基于耦合模拟退火优化最小二乘支持向量机的车轮踏面磨耗量预测
针对最小二乘支持向量机(LS-SVM)超参数优化问题,提出采用改进耦合模拟退火(CSA)算法优化LSSVM超参数。首先,耦合模拟退火算法通过并行处理多个独立模拟退火(SA)寻优过程,提高LS-SVM模型超参数优化效率;然后通过调整接受温度控制耦合项超参数的接受概率方差,降低CSA算法初始设置对LS-SVM最优超参数确定过程稳健性的影响;最后结合既有线轮轨现场的实际检测数据,开展了基于改进耦合模拟退火优化的最小二乘支持向量机(CSA LS-SVM)回归模型性能对比实验。结果表明,CSA LS-SVM回归模型达到了模型精度、算法快速性、算法鲁棒性的有效折中,所建立的LS-SVM优化模型用于现场的车...
Saved in:
Published in | 计算机应用研究 Vol. 32; no. 2; pp. 397 - 402 |
---|---|
Main Author | |
Format | Journal Article |
Language | Chinese |
Published |
华东交通大学 电气与电子工程学院,南昌,330013
2015
|
Subjects | |
Online Access | Get full text |
ISSN | 1001-3695 |
DOI | 10.3969/j.issn.1001-3695.2015.02.018 |
Cover
Summary: | 针对最小二乘支持向量机(LS-SVM)超参数优化问题,提出采用改进耦合模拟退火(CSA)算法优化LSSVM超参数。首先,耦合模拟退火算法通过并行处理多个独立模拟退火(SA)寻优过程,提高LS-SVM模型超参数优化效率;然后通过调整接受温度控制耦合项超参数的接受概率方差,降低CSA算法初始设置对LS-SVM最优超参数确定过程稳健性的影响;最后结合既有线轮轨现场的实际检测数据,开展了基于改进耦合模拟退火优化的最小二乘支持向量机(CSA LS-SVM)回归模型性能对比实验。结果表明,CSA LS-SVM回归模型达到了模型精度、算法快速性、算法鲁棒性的有效折中,所建立的LS-SVM优化模型用于现场的车轮踏面磨耗量的预测是有效的。 |
---|---|
Bibliography: | 51-1196/TP ZHONG Lu-sheng, CHEN Li-yong, GONG Jin-hong, ZHU Zhen-min, XIAO Qian (School of Electrical & Electronic Engineering, East China Jiaotong University, Nanehang 330013, China) coupled simulated annealing; least squares support vector machine; hyper-parameters optimization; tread wear This paper proposed an improved coupled simulated annealing(CSA) algorithm to optimize the hyper-parameters of least squares support vector machine (LS-SVM). First, the CSA algorithm handled multiple independent parallel simulated annealing (SA) optimization process, which improved the optimization efficiency for hyper-parameters of LS-SVM model. Second, the acceptance temperature controUed the variance of the acceptance temperature which reduced the influence of the CSA algorithm to initialization parameters. Finally, it established CSA LS-SVM regression model to predict wheel tread wear based on the field data. The simulation results show that the proposed CSA LS-SVM regression model can trade off the model fit versus the |
ISSN: | 1001-3695 |
DOI: | 10.3969/j.issn.1001-3695.2015.02.018 |