毕业设计（论文）基于MATLAB的BP神经网络设计与仿真.docVIP

SHANDONGＵＮＩＶＥＲＳＩＴＹ　ＯＦ　ＴＥＣＨＮＯＬＯＧＹ 毕业设计说明书 基于MATLAB的BP神经网络设计与仿真 学 院： 计算机科学与技术 专 业： 计算机科学与技术 学生姓名： 学 号： 指导教师： 2012 年 6 月 摘 要 设计进行地震预测采用单隐层的BP网络，通过查找往年的数据，可以提取出7个预报因子，作为输入样本为7维的输入向量。所以有7个神经元作为输入层，有15个神经元作为中间层。因为网络的输出数据只有一个，则输出层只有1个神经元，所以说网络采用的结构为7*15*1。网络以梯度下降法进行学习，学习速率为自适应。所用到的训练函数是traingdx。主要成果是：由于中间层的神经元个数能直接影响着网络预测的性能，要想达到最佳的性能，通过大量的实验数据检测，当网络达到最小的预测误差时，网络中间层神经元的个数就取得最佳值，同时也证明了一个重要的结论：如果增加中间层的神经元个数，网络的反应精度在一定程度上会有所提高，而网络的性能不一定会提高。 关键词：BP网络 神经元 梯度下降法 Abstract This design adopts the BP network of the single-hidden layer neuron to do the earthquake predication. According to searching the data of recent years, I put forward seven forecast factors which are the input vector of seven-dimensional input samples. This shows that there are 7 neurons in the input-layer and 15 neurons in the Middle Tier. Since the network has only one output datum, the output layer has only one neuron and the network structure adopts the 7*15*1. The training function of the network is traingdx. It functions through the gradient descent algorithm and adopts the study velocity. The main achievements: Since the number of the neurons in the Middle Tier are not only difficult to confirm but also influence the predictable performance, it can illustrates through a large number of the experimental data that when the predictable errors come to the least, the numbers of the neurons reach their maximum. This also testifies an important conclusion: Although increasing the number of neurons can enhance the network’s reaction precise, it can not always advance the capacity of the network. Keywords : BP network , neurons , the gradient descent algorithm 目 录 摘 要…………………………………………………………………………….. 2 Abstract（英文摘要）………………………………………………………. 3 目 录…………………………………………………………………………….. 4 引言………………………………………………………………………………… 6 课题的背景和意义………………………………………………………... 6 人工神经网络的发展……………………………………………………... 6 人工神经网络的应用………………