【作者】 张飞；

【导师】 刘文生；

【作者基本信息】 辽宁工程技术大学 ， 大地测量学与测量工程， 2011， 硕士

【摘要】 伴随着改革开放三十年我国的经济腾飞,水电建设产业高速发展,水坝的规模不断扩大和增高,同时大坝的安全问题也显得更为突出和重要。大坝变形预测是大坝安全监测系统的重要一部分,对于确保大坝安全运行具有非常重要的作用。大坝变形预测就是根据已有的监测资料预测未来的大坝变形量。大坝本身内部结构复杂、工作环境恶劣,并且还有繁多的不确定性影响因素,这些因素和大坝变形量之间的关系难以定量确定。因此,本文将具有自组织性、自学习性、自适应性和模糊推理能力等特点的BP神经网络应用于大坝变形预测,利用它的非线性函数逼近能力来有效地拟合大坝的变形量与影响因素之间的非线性函数关系。但在BP神经网络应用于大坝变形预测的实际过程中,也存在着一些不足之处,因此需要采取一定的方法和措施来改善BP神经网络的算法性能。本文通过对BP网络模型结构及学习规则的研究,针对BP神经网络的初始化权值和阈值的随机性,导致训练速度慢和易落入局部极小等弱点,运用具有并行特性和全局优化能力的粒子群算法(PSO)对BP神经网络的权值和阈值进行优化,建立了粒子群BP神经网络模型,并以丰满大坝多年监测的坝顶水平位移资料为例进行实证分析,采用MATLAB软件编制模型的程序,并与经典的BP神经网络模型的拟合预测结果相比较,分析得出PSO-BP模型的收敛速度更快、预测精度更高。另外,本文尝试对大坝的变形区间进行定级预测,并建立相应的基于PSO-BP神经网络的大坝变形区间预测模型,同样采用丰满大坝坝顶水平位移资料,运用MATLAB软件编程实现仿真,通过对拟合预测结果分析显示,基于PSO-BP网络的大坝变形区间预测模型可以满足生产实际的精度需求,也是一种值得采用的预测模型。更多还原

【Abstract】 As China’s economy boom in the in the recent three decades after opening-up and reform, the water power engineering construetion develops at a high speed, the body of dams beeomes huger and higher, and the safety of dams also has taken more and more attention. Dam deformation prediction is an important part of dam safety monitoring system, and plays a very important role in safeguarding the seeurity of dam.Dam deformation prediction is based on known monitoring data to prediction the future deformation. The internal structure and the working condition of the dam is complex, and there are various uncertainty factors,these factors can’t be described by certain ration relation in traditional models. It is hard to quantitatively determine the relationship of these factors and dam deformation. Therefore, this paper applys BP Neural Network, which has organization capability, self-educated capability, adapt capability and fuzzy ratiocinative capability, in the filed of dam deformation prediction, uses its nonlinear function approaeh ability to simulate the nonlinear relationship of the deformation of the dam and the influencing factors. But during the using of BP Neural Network in dam deformation predictioning, we find some shortcoming. So, we must take some methods to improve performance of the BP Neural Network.Initialized weights and threshold of the BP neural network is random, which results in slow convergence and easily convergence to local optima. According to these characteristics, this paper applys the Particle Swarm Optimization (PSO), which has a strong global searching ability, to optimize the weights and threshold of the BP neural network. This paper utilizes the transverse displacement monitoring data of Fengman Dam, establishs a PSO-BP model and applys MATLAB to simulate it, and then contrast the result with classic BP neural network mode. Results show that PSO-BP model is faster in training and more prediction accurate.In addition, the paper attempts to prediction the dam deformation range, which is more in line with the actual deformation in theory, establishs an appropriate PSO-BP neural network prediction model of the dam deformation range and applys MATLAB to simulate it. Through the analysis of the prediction results, we get a conclusion that the PSO-BP neural network model is feasible to prediction the dam deformation range.更多还原