孔凡胜+王竹林
摘要:基于电子系统状态监测为研究背景,传统的Kernel Principal Component Analysis(核主成份分析法,简称KPCA)在状态监测过程中做数据特征降维处理,使得电路状态数据在消除冗余信息的同时,也能在相应的模型算法计算中很大程度的减少计算步骤,但是KPCA法的降维数据处理过程对数据样本贡献率的识别能力有不足之处,虽然达到了降维的目的,但是对特征样本数据的信息保留能力存在不足。本文中采用经验模态分解法(Empirical Mode Decomposition,简称EMD)对输出信号进行采集处理作为样本数据,设计基于Fisher准则的状态信息识别能力分析,采用Estimation of Distribution Algorithms(种群算法,简称EDA)对KPCA分析法进行改进研究,通过对数据处理,最大限度的保留状态主信息,使得在电路系统状态监测过程中减小实验误差,为后续故障预测打下基础。
关键词:KPCA;EDA;Fisher准则;EMD;信息识别;
中图分类号:TP 文献标识码:A
Electronic System Based on EDA Algorithm improve the KPCA Condition Monitoring
and Fault Prediction Research
Kong Fan-sheng , Wang Zhu-lin
(Ordnance Engineering College, Shi Jiazhuang , Hebei, 050003)
Abstract: Condition monitoring based on electronic system as the research background, the traditional Kernel Principal Component Analysis (Kernel Principal Component Analysis, KPCA) do in the process of condition monitoring data feature dimension reduction process, makes the circuit state data at the same time of eliminating redundant information, as well as the corresponding calculation model algorithm greatly reduces computation steps, but KPCA method of dimension reduction data processing for the contribution rate of the data sample inadequacies in the ability to recognize, though achieved the purpose of dimension reduction, but information on the characteristics of the sample data retention capability shortcomings.This article USES the method of Empirical Mode Decomposition (Empirical Mode Decomposition, the EMD) was carried out on the output signal as sample data collection and processing, design based on Fisher criterion of state information recognition ability analysis, the Estimation of Distribution Algorithms (population algorithm, referred to as EDA) to improve the KPCA analysis research, through the data processing, maximum retention state master information, make the circuit system decrease experimental error in the process of condition monitoring, fault prediction to lay the foundation for the follow-up.
Key word: KPCA; EDA; Fisher criterion; EMD;Information identification;
1 摘要
某型测角仪是装备训练的重要控制设备,主要对装备飞行过程中通过对误差信息的接收处理,及时输出调整信号到主控机,主控机输出控制指令,从而达到提高装备命中精度的功能。
基于对某型测角仪的状态监测与故障预测研究过程,选取一定的模型算法对设备的电子信号处理模块进行分析研究,通过对采集的数据进行提取降维等一系列算法处理,从而达到信息特征状态的提取分析,为下一步电子信号模块的状态监测与故障预测研究打下基础[1]。
2 研究内容
本文主要是针对某型测角仪TV4信号处理模块的状态监测与故障预测研究,采用HSMM为状态监测模型基础,通过EMD(经验模态分解)信号特征提取作为数据特征提取方法,应用KPCA做为数据特征降维处理,根据KPCA具有的局限性,采用EDA算法基于fisher准则进行改进处理,使得采用KPCA降维的同时最大限度保证数据主信息的完整性。
3 实验理论
3.1 KPCA分析法
本文是基于HSMM的电子系统信号处理模块研究,由于提取的特征信号具有冗余和高维的特点[2],若直接应用到实验中,会很大限度的降低状态监测能力,特征降维在于提取包含更多类别信息的状态特征,大幅度的消除特征的冗余性,提高状态监测能力。