基于多极限学习机在线集成的柴油机故障诊断方法研究  

Diesel Engine Fault Diagnosis Based on Online Ensemble of Multiple Extreme Learning Machine

作　　者：张英堂[1] 马超[1] 尹刚[1] 李志宁[1] 任国全[1] 

机构地区：[1]军械工程学院火炮工程系,河北石家庄050003

出　　处：《车用发动机》2012年第6期85-89,共5页Vehicle Engine

基　　金：河北省自然科学基金资助项目(E20007001048);军内科研项目资助

摘　　要：针对在线贯序极限学习机(OS-ELM)输出不稳定和过学习问题,提出了基于贝叶斯框架的多OS-ELM融合算法。首先通过在目标函数中引入输出矩阵的二范数,将线性回归问题转化为岭回归问题,改善OS-ELM的过学习问题。其次,构建多个OS-ELM分类器对训练样本进行学习,在贝叶斯框架下实现多分类器的在线集成,以提高分类器的输出稳定性。UCI数据集的试验表明,与改进前相比,本算法的分类准确率提高了1.07%～3.35%,100次试验的标准差降低了0.001 5～0.021 4。柴油机11种工况的故障识别准确率可达到96.86%。For output instability and over-learning problem of online sequential extreme learning machine(OS-ELM),the ensemble algorithm of multiple OS-ELM based on the Bayesian method was proposed.Firstly,the linear regression problem was changed to the ridge regression problem by introducing the 2-norm of output matrix to the target function,and the over-learning problem was improved.Secondly,multiple OS-ELM classifiers were constructed to train the samples and the online ensemble of multi-classifier was carried out under the Bayesian framework to improve the stability of classifier output.The trials on UCI data show that,compared with the unimproved algorithm,the classification accuracy of improved algorithm increases by 1.07%-3.35% and the standard deviation of 100 trials decrease 0.001 5-0.021 4.The accuracy of fault recognition for diesel engine eleven working conditions can reach 96.86%.

关 键 词：柴油机 极限学习机 贝叶斯方法 集成学习 多尺度主元分析 故障诊断 

分 类 号：TK428[动力工程及工程热物理—动力机械及工程]