一种基于自适应遗传算法优化BP神经网络的漏磁缺陷量化框架

基金项目:

重点研发计划资助项目（2017YFF0108800）

详细信息

MFL defect quantification framework based on combination of adaptive genetic algorithm and BP neural network

摘要: 由于传统遗传算法在优化BP神经网络应用中训练速度慢、拟合效果不好，所以神经网络对管道漏磁缺陷的量化精度差。将自适应遗传算法引入到BP神经网络中，进行漏磁缺陷的尺寸反演，根据实测漏磁缺陷的数据特点，自适应设定交叉算子和变异算子的交叉率和变异率，进而优化BP网络的初始权值；采用不同尺寸的缺陷特征库训练网络，实现对缺陷长度、宽度、深度的量化。
- 漏磁缺陷 /
- 尺寸量化 /
- 自适应遗传算法 /
- BP神经网络
Abstract: In order to overcome the disadvantages of BP neural network based on traditional genetic algorithm, such as slow training speed, low quantitative accuracy, based on improving the computing method of the rate of crossover and mutation, also taking current stage of the population's evolution into consideration, an improved adaptive genetic algorithm is introduced. This new algorithm is used to optimize BP network's initial weights, and the defect features from magnetic flux leakage are used to train the network, so as to achieve the quantification of length, width, depth of the defects. With this model, the training time of net work can be saved and the computing accuracy can be improved as well.
- MFL defect /
- defect quantification /
- adaptive genetic algorithm /
- BP neural network

[1]	吴欣怡, 赵伟,黄松岭. 基于漏磁检测的缺陷量化方法[J]. 电测与仪表,2008(5):20-22,37.
[2]	韩文花, 阙沛文,梁巍. 改进的遗传局部搜索算法在漏磁逆问题中的应用研究[J]. 上海交通大学学报,2007,41(5):751-754.
[3]	陈政. 一种改进的遗传算法优化BP网络的研究及应用[D]. 广州:暨南大学,2010.
[4]	韩文花, 靳希,任海霞. 基于遗传模拟退火算法的漏磁缺陷重构[J]. 应用基础与工程科学学报,2007,15(2):257-262.
[5]	国嘉, 王瑞嘉,王海家. 基于遗传算法的神经网络学习方法[J]. 计算机与数字工程,2004(5):98.
[6]	魏海坤. 神经网络结构设计的理论与方法[M]. 北京:国防工业出版社,2005.
[7]	RAMUHALLI P,UPDA S.Neural network-based inversion algorithms in magnetic flux leakage nondestruction evaluation[J].Journal of Applied Physics,2003,93(10):8274-8275.
[8]	高宏宾, 焦东升,彭商濂. 一种基于遗传算法的改进的BP算法[J]. 计算机与现代化,2006(3):6-8.
[9]	蒋奇. 管道缺陷漏磁检测量化技术及其应用研究[D].天津:天津大学,2002.