文章摘要
岳忠奇,吴涛,顾宏.基于SaCE-ELM的地铁牵引控制单元快速故障诊断[J].,2016,56(3):270-278
基于SaCE-ELM的地铁牵引控制单元快速故障诊断
Fast fault diagnosis of metro traction control unit based on SaCE-ELM
  
DOI:10.7511/dllgxb201603008
中文关键词: 牵引控制单元  故障诊断  极限学习机  差分进化算法
英文关键词: traction control unit  fault diagnosis  extreme learning machine  differential evolution algorithm
基金项目:国家自然科学基金资助项目(U1560102);高等学校博士学科点专项科研基金资助项目(20120041110008).
作者单位
岳忠奇,吴涛,顾宏  
摘要点击次数: 822
全文下载次数: 799
中文摘要:
      地铁牵引控制单元(TCU)在地铁运行过程中有重要的作用,及时有效地对其进行故障诊断,是保证地铁正常运行的重要环节.针对传统故障诊断方法的学习速度慢、易陷入局部最优、预测精度较差等缺点,提出一种使用自适应差分进化算法(SaCE)进行优化的极限学习机(SaCE-ELM),即通过自适应差分进化算法对极限学习机的输入权重、隐含层参数和输出权重进行优化.其中,差分进化算法的变异策略通过基于混沌序列的自适应机制产生,其他参数使用正态分布随机生成;网络的输出权重使用Moore-Penrose广义逆矩阵计算得出. SaCE-ELM 不需要人工选择变异策略和参数,自适应策略比SaE-ELM更加简单.实验结果表明,与E-ELM、SaE-ELM、LM-NN、SVM相比,SaCE-ELM具有更好的预测精度.此外, SaCE-ELM 在所有数据集上训练时间比SaE-ELM和SVM更少,有效地改善了生成模型的效率.
英文摘要:
      Metro traction control unit (TCU) plays a key role in the operation of subway. It is important for the normal operation of subway to diagnose the TCU fault timely and effectively. However, the traditional fault diagnosis methods usually have some disadvantages, such as slow learning speed, falling into local optimum easily and poor prediction accuracy. To solve these problems, extreme learning machine based on adaptive differential evolution algorithm (SaCE-ELM) is proposed. The input weights, the implicit layer parameters and the output weights of the extreme learning machine are optimized by adaptive differential evolution algorithm. The variation strategy of differential evolution algorithm is generated by the adaptive mechanism based on chaotic sequence, and other parameters are randomly generated using normal distribution. The output weights of the network are calculated using Moore-Penrose generalized inverse matrix. SaCE-ELM doesn′t need artificial selection of variation strategy and parameters, and its adaptive strategy is simpler than that of SaE-ELM. Experimental results show that SaCE-ELM has better prediction accuracy compared with E-ELM, SaE-ELM, LM-NN and SVM. Moreover, the training time of SaCE-ELM is shorter than that of SaE-ELM and SVM in all experimental datasets,which demonstrates that the efficiency of model generation has been improved.
查看全文   查看/发表评论  下载PDF阅读器
关闭