文章快速检索 高级检索

1. 海军航空工程学院 研究生管理大队, 山东 烟台 264001;
2. 海军航空工程学院 科研部, 山东 烟台 264001

A new negative selection algorithm based on Extenics and its application in fault diagnosis
WEN Tianzhu1 , XU Aiqiang2, DENG Lu1
1. Graduate Student's Brigade, Naval Aeronautical and Astronautical University, Yantai 264001, China;
2. Department of Scientific Research, Naval Aeronautical and Astronautical University, Yantai 264001, China
Abstract:In this paper, the extension negative selection algorithm is proposed by fusing Extenics and negative selection algorithm, aiming at the problem that traditional diagnosis algorithm can hardly solve fault detection by using normal state data. The basic elements are adopted to describe the models of problem domain, detectors and training samples, the dependent function is used to define the affinity calculation formula, and the extension detector generation and optimization algorithm are designed. In the phase of extension detector generation, the mature detectors are taken through the self-tolerance and in the phase of extension detector optimization, less mature detectors are taken through merging the detectors. The influence of threshold value of the degree of affinity on the coverage rate and detection rate of detectors are discussed in the parameter analysis. Finally, the proposed algorithm is used for fault detection of an integrated display and control platform. The obtained mature detectors not only have less numbers and are non-redundant, but also have high detection rate. The results showed that the algorithm can solve the fault detection problem in the condition of no fault state data and the detection results are consistent with the practive.
Key words: extenics     negative selection algorithm     detector generation     detector optimization     fault diagnosis
o2

1 否定选择原理

 图 1 否定选择算法原理 Fig. 1 Principle of negative selection algorithm
2 可拓否定选择算法

2.1 模型的基元描述

ρ(vi,Vi)=ρ(vi,V)且vVi时，

D(vi,Vi,V)表示点vi与区间Vi和区间V组成区间套的位置。

A(s,d)≥0时表示检测器和样本匹配，当A(s,d)＜0时表示检测器和样本不匹配。

2.2 可拓检测器生成算法

1)初始化问题空间U，自己空间Us，将整个问题空间看成是一个未成熟检测器dy，此时分割级数c=1；

2)对未成熟检测器dy进行分割，得到检测器d1,d2,…,dm

3)对得到的每一个检测器dii=1,2,…,m，计算其与所有自己空间样本的亲和度A(s,d)，如果对于任意sUs，都满足A(s,d)＜δ，则将此检测器记为成熟检测器dmjj=1,2,…,m1，否则记为未成熟检测器dykk=1,2,…,m2，且m1+m2=m

4)判断分割级数c是否达到要求，如果是算法结束，否则分割级数c=c+1，对得到的每一个未成熟检测器dykk=1,2,…,m2，返回步骤2)。

2.3 检测器优化算法

1)对整个问题空间按照最大分割级数c进行最小单元分割，得到m个检测器d1,d2,…,dm

2)对得到的每一个检测器dii=1,2,…,m，如果存在成熟检测器dm，使得A(di,dm)>0，则记为成熟检测器dmjj=1,2,…,m1，否则记为未成熟检测器dykk=1,2,…,m2，且m1+m2=m

3)按照检测器标号的顺序，依此在特征向量x1,x2,…,xn的方向上对检测器进行合并，得到合并后的成熟检测器dmjj=1,2,…,m1。 4)重复步骤3)直到所有成熟检测器都无法合并为止。

2.4 参数分析

3 应用实例

1)测试数据标准化。

v=vmin时，v′=0；当v=vmax时，v′=1。vminvmax是根据测试需求确定的，当测量值v>vmax或者vvmin时，表示被测对象故障，当vmaxvvmin时，无法直接判断被测对象是否故障，需要使用算法进行故障检测，标准化后的测试数据，见表 1。前20组是正常状态数据用于生成成熟检测器，后10组数据中包含故障数据和正常数据各5组，用于故障检测。

 序号 v1 v2 v3 v4 v5 v6 状态 1 0.538 0.870 0.295 0.308 0.318 0.670 正常 2 0.537 0.885 0.303 0.315 0.334 0.670 正常 3 0.537 0.845 0.305 0.313 0.322 0.670 正常 4 0.514 0.975 0.295 0.280 0.188 0.670 正常 5 0.540 0.535 0.435 0.363 0.496 0.700 正常 6 0.538 0.720 0.395 0.360 0.484 0.700 正常 7 0.538 0.965 0.288 0.310 0.328 0.670 正常 8 0.540 0.850 0.300 0.310 0.320 0.670 正常 9 0.519 0.765 0.488 0.385 0.574 0.710 正常 10 0.542 0.555 0.413 0.370 0.512 0.705 正常 11 0.546 0.940 0.470 0.365 0.508 0.700 正常 12 0.523 0.630 0.435 0.455 0.812 0.760 正常 13 0.536 0.945 0.498 0.373 0.524 0.705 正常 14 0.519 0.780 0.465 0.380 0.554 0.710 正常 15 0.539 0.540 0.390 0.375 0.534 0.710 正常 16 0.517 0.780 0.615 0.350 0.452 0.690 正常 17 0.539 0.545 0.608 0.370 0.510 0.705 正常 18 0.509 0.770 0.625 0.370 0.528 0.705 正常 19 0.536 0.515 0.613 0.428 0.710 0.740 正常 20 0.534 0.510 0.610 0.428 0.710 0.740 正常 21 0.481 0.475 0.610 0.363 0.502 0.705 故障 22 0.189 0.555 0.615 0.385 0.586 0.715 故障 23 0.410 0.575 0.613 0.365 0.510 0.705 故障 24 0.538 1.420 0.293 0.305 0.302 0.670 故障 25 0.515 0.475 0.610 0.440 0.756 0.745 故障 26 0.517 0.770 0.293 0.330 0.388 0.685 正常 27 0.540 0.770 0.295 0.315 0.338 0.675 正常 28 0.517 0.750 0.298 0.310 0.316 0.670 正常 29 0.518 0.735 0.300 0.310 0.320 0.670 正常 30 0.540 0.760 0.300 0.318 0.344 0.675 正常

2)检测器生成。

3)检测器优化。

 序号 d1 d2 d3 d4 V1 [0,0.5] [0.5,1] [0.5,1] [0.5,1] V2 [0,1] [0,0.5] [0.5,1] [0.5,1] V3 [0,1] [0,1] [0,1] [0,1] V4 [0,1] [0,1] [0.5,1] [0,0.5] V5 [0,1] [0,1] [0,1] [0,1] V6 [0,1] [0,1] [0,1] [0,0.5]

4)抗原检测。

 序号 d1 d2 d3 d4 状态 21 0.343 -5.741 -17.864 -17.889 故障 22 0.382 -11.877 -11.943 -11.985 故障 23 0.357 -11.799 -11.855 -11.882 故障 24 ＋∞ -6.089 ＋∞ ＋∞ 故障 25 -5.704 0.235 -11.875 -11.959 故障 26 -5.746 -5.864 -5.809 -5.809 正常 27 -5.763 -5.864 -5.812 -5.801 正常 28 -5.755 -5.869 -5.817 -5.802 正常 29 -5.752 -5.863 -5.821 -5.806 正常 30 -5.759 -5.859 -5.806 -5.797 正常

4 结束语

 [1] DASGUPTAA D, YUA S, NINO F. Recent advances in artificial immune system: models and application[J]. Applied Soft Computing, 2011, 11(2): 1574-1587. [2] 李红芳, 张清华, 谢克明. 一种新型免疫学习算法在故障诊断中的应用[J]. 智能系统学报, 2008, 3(5): 449-454.LI Hongfang, ZHANG Qinghua, XIE Keming. Application of a novel immune network learning algorithm to fault diagnosis[J]. CAAI Transactions on Intelligence Systems, 2008, 3(5): 449-454. [3] 刘勇, 尚永爽, 王怡苹. 基于免疫模型的故障诊断方法及应用[J]. 计算机工程, 2011, 37(16): 5-7.LI Yong, SHANG Yongshuang, WANG Yiping. Fault diagnosis method based on immune model and its application[J]. Computer Engineering, 2011, 37(16): 5-7. [4] CHEN Wen, LI Tao, LIU Xiaojie, et al. A negative selection algorithm based on hierarchical clustering of self set[J]. Science China: Information Sciences, 2013, 56(8): 1-13. [5] CHEN Guangzhu, ZHANG Lei, BAO Jiusheng. An improved negative selection algorithm and its application in the fault diagnosis of vibrating screen by wireless sensor networks[J]. Journal of Computational and Theoretical Nanoscience, 2013, 10(10): 2418-2426. [6] GAO X Z, WANG X, ZENGER K. Motor fault diagnosis using negative selection algorithm[J]. Journal of Computing and Application, 2014, 25(1): 55-65. [7] 金章赞, 廖明宏, 肖明. 否定选择算法综述[J]. 通信学报, 2013, 34(1): 159-170.JIN Zhangzan, LIAO Minghong, XIAO Ming. Survey of negative selection algorithms[J]. Journal on Communications, 2013,34(1):159-170. [8] GONZALEZ F, DASGUPTA D, GOMEZ J. The effect of binary matching rules in negative selection[C]//Proceedings of the Genetic and Evolutionary Computation Conference. Chicago, USA, 2003: 195-206. [9] 杨春燕, 蔡文. 可拓工程[M]. 北京: 科学出版社, 2010: 18-97. [10] 向长城. 基于免疫网络算法关联函数经典域优化[J]. 湖北民族学院学报:自然科学版, 2009, 27(4): 141-146.XIANG Changcheng. Classical fields optimum of independent function based on artificial immune network algorithm[J]. Journal of Hubei University for Nationalities: Natural Science Edition, 2009, 27(4): 141-146. [11] 向长城, 黄席樾. 可拓免疫算法在汽轮机故障诊断中的应用[J]. 四川大学学报:工程科学版, 2008, 40(2): 141-146.XIANG Changcheng, HUANG Xiyue. Application of extenics immunity algorithm to turbo generator fault diagnosis[J]. Jouranl of Sichuan University: Engineering Science, 2008, 40(2): 141-146. [12] GONZALEZ F, DASGUPTA D, NINO L F. A randomized real-valued negative selection algorithm[C]//Proceedings of the 2nd International Conference on Artificial Immune Systems. Edinburgh, UK, 2003: 261-272. [13] GONZALEZ F, DASGUPTA D, KOZMA R. Combining negative selection and classification techniques for anomaly detection[C]//Proceedings of the 2002 Congress on Evolutionary Computation. Honolulu, USA, 2002: 705-710. [14] AYARA M, TIMMIS J, De LEMOS R, et al. Negative selection: how to generate detectors[C]//Proceedings of 1st International Conference on Artificial Immune Systems. Canterbury, UK, 2002, 1: 89-98. [15] 徐学邈, 王如根, 侯胜利. 基于反面选择原理的智能融合故障检测模型及其应用[J]. 系统工程与电子技术, 2009, 31(8): 2029-2032.XU Xuemiao, WANG Rugen, HOU Shengli. Intelligence fusion approach to fault detection based on negative selection principle and its application[J]. Systems Engineering and Electronics, 2009, 31(8): 2029-2032.
DOI: 10.3969/j.issn.1673-4785.201402020

0

#### 文章信息

WEN Tianzhu, XU Aiqiang, DENG Lu

A new negative selection algorithm based on Extenics and its application in fault diagnosis

CAAI Transactions on Intelligent Systems, 2015, 10(03): 488-493.
DOI: 10.3969/j.issn.1673-4785.201402020