (SVM) is also suggested.Biological basis, basic principle and the process of implementation of the CSA are described in detail in the text. Based on the better understanding of the mechanism of the CSA, the influence on algorithm performance by related parameters of the CSA is discussed, and value range of the related parameters is defined further so as to ensure the excellent performance of the CSA. At the same time, the performance of the CSA for complex constrained optimization issues is also analyzed. It is found by simulation test that the application of the CSA for complex optimization brings many characteristics like multiple point operation, parallel computing, fast convergence, powerful searching ability and well stability.Mechanical properties of geotechnical media materials come before the solution of the geotechnical issues; then the methods of backward analysis for displacements are effective ways to acquire the mechanical properties of geotechnical media materials. There exist some disadvantages in the traditional method of backward analysis for displacements, such as much amount of calculation, slow convergence speed, bad stability and unease mastering. When this method is adopted, the accuracy of mechanical properties of geotechnical media materials is often unsatisfactory. The author presents a new solution method for backward analysis of displacements based on the CSA. Its key is that BP neural network is applied as a tool in the forward process instead of the finite element method. The new method can be used to describe nonlinear mapping relationship between mechanical parameters of geotechnical media materials and displacements, thus to reduce the amount of calculation. And then the CSA is used as a backward tool to optimize the mechanical parameters of geotechnical media materials. The results show that higher precise can be obtained. Higher precise displacements can be estimated according to the higher precise mechanical parameters of geotechnical media materials. So it benefits for construction guiding and process monitoring of deformation in foundation pit engineering. It is proved the new method is feasible and effective according to comparative analysis of simulation test.The SVM technique is a useful tool of processing small sample data modeling. But the performance of the SVM technique is rather sensitive to its related parameters, that is to say that small changes of its related parameters will make great influence on the performance of the SVM. The CSA is proposed in this text to process the related parameters of the SVM to improve its performance. Correct classification rate is choice as an objective function for classification issue, and mean square error between fitting numerical value and real numerical value is choice as an objective function for regression issue. It is proved by simulation experiment that the optimum matching of related parameters of the SVM can be achieved with the CSA. Thus high precise results can be achieved applying the SVM based on the CSA in solving ground modulus, modeling of dynamic response and prediction of foundation pit settlement.Further research on the CSA is theoretically significant to its development. It is valuable both in theory and engineering practice that the CSA with BP neural network is applied to solve the issue of backward analysis of displacements in the foundation pit engineering, and the CSA with SVM to solve ground modulus, modeling of dynamic response and deformation and settlement in the foundation engineering. The work done in this dissertation not only establish a model and frame to solve corresponding practical issues in the foundation engineering but also lay a good basis to solve similar issues in other engineering field.