摘要
多元LS-SVM算法可以直接用于多分类模式识别问题,通过对该算法的误差变量进行加权,消除了训练样本中异常值或非高斯噪声的影响,增强了多元LS-SVM算法的鲁棒性。然后,利用改进算法建立特征向量与故障模式之间的映射关系,得到齿轮箱故障诊断模型。仿真表明:与传统BP神经网络相比,鲁棒多元LS-SVM算法对齿轮箱的故障诊断的精度更高,抗干扰能力和鲁棒性更强,是一种在齿轮箱故障诊断中值得推广和采用的算法。
The multi-classification LS-SVM algorithm can be directly used for multi-classification pattern recognition problems,this paper removes the influence of outliers and non-guassian noise in training samples to force the robustness of multi-classification LS-SVM by weighting the error variable.Then based on this improved algorithm,models the gearbox fault diagnosis by mapping the feature vectors to the corresponding fault modes.The simulation result indicate this algorithm,compared with the conventional BP neural network,has a better diagnosis of gearbox fault with features of simplicity,precision,robustness,and is a good method which is worthy of popularizing and promoting in gearbox fault diagnosis.
出处
《火力与指挥控制》
CSCD
北大核心
2010年第5期93-96,102,共5页
Fire Control & Command Control
关键词
最小二乘
支持向量机
齿轮箱
故障诊断
least square
support vector machine
gearbox
fault diagnosis
