《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (5): 1342-1348.DOI: 10.11772/j.issn.1001-9081.2022030429
所属专题: 第九届中国数据挖掘会议(CCDM 2022)
收稿日期:
2022-03-17
修回日期:
2023-02-03
接受日期:
2023-02-06
发布日期:
2023-05-08
出版日期:
2023-05-10
通讯作者:
鲁淑霞
作者简介:
吕新伟(1997—),男,山东济宁人,硕士研究生,主要研究方向:机器学习基金资助:
Xinwei LYU1,2, Shuxia LU1,2()
Received:
2022-03-17
Revised:
2023-02-03
Accepted:
2023-02-06
Online:
2023-05-08
Published:
2023-05-10
Contact:
Shuxia LU
About author:
LYU Xinwei, born in 1997, M. S. candidate. His research interests include machine learning.Supported by:
摘要:
极限学习机(ELM)的许多变体都致力于提高ELM对异常点的鲁棒性,而传统的鲁棒极限学习机(RELM)对异常点非常敏感,如何处理数据中的过多极端异常点变成构建RELM模型的棘手问题。对于残差较大的异常点,采用有界损失函数消除异常点对模型的污染;为了解决异常点过多的问题,采用迭代修正技术修改数据以降低由异常点过多带来的影响。结合这两种方法,提出迭代修正鲁棒极限学习机(IMRELM)。IMRELM通过迭代的方式求解,在每次的迭代中,通过对样本重加权减小异常点的影响,在不断修正的过程中避免算法出现欠拟合。在具有不同异常点水平的人工数据集和真实数据集上对比了IMRELM、ELM、加权极限学习机(WELM)、迭代重加权极限学习机(IRWELM)和迭代重加权正则化极限学习机(IRRELM)。在异常点占比为80%的人工数据集上,IRRELM的均方误差(MSE)为2.450 44,而IMRELM的MSE为0.000 79。实验结果表明,IMRELM在具有过多极端异常点的数据上具有良好的预测精度和鲁棒性。
中图分类号:
吕新伟, 鲁淑霞. 迭代修正鲁棒极限学习机[J]. 计算机应用, 2023, 43(5): 1342-1348.
Xinwei LYU, Shuxia LU. Iteratively modified robust extreme learning machine[J]. Journal of Computer Applications, 2023, 43(5): 1342-1348.
异常点水平/% | ELM(MSE | WELM(MSE | IRWELM(MSE | IRRELM(MSE | IMRELM(MSE |
---|---|---|---|---|---|
序值 | 4.5 | 4.1 | 3.1 | 2.2 | 1 |
0 | 0.000 20 | 0.000 23 | 0.000 22 | 0.000 21 | 0.000 20 |
10 | 0.117 92 | 0.000 71 | 0.000 69 | 0.000 21 | 0.000 21 |
20 | 0.319 12 | 0.076 03 | 0.005 01 | 0.000 23 | 0.000 22 |
30 | 0.945 99 | 0.141 36 | 0.029 90 | 0.000 71 | 0.000 22 |
40 | 1.429 58 | 0.561 76 | 0.249 08 | 0.006 76 | 0.000 23 |
50 | 1.937 54 | 1.405 65 | 0.600 57 | 0.096 97 | 0.000 24 |
60 | 2.647 86 | 2.068 85 | 1.250 01 | 0.729 21 | 0.000 26 |
70 | 3.474 22 | 2.731 50 | 1.780 44 | 1.346 05 | 0.000 41 |
80 | 4.195 85 | 3.934 42 | 2.481 35 | 2.450 44 | 0.000 79 |
表1 具有不同异常点水平的人工数据集上的实验结果
Tab. 1 Experimental results on synthetic datasets with different outlier levels
异常点水平/% | ELM(MSE | WELM(MSE | IRWELM(MSE | IRRELM(MSE | IMRELM(MSE |
---|---|---|---|---|---|
序值 | 4.5 | 4.1 | 3.1 | 2.2 | 1 |
0 | 0.000 20 | 0.000 23 | 0.000 22 | 0.000 21 | 0.000 20 |
10 | 0.117 92 | 0.000 71 | 0.000 69 | 0.000 21 | 0.000 21 |
20 | 0.319 12 | 0.076 03 | 0.005 01 | 0.000 23 | 0.000 22 |
30 | 0.945 99 | 0.141 36 | 0.029 90 | 0.000 71 | 0.000 22 |
40 | 1.429 58 | 0.561 76 | 0.249 08 | 0.006 76 | 0.000 23 |
50 | 1.937 54 | 1.405 65 | 0.600 57 | 0.096 97 | 0.000 24 |
60 | 2.647 86 | 2.068 85 | 1.250 01 | 0.729 21 | 0.000 26 |
70 | 3.474 22 | 2.731 50 | 1.780 44 | 1.346 05 | 0.000 41 |
80 | 4.195 85 | 3.934 42 | 2.481 35 | 2.450 44 | 0.000 79 |
数据集 | 样本数 | 特征数 | 最佳 神经元数 | 数据集 | 样本数 | 特征数 | 最佳 神经元数 | ||
---|---|---|---|---|---|---|---|---|---|
训练集 | 测试集 | 训练集 | 测试集 | ||||||
Fatbody | 166 | 86 | 12 | 250 | Abalone | 2 784 | 1 393 | 8 | 210 |
Pollution | 260 | 130 | 16 | 70 | CO and NO | 24 489 | 12 244 | 11 | 240 |
Machine-CPU | 680 | 340 | 7 | 200 | CCPP (Combined Cycle Power Plant) | 6 379 | 3 189 | 5 | 120 |
Concrete Compressive | 680 | 340 | 9 | 580 | Forest Fires Data Set | 345 | 172 | 13 | 600 |
AutoMPG | 266 | 130 | 8 | 135 | Airfoil Self-Noise Data | 1 002 | 501 | 6 | 160 |
Residential-Building | 250 | 125 | 109 | 600 | Bias modified | 5 167 | 2 583 | 25 | 370 |
表2 真实数据集
Tab. 2 Real datasets
数据集 | 样本数 | 特征数 | 最佳 神经元数 | 数据集 | 样本数 | 特征数 | 最佳 神经元数 | ||
---|---|---|---|---|---|---|---|---|---|
训练集 | 测试集 | 训练集 | 测试集 | ||||||
Fatbody | 166 | 86 | 12 | 250 | Abalone | 2 784 | 1 393 | 8 | 210 |
Pollution | 260 | 130 | 16 | 70 | CO and NO | 24 489 | 12 244 | 11 | 240 |
Machine-CPU | 680 | 340 | 7 | 200 | CCPP (Combined Cycle Power Plant) | 6 379 | 3 189 | 5 | 120 |
Concrete Compressive | 680 | 340 | 9 | 580 | Forest Fires Data Set | 345 | 172 | 13 | 600 |
AutoMPG | 266 | 130 | 8 | 135 | Airfoil Self-Noise Data | 1 002 | 501 | 6 | 160 |
Residential-Building | 250 | 125 | 109 | 600 | Bias modified | 5 167 | 2 583 | 25 | 370 |
数据集 | 异常点 水平/% | ELM (MSE | WELM (MSE | IRWELM (MSE | IRRELM (MSE | IMRELM (MSE |
---|---|---|---|---|---|---|
Fatbody | 0 | 0.006 5 | 0.006 7 | 0.006 8 | 0.006 5 | 0.006 4 |
10 | 1.693 3 | 0.647 8 | 0.219 9 | 0.038 3 | 0.007 8 | |
20 | 2.406 3 | 2.164 9 | 1.046 9 | 0.091 7 | 0.009 4 | |
30 | 3.184 8 | 4.131 1 | 3.481 6 | 0.630 6 | 0.031 5 | |
40 | 6.271 8 | 5.292 7 | 4.840 1 | 1.115 0 | 0.206 2 | |
Pollution | 0 | 0.004 9 | 0.005 2 | 0.004 9 | 0.005 0 | 0.005 1 |
10 | 4.524 9 | 2.671 9 | 2.001 8 | 0.523 8 | 0.025 3 | |
20 | 7.289 4 | 5.956 7 | 5.155 2 | 2.650 6 | 1.007 5 | |
30 | 8.261 1 | 8.449 5 | 6.764 9 | 3.984 7 | 2.342 9 | |
40 | 12.219 1 | 9.569 1 | 7.839 2 | 5.664 7 | 2.764 5 | |
Machine-CPU | 0 | 0.001 8 | 0.002 2 | 0.002 3 | 0.001 9 | 0.001 8 |
10 | 1.724 3 | 0.709 8 | 0.426 8 | 0.168 4 | 0.004 8 | |
20 | 2.279 0 | 2.201 3 | 1.773 7 | 0.673 8 | 0.010 7 | |
30 | 2.551 9 | 2.512 7 | 2.476 3 | 1.419 8 | 0.020 8 | |
40 | 5.653 4 | 5.542 5 | 5.452 9 | 3.051 8 | 0.048 0 | |
Concrete Compressive | 0 | 0.010 1 | 0.010 3 | 0.010 1 | 0.010 3 | 0.010 2 |
10 | 0.552 0 | 0.137 7 | 0.013 8 | 0.015 8 | 0.011 5 | |
20 | 0.724 9 | 0.535 6 | 0.437 5 | 0.031 9 | 0.025 2 | |
30 | 0.965 1 | 0.821 4 | 0.630 3 | 0.062 9 | 0.015 9 | |
40 | 1.396 0 | 1.175 9 | 0.850 9 | 0.118 8 | 0.051 9 | |
AutoMPG | 0 | 0.071 0 | 0.075 2 | 0.091 1 | 0.080 0 | 0.102 3 |
10 | 1.753 8 | 1.481 9 | 0.312 3 | 0.126 1 | 0.115 2 | |
20 | 1.591 5 | 0.943 5 | 0.871 6 | 0.192 3 | 0.117 5 | |
30 | 1.823 0 | 1.494 0 | 1.417 3 | 0.706 8 | 0.152 2 | |
40 | 2.811 6 | 2.260 7 | 2.270 2 | 1.037 1 | 0.338 3 | |
Abalone | 0 | 0.007 2 | 0.007 4 | 0.007 9 | 0.007 3 | 0.007 2 |
10 | 0.935 6 | 0.299 8 | 0.180 8 | 0.040 5 | 0.007 7 | |
20 | 1.243 5 | 0.941 5 | 0.665 4 | 0.013 6 | 0.010 7 | |
30 | 1.431 5 | 1.391 1 | 1.026 3 | 0.206 9 | 0.014 3 | |
40 | 1.770 4 | 1.746 2 | 1.771 8 | 0.756 4 | 0.065 3 | |
Residential-Building | 0 | 0.003 4 | 0.004 5 | 0.005 7 | 0.003 5 | 0.003 7 |
10 | 0.851 4 | 0.857 7 | 0.015 3 | 0.004 3 | 0.004 3 | |
20 | 1.993 2 | 1.891 4 | 2.559 5 | 0.123 3 | 0.008 5 | |
30 | 2.644 7 | 2.366 0 | 2.054 4 | 0.194 4 | 0.024 3 | |
40 | 2.761 8 | 2.734 3 | 2.961 6 | 1.308 3 | 0.032 3 | |
CO and NO | 0 | 0.002 1 | 0.002 3 | 0.002 6 | 0.002 2 | 0.002 1 |
10 | 0.790 4 | 0.198 2 | 0.004 7 | 0.008 3 | 0.006 5 | |
20 | 1.234 1 | 1.074 0 | 0.745 8 | 0.100 3 | 0.014 1 | |
30 | 1.807 3 | 1.738 2 | 1.884 2 | 0.491 2 | 0.028 9 | |
40 | 2.250 2 | 2.356 7 | 2.533 9 | 0.721 4 | 0.065 2 | |
CCPP | 0 | 0.002 8 | 0.002 9 | 0.002 9 | 0.002 9 | 0.002 9 |
10 | 0.681 2 | 0.003 3 | 0.003 1 | 0.003 2 | 0.003 0 | |
20 | 0.909 5 | 0.373 9 | 0.053 2 | 0.003 7 | 0.003 8 | |
30 | 1.019 5 | 1.818 5 | 2.073 3 | 0.045 0 | 0.021 5 | |
40 | 1.328 4 | 2.138 6 | 2.708 8 | 0.168 9 | 0.043 5 | |
Forest Fire Data Set | 0 | 0.001 8 | 0.002 1 | 0.001 9 | 0.002 1 | 0.002 0 |
10 | 0.681 0 | 0.075 5 | 0.002 5 | 0.003 4 | 0.002 7 | |
20 | 1.120 9 | 1.098 1 | 0.061 1 | 0.005 2 | 0.003 2 | |
30 | 1.387 9 | 1.298 1 | 0.757 8 | 0.009 8 | 0.005 3 | |
40 | 1.789 9 | 1.501 6 | 1.342 6 | 0.308 8 | 0.025 9 | |
Airfoil Self-Noise Data | 0 | 0.005 8 | 0.006 1 | 0.006 2 | 0.005 9 | 0.005 8 |
10 | 0.381 0 | 0.075 5 | 0.010 8 | 0.010 6 | 0.012 6 | |
20 | 1.643 8 | 0.875 4 | 0.971 4 | 0.270 8 | 0.066 7 | |
30 | 1.895 4 | 1.450 5 | 1.236 9 | 0.653 3 | 0.073 7 | |
40 | 2.362 2 | 1.755 3 | 1.860 1 | 0.934 5 | 0.012 1 | |
Bias modified | 0 | 0.006 1 | 0.006 2 | 0.006 0 | 0.006 3 | 0.006 2 |
10 | 1.059 9 | 0.500 7 | 0.016 9 | 0.018 0 | 0.008 3 | |
20 | 0.810 2 | 0.585 8 | 0.295 1 | 0.062 3 | 0.022 3 | |
30 | 1.810 0 | 1.138 2 | 0.834 6 | 0.284 0 | 0.062 4 | |
40 | 1.838 5 | 1.778 4 | 1.740 2 | 0.509 6 | 0.177 2 |
表3 具有不同异常点水平的真实数据集上的实验结果
Tab. 3 Experimental results on real datasets with different outlier levels
数据集 | 异常点 水平/% | ELM (MSE | WELM (MSE | IRWELM (MSE | IRRELM (MSE | IMRELM (MSE |
---|---|---|---|---|---|---|
Fatbody | 0 | 0.006 5 | 0.006 7 | 0.006 8 | 0.006 5 | 0.006 4 |
10 | 1.693 3 | 0.647 8 | 0.219 9 | 0.038 3 | 0.007 8 | |
20 | 2.406 3 | 2.164 9 | 1.046 9 | 0.091 7 | 0.009 4 | |
30 | 3.184 8 | 4.131 1 | 3.481 6 | 0.630 6 | 0.031 5 | |
40 | 6.271 8 | 5.292 7 | 4.840 1 | 1.115 0 | 0.206 2 | |
Pollution | 0 | 0.004 9 | 0.005 2 | 0.004 9 | 0.005 0 | 0.005 1 |
10 | 4.524 9 | 2.671 9 | 2.001 8 | 0.523 8 | 0.025 3 | |
20 | 7.289 4 | 5.956 7 | 5.155 2 | 2.650 6 | 1.007 5 | |
30 | 8.261 1 | 8.449 5 | 6.764 9 | 3.984 7 | 2.342 9 | |
40 | 12.219 1 | 9.569 1 | 7.839 2 | 5.664 7 | 2.764 5 | |
Machine-CPU | 0 | 0.001 8 | 0.002 2 | 0.002 3 | 0.001 9 | 0.001 8 |
10 | 1.724 3 | 0.709 8 | 0.426 8 | 0.168 4 | 0.004 8 | |
20 | 2.279 0 | 2.201 3 | 1.773 7 | 0.673 8 | 0.010 7 | |
30 | 2.551 9 | 2.512 7 | 2.476 3 | 1.419 8 | 0.020 8 | |
40 | 5.653 4 | 5.542 5 | 5.452 9 | 3.051 8 | 0.048 0 | |
Concrete Compressive | 0 | 0.010 1 | 0.010 3 | 0.010 1 | 0.010 3 | 0.010 2 |
10 | 0.552 0 | 0.137 7 | 0.013 8 | 0.015 8 | 0.011 5 | |
20 | 0.724 9 | 0.535 6 | 0.437 5 | 0.031 9 | 0.025 2 | |
30 | 0.965 1 | 0.821 4 | 0.630 3 | 0.062 9 | 0.015 9 | |
40 | 1.396 0 | 1.175 9 | 0.850 9 | 0.118 8 | 0.051 9 | |
AutoMPG | 0 | 0.071 0 | 0.075 2 | 0.091 1 | 0.080 0 | 0.102 3 |
10 | 1.753 8 | 1.481 9 | 0.312 3 | 0.126 1 | 0.115 2 | |
20 | 1.591 5 | 0.943 5 | 0.871 6 | 0.192 3 | 0.117 5 | |
30 | 1.823 0 | 1.494 0 | 1.417 3 | 0.706 8 | 0.152 2 | |
40 | 2.811 6 | 2.260 7 | 2.270 2 | 1.037 1 | 0.338 3 | |
Abalone | 0 | 0.007 2 | 0.007 4 | 0.007 9 | 0.007 3 | 0.007 2 |
10 | 0.935 6 | 0.299 8 | 0.180 8 | 0.040 5 | 0.007 7 | |
20 | 1.243 5 | 0.941 5 | 0.665 4 | 0.013 6 | 0.010 7 | |
30 | 1.431 5 | 1.391 1 | 1.026 3 | 0.206 9 | 0.014 3 | |
40 | 1.770 4 | 1.746 2 | 1.771 8 | 0.756 4 | 0.065 3 | |
Residential-Building | 0 | 0.003 4 | 0.004 5 | 0.005 7 | 0.003 5 | 0.003 7 |
10 | 0.851 4 | 0.857 7 | 0.015 3 | 0.004 3 | 0.004 3 | |
20 | 1.993 2 | 1.891 4 | 2.559 5 | 0.123 3 | 0.008 5 | |
30 | 2.644 7 | 2.366 0 | 2.054 4 | 0.194 4 | 0.024 3 | |
40 | 2.761 8 | 2.734 3 | 2.961 6 | 1.308 3 | 0.032 3 | |
CO and NO | 0 | 0.002 1 | 0.002 3 | 0.002 6 | 0.002 2 | 0.002 1 |
10 | 0.790 4 | 0.198 2 | 0.004 7 | 0.008 3 | 0.006 5 | |
20 | 1.234 1 | 1.074 0 | 0.745 8 | 0.100 3 | 0.014 1 | |
30 | 1.807 3 | 1.738 2 | 1.884 2 | 0.491 2 | 0.028 9 | |
40 | 2.250 2 | 2.356 7 | 2.533 9 | 0.721 4 | 0.065 2 | |
CCPP | 0 | 0.002 8 | 0.002 9 | 0.002 9 | 0.002 9 | 0.002 9 |
10 | 0.681 2 | 0.003 3 | 0.003 1 | 0.003 2 | 0.003 0 | |
20 | 0.909 5 | 0.373 9 | 0.053 2 | 0.003 7 | 0.003 8 | |
30 | 1.019 5 | 1.818 5 | 2.073 3 | 0.045 0 | 0.021 5 | |
40 | 1.328 4 | 2.138 6 | 2.708 8 | 0.168 9 | 0.043 5 | |
Forest Fire Data Set | 0 | 0.001 8 | 0.002 1 | 0.001 9 | 0.002 1 | 0.002 0 |
10 | 0.681 0 | 0.075 5 | 0.002 5 | 0.003 4 | 0.002 7 | |
20 | 1.120 9 | 1.098 1 | 0.061 1 | 0.005 2 | 0.003 2 | |
30 | 1.387 9 | 1.298 1 | 0.757 8 | 0.009 8 | 0.005 3 | |
40 | 1.789 9 | 1.501 6 | 1.342 6 | 0.308 8 | 0.025 9 | |
Airfoil Self-Noise Data | 0 | 0.005 8 | 0.006 1 | 0.006 2 | 0.005 9 | 0.005 8 |
10 | 0.381 0 | 0.075 5 | 0.010 8 | 0.010 6 | 0.012 6 | |
20 | 1.643 8 | 0.875 4 | 0.971 4 | 0.270 8 | 0.066 7 | |
30 | 1.895 4 | 1.450 5 | 1.236 9 | 0.653 3 | 0.073 7 | |
40 | 2.362 2 | 1.755 3 | 1.860 1 | 0.934 5 | 0.012 1 | |
Bias modified | 0 | 0.006 1 | 0.006 2 | 0.006 0 | 0.006 3 | 0.006 2 |
10 | 1.059 9 | 0.500 7 | 0.016 9 | 0.018 0 | 0.008 3 | |
20 | 0.810 2 | 0.585 8 | 0.295 1 | 0.062 3 | 0.022 3 | |
30 | 1.810 0 | 1.138 2 | 0.834 6 | 0.284 0 | 0.062 4 | |
40 | 1.838 5 | 1.778 4 | 1.740 2 | 0.509 6 | 0.177 2 |
数据集 | ELM | WELM | IRWELM | IRRELM | IMRELM |
---|---|---|---|---|---|
Fatbody | 4.4 | 4.0 | 3.2 | 2.1 | 1.1 |
Pollution | 4.4 | 4.1 | 2.8 | 2.1 | 1.4 |
Machine-CPU | 4.4 | 4.0 | 3.2 | 2.1 | 1.1 |
Concrete Compressive | 4.4 | 4.0 | 2.8 | 2.4 | 1.2 |
AutoMPG | 4.4 | 3.7 | 3.1 | 2.1 | 1.5 |
Abalone | 4.4 | 3.7 | 3.0 | 2.2 | 1.5 |
Residential-Building | 4.4 | 4.0 | 3.2 | 2.0 | 1.2 |
CO and NO | 4.6 | 4.0 | 3.0 | 2.1 | 1.3 |
CCPP (Combined Cycle Power Plant) | 4.6 | 3.5 | 2.8 | 2.8 | 1.1 |
Forest Fires Data Set | 4.6 | 4.0 | 2.5 | 2.5 | 1.4 |
Airfoil Self-Noise Data | 4.6 | 4.0 | 3.0 | 2.2 | 1.2 |
Bias modified | 4.6 | 4.0 | 2.6 | 2.6 | 1.2 |
表4 五种算法在12个真实数据集上的平均序值
Tab. 4 Average order values of five algorithms on 12 real datasets
数据集 | ELM | WELM | IRWELM | IRRELM | IMRELM |
---|---|---|---|---|---|
Fatbody | 4.4 | 4.0 | 3.2 | 2.1 | 1.1 |
Pollution | 4.4 | 4.1 | 2.8 | 2.1 | 1.4 |
Machine-CPU | 4.4 | 4.0 | 3.2 | 2.1 | 1.1 |
Concrete Compressive | 4.4 | 4.0 | 2.8 | 2.4 | 1.2 |
AutoMPG | 4.4 | 3.7 | 3.1 | 2.1 | 1.5 |
Abalone | 4.4 | 3.7 | 3.0 | 2.2 | 1.5 |
Residential-Building | 4.4 | 4.0 | 3.2 | 2.0 | 1.2 |
CO and NO | 4.6 | 4.0 | 3.0 | 2.1 | 1.3 |
CCPP (Combined Cycle Power Plant) | 4.6 | 3.5 | 2.8 | 2.8 | 1.1 |
Forest Fires Data Set | 4.6 | 4.0 | 2.5 | 2.5 | 1.4 |
Airfoil Self-Noise Data | 4.6 | 4.0 | 3.0 | 2.2 | 1.2 |
Bias modified | 4.6 | 4.0 | 2.6 | 2.6 | 1.2 |
数据集 | CD | Δ(IMRELM-ELM) | Δ(IMRELM-WELM) | Δ(IMRELM-IRWELM) | Δ(IMRELM-IRRELM) | ||||
---|---|---|---|---|---|---|---|---|---|
Fatbody | 27.144 0 | 24.520 3 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.9 | 1.7 | 1.0 |
Pollution | 19.368 0 | 9.316 0 | 2.606 | 2.459 | 1.832 8 | 3.0 | 2.8 | 1.4 | 0.7 |
Machine-CPU | 22.392 0 | 13.164 0 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.9 | 1.7 | 1.0 |
Concrete Compressive | 19.439 9 | 9.391 3 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.8 | 1.7 | 1.2 |
AutoMPG | 15.552 0 | 6.084 5 | 2.606 | 2.459 | 1.832 8 | 2.9 | 2.2 | 1.7 | 0.6 |
Abalone | 14.904 0 | 5.651 8 | 2.606 | 2.459 | 1.832 8 | 2.9 | 2.2 | 1.6 | 0.7 |
Residential-Building | 21.743 9 | 12.202 0 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.8 | 2.0 | 0.8 |
CO and NO | 26.135 9 | 21.197 0 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.7 | 1.7 | 0.7 |
CCPP | 19.079 9 | 9.021 2 | 2.606 | 2.459 | 1.832 8 | 3.5 | 2.4 | 1.7 | 1.7 |
Forest Fires Data Set | 23.831 9 | 15.668 6 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.6 | 1.1 | 1.1 |
Airfoil Self-Noise Data | 26.783 9 | 23.249 9 | 2.606 | 2.459 | 1.832 8 | 3.4 | 2.8 | 1.8 | 1.0 |
Bias modified | 25.631 9 | 19.777 7 | 2.606 | 2.459 | 1.832 8 | 3.4 | 2.8 | 1.4 | 1.4 |
表5 五种算法在12个真实数据集上Friedman测试的相关数据
Tab. 5 Friedman test data of five algorithms on 12 real datasets
数据集 | CD | Δ(IMRELM-ELM) | Δ(IMRELM-WELM) | Δ(IMRELM-IRWELM) | Δ(IMRELM-IRRELM) | ||||
---|---|---|---|---|---|---|---|---|---|
Fatbody | 27.144 0 | 24.520 3 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.9 | 1.7 | 1.0 |
Pollution | 19.368 0 | 9.316 0 | 2.606 | 2.459 | 1.832 8 | 3.0 | 2.8 | 1.4 | 0.7 |
Machine-CPU | 22.392 0 | 13.164 0 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.9 | 1.7 | 1.0 |
Concrete Compressive | 19.439 9 | 9.391 3 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.8 | 1.7 | 1.2 |
AutoMPG | 15.552 0 | 6.084 5 | 2.606 | 2.459 | 1.832 8 | 2.9 | 2.2 | 1.7 | 0.6 |
Abalone | 14.904 0 | 5.651 8 | 2.606 | 2.459 | 1.832 8 | 2.9 | 2.2 | 1.6 | 0.7 |
Residential-Building | 21.743 9 | 12.202 0 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.8 | 2.0 | 0.8 |
CO and NO | 26.135 9 | 21.197 0 | 2.606 | 2.459 | 1.832 8 | 3.3 | 2.7 | 1.7 | 0.7 |
CCPP | 19.079 9 | 9.021 2 | 2.606 | 2.459 | 1.832 8 | 3.5 | 2.4 | 1.7 | 1.7 |
Forest Fires Data Set | 23.831 9 | 15.668 6 | 2.606 | 2.459 | 1.832 8 | 3.2 | 2.6 | 1.1 | 1.1 |
Airfoil Self-Noise Data | 26.783 9 | 23.249 9 | 2.606 | 2.459 | 1.832 8 | 3.4 | 2.8 | 1.8 | 1.0 |
Bias modified | 25.631 9 | 19.777 7 | 2.606 | 2.459 | 1.832 8 | 3.4 | 2.8 | 1.4 | 1.4 |
1 | HUANG G B, CHEN L, SIEW C K. Universal approximation using incremental constructive feedforward networks with random hidden nodes[J]. IEEE Transactions on Neural Networks, 2006, 17(4): 879-892. 10.1109/tnn.2006.875977 |
2 | ZHAO J W, WANG Z H, PARK D S. Online sequential extreme learning machine with forgetting mechanism[J]. Neurocomputing, 2012, 87: 79-89. 10.1016/j.neucom.2012.02.003 |
3 | SHANG Z H, HE Z S, CHEN Y, et al. Short-term wind speed forecasting system based on multivariate time series and multi-objective optimization[J]. Energy, 2022, 238(Pt C): No.122024. 10.1016/j.energy.2021.122024 |
4 | 焦广利,张璐,钟麦英. 基于鲁棒极限学习机的污泥膨胀智能检测方法[J]. 山东科技大学学报(自然科学版), 2022, 41(3):111-120. |
JIAO G L, ZHANG L, ZHONG M Y. Intelligent detection method of sludge bulking based on robust extreme learning machine[J]. Journal of Shandong University of Science and Technology (Natural Science), 2022, 41(3):111-120. | |
5 | 王桥, 魏孟,叶敏,等. 基于灰狼算法优化极限学习机的锂离子电池 SOC 估计[J]. 储能科学与技术, 2021, 10(2): 744-751. 10.19799/j.cnki.2095-4239.2020.0389 |
WANG Q, WEI M, YE M, et al. Estimation of lithium-ion battery SOC based on GWO-optimized extreme learning machine[J]. Energy Storage Science and Technology, 2021, 10(2): 744-751. 10.19799/j.cnki.2095-4239.2020.0389 | |
6 | HUANG G B, ZHU Q Y, SIEW C K. Extreme learning machine: theory and applications[J]. Neurocomputing, 2006, 70(1/2/3): 489-501. 10.1016/j.neucom.2005.12.126 |
7 | CHEN X, DONG Z Y, MENG K, et al. Electricity price forecasting with extreme learning machine and bootstrapping[J]. IEEE Transactions on Power Systems, 2012, 27(4): 2055-2062. 10.1109/tpwrs.2012.2190627 |
8 | HUANG G B, ZHOU H M, DING X J, et al. Extreme learning machine for regression and multiclass classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2012, 42(2): 513-529. 10.1109/tsmcb.2011.2168604 |
9 | YANG Y, ZHOU H, GAO Y C, et al. Robust penalized extreme learning machine regression with applications in wind speed forecasting[J]. Neural Computing and Applications, 2022, 34(1): 391-407. 10.1007/s00521-021-06370-3 |
10 | DENG W Y, ZHENG Q H, CHEN L. Regularized extreme learning machine[C]// Proceedings of the 2009 IEEE Symposium on Computational Intelligence and Data Mining. Piscataway: IEEE, 2009: 389-395. 10.1109/cidm.2009.4938676 |
11 | HORATA P, CHIEWCHANWATTANA S, SUNAT K. Robust extreme learning machine[J]. Neurocomputing, 2013, 102: 31-44. 10.1016/j.neucom.2011.12.045 |
12 | CHEN K, LV Q, LU Y, et al. Robust regularized extreme learning machine for regression using iteratively reweighted least squares[J]. Neurocomputing, 2017, 230: 345-358. 10.1016/j.neucom.2016.12.029 |
13 | ZHA L L, MA K, LI G Q, et al. A robust double-parallel extreme learning machine based on an improved M-estimation algorithm[J]. Advanced Engineering Informatics, 2022, 52: No.101606. 10.1016/j.aei.2022.101606 |
14 | ZHANG K, LUO M X. Outlier-robust extreme learning machine for regression problems[J]. Neurocomputing, 2015, 151(Pt 3): 1519-1527. 10.1016/j.neucom.2014.09.022 |
15 | YANG Y, ZHOU H, WU J R, et al. Robustified extreme learning machine regression with applications in outlier-blended wind-speed forecasting[J]. Applied Soft Computing, 2022, 122: No.108814. 10.1016/j.asoc.2022.108814 |
16 | WU Q, FU Y L, CUI D S, et al. C-loss-based doubly regularized extreme learning machine[J]. Cognitive Computation, 2022: 1-24. 10.1007/s12559-022-10050-2 |
17 | XING H J, WANG X M. Training extreme learning machine via regularized correntropy criterion[J]. Neural Computing and Applications, 2013, 23(7/8): 1977-1986. 10.1007/s00521-012-1184-y |
18 | REN Z, YANG L R. Robust extreme learning machines with different loss functions[J]. Neural Processing Letters, 2019, 49(3): 1543-1565. 10.1007/s11063-018-9890-9 |
19 | WANG K N, CAO J D, PEI H M. Robust extreme learning machine in the presence of outliers by iterative reweighted algorithm[J]. Applied Mathematics and Computation, 2020, 377: No.125186. 10.1016/j.amc.2020.125186 |
20 | UCI [DB/OL]. [2022-03-01]. ∼mlearn/MLRepository.html. |
[1] | 付帅, 郭小英, 白茹意, 闫涛, 陈斌. 改进的CloFormer模型与有序回归相结合的年龄评估方法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2372-2380. |
[2] | 孙淳, 胡春龙, 黄树成. 一致性保留的集成排序年龄估计方法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2381-2386. |
[3] | 刘青, 陈艳平, 邹安琪, 黄瑞章, 秦永彬. 面向机器阅读理解的边界感知方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2004-2010. |
[4] | 雷明珠, 王浩, 贾蓉, 白琳, 潘晓英. 基于特征间关系合成少数类样本的过采样算法[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1428-1436. |
[5] | 邹博士, 杨铭, 宗辰辰, 谢明昆, 黄圣君. 基于负学习的样本重加权鲁棒学习方法[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1479-1484. |
[6] | 李宗禹, 强思维, 郭晓波, 朱振峰. 重加权的对抗变分自编码器及其在工业因果效应估计中的应用[J]. 《计算机应用》唯一官方网站, 2024, 44(4): 1099-1106. |
[7] | 任帅, 纪元法, 孙希延, 韦照川, 林子安. 基于改进灰狼优化与支持向量回归的滑坡位移预测[J]. 《计算机应用》唯一官方网站, 2024, 44(3): 972-982. |
[8] | 黄雨鑫, 黄贻望, 黄辉. 基于浅层网络预测的元标签校正方法[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3364-3370. |
[9] | 孙凯明, 蔡东风, 白宇. 基于自学习的整数数列符号回归方法[J]. 《计算机应用》唯一官方网站, 2024, 44(10): 3158-3166. |
[10] | 冯程皓, 谢振平, 丁博文. 中文文本纠错软件测试用例的选择生成方法[J]. 《计算机应用》唯一官方网站, 2024, 44(1): 101-112. |
[11] | 岑黎彬, 李靖东, 林淳波, 王晓玲. 基于深度自回归模型的近似查询处理方法[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2034-2039. |
[12] | 王逸, 裴生雷, 王煜. 基于CSI和K-means-SVR的多指纹库室内定位方法[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1636-1640. |
[13] | 徐少康, 张战成, 姚浩男, 邹智伟, 张宝成. 基于姿态编码器的2D/3D脊椎医学图像实时配准方法[J]. 《计算机应用》唯一官方网站, 2023, 43(2): 589-594. |
[14] | 郭子豪, 董乐乐, 曲志坚. 基于改进Faster RCNN的节肢动物目标检测方法[J]. 《计算机应用》唯一官方网站, 2023, 43(1): 88-97. |
[15] | 李兴佳, 杨秋辉, 洪玫, 潘春霞, 刘瑞航. 基于历史数据和多目标优化的测试用例排序方法[J]. 《计算机应用》唯一官方网站, 2023, 43(1): 221-226. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||