【作者】 熊光赤；

【导师】 阮永芬；

【作者基本信息】 昆明理工大学 ， 结构工程， 2006， 硕士

【摘要】 本论文运用人工智能的基本知识,结合研究边坡稳定性评价及位移预报,建立了基于遗传—模拟退火算法的边坡工程稳定性分析评价系统以及基于遗传—改进BP神经网络模型的边坡变形的非线性时间序列的预测模型,并编制了相应的计算程序,使边坡稳定性和变形的研究在一定程度上达到智能化。 本文分别对圆弧滑动面的圆心、半径进行遗传进化计算,并用所得结果与传统计算方法所得的结果进行比较,表明遗传进化算法在边坡工程的计算结果是可靠的,采用遗传进化算法有效的克服了传统方法易陷入局部最优解的缺点,且遗传算法对目标函数的形式和性质要求较低,有很好的重复性。此外,遗传算法没有边坡危险滑动面一定要通过坡脚的假定,而是随机的搜索确定的。 本论文采用简化Bishop法与遗传—模拟退火算法相耦合的计算方法,通过分别对圆弧滑动面的圆心、半径进行遗传—模拟退火的耦合计算,并且与其他传统方法的计算结果进行比较,表明该方法所得的计算结果是可靠的。在计算力学模型相同的条件下,遗传—模拟退火算法的计算结果基本不会大于遗传算法的计算结果。这充分体现了遗传—模拟退火算法不仅较好的利用了遗传算法的全局寻优的优势,而且利用了模拟退火算法的较强的局部寻优能力的特点。 在传统的非线性时间序列边坡变形预测模型的基础上,本论文考虑降雨对边坡变形的影响而建立的遗传—改进BP神经网络预测模型在新滩滑坡的变形预测中取得了较好的预测效果,结果表明考虑降雨因素影响的预测模型比不考虑降雨因素影响的预测模型的预测精度高,能满足中长期的边坡变形预报。 以上所提出的边坡变形计算模型在三峡永久船闸三闸首中隔墩岩体的开挖变形、小湾电站边坡变形以及新滩的滑坡变形预测中得到了验证,预测结果均与工程实际稳定性状况和变形实测值相符合,预测的相对误差基本小于5%,从而证明上述所提出的理论方法的可靠性和有效性。更多还原

【Abstract】 This paper Applies the theory of Artificial intelligence and associating to study slope stability analysis and deformation prediction, it renders the slope stability analysis system based on Genetic Simulated Annealing Algorithms and Non-linear Time series predictive model based on Improved BP Genetic Natural Network model, and compiles relevant programs, so that make the studying on slope stability and deformation intelligent in a sense.The search of critical surface of circular slip can be transformed into The Genetic Algorithms’ calculation on the search of coordinates of circle center and it’s radius. By comparing the calculative results of the traditional calculation method, it shows that the results of the Genetic Algorithms is reliable in the analysis of slope engineering. The application of the Genetic Algorithms effectively overcome the disadvantage that the traditional method easily tap in local optimal solution, and it almost has not requirement about the form and character of objective function, it has better repeatability. Moreover, the Genetic Algorithms does not assume that the most dangerous sliding surface must pass foot of slope, which is determined by the searching course at random.This text adopt the coupling calculation of simplified Bishop method and the Genetic Algorithms method by comparing with other traditional method, it shows that the coupling calculation is reliable. On the condition of sample calculation mechanical model, Genetic simulated Annealing Algorithms method’s calculation results of all more superiority than the calculation results of Genetic Algorithms. It indicates that Genetic Simulated Annealing Algorithms not only take advantage of the advantage of global optimization of Genetic Algorithms, but also take advantage of the characteristic of local optimization of simulated Annealing Algorithms.This text consider rainfall as important factor deformation and establish put forward Improved BP Genetic Natural Network predictive model based on the slope deformation’s predictive Model of traditional Non-Linear Time Series. This method obtains better results in the application of deformation prediction of Xintan Landslide, it’s results shows that prediction model of considering rainfall as influence factors ofslope deformation’s precision of prediction more superiority than the prediction model of take no account of rainfall influence factors. It can be applied to long-term prediction of slope deformation.All prediction models of slope deformation above have been applied and verified in the rock mass deformation prediction of Three Gorges Project > the slope deformation prediction of Xintan landslide and the slope deformation prediction of Xiaowan Hydropower station, all the results of analysis are basically agreed with the practical stability situation and deformation situation. It’s prediction relative error is basically less than 0.05. Therefore, it is proved that the prediction theories and methods put forward in the dissertation are reliable and effective.更多还原