Modern Defense Technology ›› 2022, Vol. 50 ›› Issue (2): 84-95.DOI: 10.3969/j.issn.1009-086x.2022.02.012

• INTEGRATED LOGISTICS SUPPORT TECHNOLOGY • Previous Articles     Next Articles

Evaluation of Test Equipment Maintenance Ability Based on GA-Wavelet-BP Neural Network

Li-xing ZHAO, Xing-ming HOU, Zhao-wen XU, Lin-zi HE   

  1. University of Aerospace Engineering,Department of Aerospace Support,Beijing 102206,China
  • Received:2021-09-06 Revised:2021-11-23 Online:2022-04-28 Published:2022-04-29

基于GA-小波-BP神经网络的装备维修能力评估

赵黎兴, 侯兴明, 徐兆文, 和林子   

  1. 航天工程大学 航天保障系,北京 102206
  • 作者简介:赵黎兴(1984-),男,甘肃平凉人。硕士生,主要从事装备维修保障研究。通信地址:102206 北京市昌平区沙河镇小沙河村 215 号北门 E-mail:454667353@qq.com

Abstract:

Aiming at the characteristics of strong subjectivity and weak adaptability of traditional test equipment maintenance support capability evaluation, a maintenance capability evaluation method based on error back propagation (BP) feedforward neural network is proposed based on test equipment maintenance records as sample data. In order to eliminate the attribute redundancy of the maintenance records, the rough set principal component reduction algorithm is selected to reduce the sample attributes. In order to avoid the influence of the local convergence and the slow convergence speed of the BP neural network due to the local extremes,the initial weights and thresholds of neural network are optimized by using the global search ability of genetic algorithm (GA). In order to increase the learning function of the network and improve the network training rate and training accuracy, the wavelet function is selected as the transfer function of the hidden layer, and the Sigmoid function is the activation function of the output layer. It is shown that the performance of the BP neural network is improved through redundant attribute reduction, and the weight/threshold value of the BP neural network and the wavelet function are optimized by GA as the transfer function of the hidden layer, which avoids local convergence and improves the training rate of the network. The accuracy of the evaluation reduces the error.

Key words: maintenance capability evaluation, principal component reduction, genetic algorithm (GA), back propagation (BP) neural network, wavelet function

摘要:

针对传统装备维修保障能力评估主观性强、适应性弱的特点,提出了一种以装备维修记录为样本数据的基于误差反向传播(back propagation,BP)前馈神经网络的维修能力评估方法。为消除维修记录属性冗余,选用粗糙集主分量约简算法,对样本属性进行约简,为避免BP神经网络因局部极值导致局部收敛和收敛速度过慢的影响,利用遗传算法(genetic algorithm,GA)的全局搜索能力对神经网络初始权值和阈值进行了优化,为增加网络的学习功能,提高网络训练速率和训练精度,选取小波函数作为隐含层的传递函数,Sigmoid函数为输出层激活函数。研究表明:通过冗余属性约简提高了BP神经网络的性能,利用GA优化BP神经网络权值/阈值和小波函数作为隐含层的传递函数,避免了局部收敛,提高了网络的训练速率和评估的精度,降低了误差。

关键词: 维修能力评估, 主分量约简, 遗传算法, BP神经网络, 小波函数

CLC Number: