Dissertation > Mathematical sciences and chemical > Mathematics > Geometry, topology > Vector ( vector ) and tensor analysis

Reliability Analysis Using Evolutionary Neural Network and Its Application in Engineering

Author LinFeng
Tutor DengJian
School Central South University
Course Underground Space Science and Engineering
Keywords Reliability Genetic Algorithms Neural Networks Uniform Design Bishop method
CLC O183
Type Master's thesis
Year 2010
Downloads 145
Quotes 2
Download Dissertation

Implicit limit state function the reliability theory research hotspot issues. Reliability analysis method based on neural networks in recent years has made many research results. However, there are still two important defects does not solve, first, the traditional neural network training algorithm using error back propagation algorithm, the algorithm for the local convergence algorithm, easily lead to neural network training failed; Second, most scholars generally ignored The sample design of the neural network. These defects lead to difficult to solve practical problems. This paper tries to solve these two defects, to improve the practical applicability and accuracy of the theory based on complete theory. The main results are as follows: (1) discusses the evolutionary neural network model, evolutionary neural network model, it is the combination of the genetic algorithm and back propagation algorithm neural network model. Training of the neural network algorithm is essentially an optimization algorithm, genetic algorithm is good at global search short in partial precise search, is contrary to the traditional gradient descent algorithm. Proposed evolutionary neural network first training using genetic algorithms, the global convergence excellent weights, then will get the right value as the value of the initial weight, re-use error back propagation algorithm weights precisely adjusted, compared to traditional Artificial neural network modeling to improve the robustness and accuracy of the modeling. (2) the sample design, uniform design method is used in the experimental design of neural networks. Sample design is very important. How to design a sample that has been ignored by the vast majority of scholars. Explicitly discussed this problem, try to test design ideas for guiding neural network sample design, uniform design sample for neural network design is recommended on the basis of comprehensive analysis of the current domestic and international research. (3) propose a new method based on the mentioned evolutionary neural network reliability analysis. Numerical example proved the feasibility and correctness of the proposed method, and compared with the traditional method has robustness and greater applicability. (4) will be referred to the reliability analysis methods based on evolutionary neural network combined with the Bishop wears a new slope reliability analysis methods and for an open pit slope reliability analysis. The Bishop method widely used due to the efficient and accurate. Combined to enhance the value of its practical application, and reliability of the analytical methods based on this. (5) based on the MATLAB platform, the procedures for the preparation of all the methods mentioned in this article. These procedures a reference for researchers and engineers, and application value.

Related Dissertations
More Dissertations