摘要
针对设备故障诊断过程中构建特征参数冗余,且进行高分辨率信息压缩所需的映射通常具有非线性的问题,应用BP神经网络提取设备状态特征,给出了进行设备状态特征集约简的实施方法.然后利用最小二乘支持向量机(LS-SVM)分类器的训练过程遵循结构风险最小化原则,能够避免传统机器学习的模型选择、过学习、局部极小等问题,具有有效解决非线性和高维模式识别问题的优点,构建了故障识别模型.最后将基于BP网络和LS-SVM的特征提取和故障识别方法用于离心泵机组的四种工作状态识别,并进行了ROC曲线分析,研究结果表明诊断实验的性能评价为优.
The characteristic parameters are often constructed redundantly in fault diagnosis.The information compression mapping with the high resolution is usually nonlinear.To counter these problems,the BP networks structure is firstly designed for equipment's feature extraction,and the implement method for feature reduction is presented in the paper.Secondly,since LS-SVM employs structural risk minimization criterion and has prominent advantages in selecting model,overcoming over-fitting and local minimum,and solving the problems of the nonlinear and high-dimensional pattern recognition,a fault identification model is proposed.Finally,the method of feature extraction and fault identification based on BP networks and LS-SVM is applied to the centrifugal pump status recognition.The experimental results and the ROC curve analysis prove the effectiveness of the method.
出处
《昆明理工大学学报(理工版)》
北大核心
2010年第5期41-46,共6页
Journal of Kunming University of Science and Technology(Natural Science Edition)
关键词
特征提取
故障识别
BP网络
最小二乘支持向量机
ROC曲线
feature extraction
fault identification
BP network
least squares support vector machine
ROC curve