The Research On Slope Stability Evaluation According To The Neural Network And Genetic Algorighm Technical

With the development of large-scale construction projects, more and more high and steep slopes have appeared in all kinds of construction fields, which are always essential factors in deciding whether the project is economical and reasonable or not. Therefore, how to design a slope project or evaluate the stability of a natural slope in an economical, safety and reliable way is of great significance. Among the methods of slope stability evaluation, neural networks method is a relatively effective one. As to this method, some researches about the slope stability evaluation by neural networks method and its optimizing means were done by this thesis.Firstly, this thesis introduced the basic concepts of neural networks and Genetic Algorithms. BP neural networks, main characters and principles of self-organizing neural networks and Genetic Algorithms were also briefly discussed in the first part. Then the paper illuminated that how to take use of self-organizing neural networks in forecasting the stability of a slope and principal ways in optimizing the topological structure of BP neural networks through Genetic Algorithms. On the basis of above theories and methods, the thesis could fully explain how to choose the learning samples and optimize topological structure networks as well.This thesis adopted the self-organizing competitive neural networks to classify those collected slope samples, which made noises caused by learning samples greatly reduced. BP neural networks were learned through its classified samples, which led to improved learning efficiency and enhanced popularizing ability of networks. Furthermore, topological structure of BP neural networks optimizing was achieved by the Genetic Algorithms programming. Experimental results proved that the safety coefficient fitting as well as sample error of BP neural networks had got distinct improvement after their optimization.A general program used to appraise the stability of a slope had been established in this thesis. When applying that program to a real case, we could obtain a more accurate result, which was accord with the actual conditions and had smaller error of safety coefficient. What we only need was finding the appropriate sample set in actual application, so that the goal slope's safety coefficient and its stability conditions could be worked out by the very program.