大型IP网络时延的主成分分析

doi:10.11959/j.issn.1000-0801.2021059

摘要/Abstract

摘要：

网络时延是评估网络性能的关键指标之一。主成分分析（PCA）是数据挖掘领域常用的一种多变量分析和降维算法。通过对大型 IP 网络时延进行 PCA 分析，旨在挖掘网络时延的深层原因及网络各节点间的相互依赖关系，并搭建一个科学合理的网络时延评价体系，最终得到IP网络建设、优化改造的有效建议。对历史网络时延进行离线分析只是主成分分析方法的一种初步应用，今后可结合网络拓扑结构、现网流量流向、路由、距离等相关因素，将主成分分析方法应用到针对网络流量、网络时延、网络丢包等网络性能的实时在线监测分析中，进一步提升网络运营的效率和质量。

关键词: 网络时延, 主成分分析, 相关系数矩阵, 方差贡献率

Abstract:

Network time delay is one of the key indexes to evaluate network performance.Principal component analysis (PCA) is a kind of multivariable analysis and declination algorithm commonly used in the field of data mining.Based on PCA analysis of time delay in large IP networks, aiming to find out the deep reason of time delay and the interdependencies among nodes of the network, a scientific and reasonable network time delay evaluation system was built, and effective suggestions for IP network construction and optimization were finally got.The off-line analysis of the historical network delay is only a preliminary application of the PCA.In the future, PCA can be applied to the real-time on-line monitoring and analysis of the network performance, such as network traffic, network delay, network packet loss, etc., in combination with the network topology, current network traffic direction, routing, distance and other related factors, thus the efficiency and quality of network operations can be further improved.

Key words: network time delay, principal component analysis, coefficient matrix, proportion of variance

中图分类号:

TN915

韦烜, 黄晓莹. 大型IP网络时延的主成分分析[J]. 电信科学, 2021, 37(4): 62-72.

Xuan WEI, Xiaoying HUANG. Principal component analysis of time delay in large IP network[J]. Telecommunications Science, 2021, 37(4): 62-72.

图/表 14

图1

表1

旧平面标准化后的时延矩阵"

ID	X1	X2	X3	X4	X5	X6	X7	X8	X9
A1	-1.508 53	-1.307 03	-0.354 17	0.000 783	-0.691 53	1.156 154	-0.371 68	-0.196 41	-0.277 24
A2	-1.298 51	-1.489 23	0.017 812	0.259 75	0.173 263	0.638 81	0.010 35	0.361 49	0.104 387
A3	0.395 988	-0.003 4	-0.265 32	-2.103 84	0.189 925	-0.54925	-0.679 14	-0.034 48	-0.344 64
A4	-0.619 12	-0.384 27	-0.272 43	-0.548 64	-2.113 7	-1.218 25	-0.501 8	-0.185 44	-0.338 13
A5	1.330 479	0.517 775	0.205 578	-0.242 63	-1.268 9	-2.089 26	0.062 879	0.367 232	-0.063 88
A6	-0.703 44	-0.484 55	-0.970 18	-1.31 527	-1.007 29	-0.509 19	-1.602 12	-1.117 89	-0.903 83
A7	-0.236 73	-0.111 21	-1.553 62	-0.440 58	-0.555 73	0.160 382	-0.737 84	-1.691 46	-1.053 97
A8	-0.226 65	0.962 677	0.085 337	1.182 025	-1.441 36	-0.794 18	-0.196 16	0.133 729	0.615 7
A9	0.618 208	0.968 327	0.561 564	-0.854 65	0.481 523	-1.023 67	-1.101 59	0.160 892	0.234 075
A10	1.822 121	1.736 196	0.872 533	-0.468 4	-0.263 47	-1.237 13	0.710 479	1.005 578	1.308 761
A11	0.027 92	0.346 876	-1.539 4	-0.722 51	-0.004 2	0.318 904	-0.523 59	-1.114 23	-1.564 82
A12	-1.346 77	-1.054 68	-0.151	0.806 467	-0.518 24	0.137 491	-0.728 16	-0.312 9	0.627 321
A13	-0.489 18	-0.678 52	0.761 176	0.630 458	1.731 228	1.509 824	0.890 233	0.930 355	0.991 747
A14	-0.129 6	-0.107 44	1.180 541	1.241 943	1.779 55	1.452 024	1.471 257	1.427 137	1.595 561
A15	-0.002 31	0.419 379	-0.687 64	0.548 071	-0.279 96	-0.402 17	-0.855 86	-0.657 15	-1.083 72
A16	-0.847 17	-1.016 55	0.415 26	0.240 456	0.855 601	1.164 783	0.420 572	0.280 517	0.635 686
A17	-0.224 53	0.067 9	-0.134 41	0.083 706	-0.017 53	0.009 3	-0.988 41	-0.397 53	-0.772 28
A18	-0.314 16	-0.467 6	-1.862 81	-0.655 11	-0.265 8	-0.087 99	-0.461 86	-1.402 59	-1.460 69
A19	1.712 337	0.890 646	0.709 052	-0.641 2	-0.794 01	-1.548 45	0.621 51	0.108 132	0.479 04
A20	-0.392 12	-1.223 23	-0.769 98	-0.723 46	0.526 513	0.915 223	0.222 056	-0.406 93	-0.688 61
A21	-0.828 61	-0.011 87	-0.134 41	0.759 927	-1.362 21	-0.569 85	0.012 645	0.151 49	-0.194 5
A22	1.267 897	2.146 73	1.886 589	2.431 744	0.961 41	1.082 901	2.672 039	2.533 534	2.149 638
A23	1.079 62	1.886 85	2.425 01	2.192 072	1.898 688	1.545 878	2.308 9	2.227 943	1.699 682
A24	2.220 42	1.250 334	1.007 582	0.4779 88	0.214 086	-1.045 42	0.556 75	0.749 613	0.469 279
A25	-0.335 91	-0.760 91	-1.342 75	-0.326 63	-0.249 14	-0.266 54	-0.907 91	-1.326 32	-1.168 78
A26	0.685 033	-0.077 78	-0.524 16	-1.084 16	1.158 03	0.128 334	0.242 634	-0.491 04	-1.142 75
A27	-0.995 14	-0.835 29	0.282 58	-0.724 65	0.085 783	-0.006 72	-1.030 17	-0.637 3	-0.545 44
A28	-0.661 55	-1.044 33	0.151 677	0.215 846	0.781 452	1.128 112	0.504 7	0.257 01	0.692 397

表1

表2

新平面标准化后时延矩阵"

ID	X1	X2	X3	X4	X5	X6	X7	X8	X9
A1	-1.425 06	-1.629 89	-0.375 77	-0.193 08	-0.836 54	-0.191 69	-0.469 21	-0.265 49	-0.054
A2	-1.282 4	-1.810 22	-0.043 51	0.029 081	-0.697 38	-0.106 13.	-0.367 67	-0.170 61	-0.071 61
A3	0.083 615	0.309 951	-0.337 13	-1.652 64	1.117 554	0.041 056	-0.419 52	-0.448 74	-0.180 78
A4	-0.599 39	-0.67 248	-0.398 52	0.133 1	-2.211 62	-0.918 31	0.578 044	-0.332 89	-0.204 55
A5	-0.115	-0.0746	0.024 315	-0.376 17	-1.241 78	-1.453 41	-0.154 33	0.106 732	-0.060 17
A6	-0.547 56	-0.639 49	-0.805 9	-0.827 54	0.225 592	-0.556 11	-1.434 92	-0.947 68	-0.908
A7	-0.248 94	-0.189 51	-1.186 67	-0.572 91	-0.421 93	0.063 222	-0.884 01	-1.362 83	-1.147 47
A8	0.491 402	0.666 631	0.587 095	0.573 65	-1.480 46	-0.801 72	0.201 593	0.430 869	0.137 926
A9	0.490 485	0.676 302	0.241 529	1.227 15	0.260 924	-0.662 07	-0.059 28	0.097 716	0.256 5
A10	-0.063 63	0.461 839	0.093 429	-0.834 6	-0.683 68	-1.089 44	-0.138 13	0.416 272	0.251 939
A11	0.031 323	-0.021 13	-1.197 4	0.363 46	-0.174 6	-0.059 14	-0.800 84	-1.073 04	-1.450 33
A12	-1.217 73	-1.452 4	-0.294 64	-0.316 86	0.406 58	0.182 478	-0.658 25	-0.432 06	-0.300 52
A13	-0.355 82	-0.377 24	0.597 827	0.864 526	0.919 981	0.896 684	0.804 347	0.708 21	1.014 813
A14	-0.263 16	-0.200 89	0.690 55	0.988 784	1.086 548	1.001 753	0.946 393	0.797 938	1.153 918
A15	0.0634 32	0.087 524	-0.959 58	-0.640 69	0.082 82	-0.337 11	-0.866 73	-1.041 27	-1.181 81
A16	-0.787 46	-1.162 8 5	0.189 157	0.495 518	0.071 283	2.251 502	0.215 096	0.169 413	0.438 585
Z17	-0.385 18	-0.411 94	-0.632 05	-0.758 39	0.420 28	-0.394 74	-0.948 82	-0.702 11	-0.644 32
A18	-0.296 65	0.152 37 5	-1.428 78	-0.472 19	-0.424 81	0.061 892	-0.637 19	-1.129 71	-1.212 18
A19	-0.240 69	0.179 112	-0.067 55	-0.538 08	-1.003 83	-1.243 27	-0.501 62	-0.701 87	-0.069 41
A20	-0.905 81	-0.498 41	-0.559 5	-0.063 64	0.500 319	0.017 559	0.068 188	-0.694 81	-0.772 42
A21	-0.100 78	-0.620 9	0.059 516	0.764 273	-1.185 54	-0.449 71	0.145 422	-0.004 46	0.236 531
A22	1.323 031	1.876 042	1.355 068	1.768 219	0.584 684	-0.045 84	1.588 574	1.325 572	1.328 239
A23	2.827 118	2.203 71	3.127 975	3.223 071	2.822 161	2.413 318	3.746 281	2.899 031	2.870 716
A24	2.501 439	0.982 921	2.003 703	0.581 181	0.454 892	-0.830 53	0.200 513	2.166 181	1.930 439
A25	-0.153 53	0.074 44	-1.076 8	-0.664 22	-0.225 8	-0.054 26	-0.965 03	-1.252 07	-1.094 65
A26	0.680 846	0.745 135	-0.720 48	-1.063 82	1.061 311	0.662 605	-0.280 72	-0.468 13	-0.725 32
A27	-0.766 36	-1.124 73	0.526 997	0.852 759	0.105 894	-0.654 97	0.500 809	0.829 278	0.412 173
A28	1.262 482	1.911 881	0.618 003	0.291 246	0.467 15	2.256 379	0.591 006	0.490 544	0.558 761

表2

表3

表4

图2

表5

表6

图3

图4

图5

图6

图7

图8

参考文献 6

[1]	陈小峰, 赵雅迪, 张利鹏 ,等. 电力潜在敏感客户预测的大数据方法应用[J]. 电信科学, 2019,35(11): 117-124.
	CHEN X F , ZHAO Y D , ZHANG L P ,et al. Application of big data method in forecasting potential sensitive customers of electric power[J]. Telecommunications Science, 2019,35(11): 117-124.
[2]	许方敏, 伍丽娇, 杨帆 ,等. 时间敏感网络（TSN）及无线TSN技术[J]. 电信科学, 2020,36(8): 81-91.
	XU F M , WU L J , YANG F ,et al. Time-sensitive networking (TSN) and wireless TSN technology[J]. Telecommunications Science, 2020,36(8): 81-91.
[3]	李昊奇, 应娜, 郭春生 ,等. 基于深度信念网络和线性单分类SVM的高维异常检测[J]. 电信科学, 2018,34(1): 34-42.
	LI Q H , YING N , GUO C S ,et al. High-dimensional outlier detection based on deep belief network and linear one-class SVM[J]. Telecommunications Science, 2018,34(1): 34-42.
[4]	杨明极, 刘恺怿, 邵丹 . 用于WLAN室内定位的PCA聚类算法[J]. 电信科学, 2016,32(7): 21-26.
	YANG M J , LIU K Y , SHAO D . PCA clustering algorithm for indoor positioning in WLAN[J]. Telecommunications Science, 2016,32(7): 21-26.
[5]	林川, 赵海, 毕远国 ,等. 互联网网络时延特征研究[J]. 通信学报, 2015,36(3): 149-160.
	LIN C , ZHAO H , BI Y G ,et al. Reasearch on network delay of Internet[J]. Journal on Communications, 2015,36(3): 149-160.
[6]	牟澄 . 互联网流量特征智能提取关键技术研究[D]. 北京:北京邮电大学, 2014.
	MU C . Research on key technologies of intelligent extraction of Internet traffic characteristics[D]. Beijing:Beijing University of Posts and Telecommunications, 2014.

	X1	X2	X3	X4	X5	X6	X7	X8	X9
X1	1.000 000 0	0.844 984 45	0.455 616 1	0.081 517 4	0.150 223 24	-0.431 343 8	0.453 992 8	0.465 268 5	0.318 890 0
X2	0.844 984 4	1.000 000 0	0.544 895 8	0.366 292 6	0.076 777 72	-0.326 413 9	0.480 077 5	0.577 083 7	0.441 787 1
X3	0.455 616 1	0.544 895 75	1.000 000 0	0.617 904 4	0.464 248 96	0.182 207 3	0.766 221 69	0.943 872 5	0.909 731 5
X4	0.081 517 4	0.366 292 60	0.617 904 4	1.000 000 0	0.314 794 02	0.441 670 4	0.694 922 9	0.687 131 2	0.679 282 8
X5	0.150 223 2	0.076 777 72	0.464 249 0	0.314 794 0	1.000 000 0	0.719 740 6	0.572 707 5	0.482 343 5	0.413 956 0
X6	-0.431 343 8	-0.326 413 92	0.182 207 3	0.441 670 4	0.719 740 56	1.000 000 0	0.428 194 2	0.257 572 2	0.269 500 4
X7	0.453 992 8	0.480 077 49	0.762 216 9	0.694 922 9	0.572 707 47	0.428 194 2	1.000 000 0	0.866 955 1	0.787 833 5
X8	0.465 268 5	0.577 083 66	0.943 872 5	0.687 131 2	0.482 343 55	0.257 572 2	0.866 955 1	1.000 000 0	0.919 055 0
X9	0.318 890 0	0.441 787 05	0.909 731 5	0.679 282 8	0.413 955 99	0.269 500 4	0.787 833 5	0.919 055 0	1.000 000

	X1	X2	X3	X4	X5	X6	X7	X8	X9
X1	1.000 000 0	0.905 770 7	0.686 749 3	0.455 820 7	0.479 148 5	0.263 908 4	0.592 238 0	0.672 006 1	0.603 464 1
X2	0.905 770 7	1.000 000 0	0.505 781 3	0.321 450 0	0.401 916 1	0.243 530 9	0.521 176 2	0.487 740 6	0.426 389 2
X3	0.686 749 3	0.505 781 3	1.000 000 0	0.791 016 7	0.444 873 4	0.321 905 0	0.860 865 1	0.986 730 0	0.973 379 9
X4	0.455 820 7	0.321 450 0	0.791 016 7	1.000 000 0	0.327 240 7	0.458 370 9	0.875 350 2	0.780 197 3	0.799 051 1
X5	0.479 148 5	0.401 916 1	0.444 873 4	0.327 240 7	1.000 000 0	0.650 277 9	0.466 149 9	0.417 715 9	0.433 087 0
X6	0.263 908 4	0.243 530 9	0.321 905 0	0.458 370 9	0.650 277 9	1.000 000 0	0.500 610 0	0.284 164 6	0.369 065 4
X7	0.592 238 0	0.521 176 2	0.860 865 1	0.875 350 2	0.466 149 9	0.500 610 0	1.000 000 0	0.851 804 8	0.858 474 1
X8	0.672 006 1	0.487 740 6	0.986 730 0	0.780 197 3	0.417 715 9	0.284 164 6	0.851 804 8	1.000 000 0	0.974 949 9
X9	0.603 464 1	0.426 389 2	0.973 379 9	0.799 051 1	0.433 087 0	0.369 065 4	0.858 474 1	0.974 949 9	1.000 000

对比项	Comp.1	Comp.2	Comp.3	Comp.4	Comp.5	Comp.6	Comp.7	Comp.8	Comp.9
标准差	2.261 982 9	1.469 033 9	0.904 578 71	0.684 403 32	0.458 152 71	0.318 612 65	0.269 466 744	0.181 465 553	0.147 453 193
方差贡献率	0.568 507 4	0.239 784 5	0.090 918 07	0.052 045 32	0.023 322 66	0.011 279 34	0.008 068 036	0.003 658 861	0.002 415 827
方差累计贡献率	0.568 5074	0.808 291 9	0.899 209 96	0.951 255 28	0.974 577 94	0.985 857 28	0.993 925 312	0.997 584 173	1.000 000 0

对比项	Comp.1	Comp.2	Comp.3	Comp.4	Comp.5	Comp.6	Comp.7	Comp.8	Comp.9
标准差	2.416 156 0	1.150 895 6	1.065 301 0	0.660 799 80	0.435 937 01	0.342 479 85	0.203 853 186	0.134 152 915	0.108 672 994
方差贡献率	0.648 645 5	0.150 663 4	0.110 096 3	0.048 517 37	0.021 115 68	0.013 032 49	0.004 617 347	0.001 999 667	0.001 312 202
方差累计贡献率	0.648 645 5	0.804 330 90	0.909 405 2	0.957 922 61	0.979 038 29	0.992 070 78	0.996 688 131	0.998 687 798	1.000 000 000