【作者】 郭静；

【导师】 秦荣；

【作者基本信息】 广西大学 ， 固体力学， 2008， 硕士

【摘要】 桥梁结构的承载力可靠性评估是目前我国工程界亟待解决的问题。神经网络由于其极强的学习能力和非线性大规模并行处理的能力,正适合预测这类复杂的问题。本文以百色华村大桥为例,基于对静载试验以及神经网络机理的认识,提出了连续钢构桥承载力预测的神经网络模型。本文首先阐述了承载力的影响因素和评定方法;其次总结了桥梁结构静载试验的一般流程方法和事项,并进行了百色华村大桥静载试验;然后介绍了神经网络的基本概念、基本原理以及BP神经网络的有关特点,并对遗传算法的相关原理进行了阐述。最后在此基础上建立了承载力预测的神经网络模型。本文重点对神经网络方法在承载力预测这一问题上的可行性进行讨论。建立了以应力作为输入层神经元、挠度作为输出层神经元的三层BP神经网络。基于MATLAB R2007平台,利用其自带的神经网络工具箱,编写了相应的程序。针对BP神经网络的缺点,采用了学习率自适应调整、试算法选择隐层神经元结点数以及遗传算法优化网络初始权重的方法对BP神经网络加以优化。最后,利用该神经网络预测模型进行了训练及预测分析,并将神经网络预测结果同静载试验结果进行了对比分析,验证了其良好的预测效果。更多还原

【Abstract】 Reliability assessment of the bridge is an urgent problem to be resolved presently in engineering field. Fortunately ,thanks to its great learning ability and non-linear massive simultaneous managing capacity, the neural network provides a model for determining the bearing capacity of bridges. This paper, based on the knowledge of the dynamical test and the mechanism of the neural network, proposes the neural network model on the prediction of bearing capacity.Firstly, this paper expands some parameters and methods of load carrying capacity. Secondly,summarized the general methods and notices in dynamical test. Then, it introduces the basic concept and fundamental principle of neural network, the relevant characteristics of BP network and the principles of the genetic algorithm. Finally, on the basis of previous work, the paper sets up the neural network model on the prediction of bearing capacity .The paper mainly focuses on the feasibility and spueirority of the neural netwok model in solving the problem. Based on the analysis of all factors concerning the bearing capacity, it esbtablishes the 3-level BP network with stresses as neuron of input level and warping as neuron of output level, and works out a relevant program, making use of the neural network toolbox of MATLAB R2007. Contrapose the disadvantages of BP network,the paper employs the self-adaption method to adjust the learning rate, provisional estimate to determine the number of neuron joints of hidden layer and using genetic algorithm to optimize the initial weight of BP network.In the end, the paper applies the neural network model, compares the result by using neural network with results of dynamical test to prove a favorable prediction effect.更多还原