CIESC Journal ›› 2015, Vol. 66 ›› Issue (1): 351-256.DOI: 10.11949/j.issn.0438-1157.20141452

Previous Articles     Next Articles

Multiple timing-driven based extreme learning machine whole process fault prediction and its application

XU Yuan, LU Yushuai, CAI Yi   

  1. College of Information Science & Technology, Beijing University of Chemical Technology, Beijing 100029, China
  • Received:2014-09-25 Revised:2014-10-05 Online:2015-01-05 Published:2015-01-05
  • Supported by:

    supported by the National Natural Science Foundation of China (61104131), the Natural Science Foundation of Beijing (4142039) and the Programs for Beijing Excellent Talents D (2013D0009016000004).

基于多元时序驱动的ELM全流程故障预测及其应用

徐圆, 卢玉帅, 才轶   

  1. 北京化工大学信息与科学技术学院, 北京 100029
  • 通讯作者: 徐圆
  • 基金资助:

    国家自然科学基金项目(61104131);北京市自然科学基金项目(4142039);北京市优秀人才培养资助D类(2013D009016000004)。

Abstract:

A multiple timing-driven modeling method is an effective way for fault prediction and state evaluation of complex system, in which the artificial neural network is an effective data-driven modeling tool to deal with the nonlinear problems. Recently, it has been widely concerned on the multiple timing-driven modeling problems. In the paper, from the perspective of the whole process, the k-nearest neighbor mutual information method is firstly used to reduce the dimension of the multiple timing variables and calculate the correlation among the variables, so as the select the characteristic variable. Second, an improved trend analysis method is proposed to monitor the system state in real time and segment the system operation state. Finally, aiming at the potential fault stage, extreme learning machine (ELM) neural network is used for fault prediction. Through the simulation experiment on penicillin fermentation process, the results verify the effectiveness of the proposed method.

Key words: extreme learning machine, mutual information, fault prediction, multi-timing, trend analysis

摘要:

多元时序驱动建模方法是复杂系统故障预测和系统状态评估的一种有效途径, 其中人工神经网络作为一种数据驱动的处理非线性问题的有效建模工具, 近年来在处理多元时序建模这个问题上得到了较广泛的关注。从全流程的角度出发, 首先, 运用k-近邻互信息方法对多元时序变量进行降维与相关性计算, 从而选择特征变量;其次, 提出了一种改进的趋势分析方法对系统的状态进行实时监测, 并对系统运行状态进行有效细分;最后, 针对系统潜在故障阶段, 应用极限学习机(extreme learning machine, ELM)神经网络方法对其进行故障预测。通过对青霉素发酵过程(penicillin fermentation process)进行仿真实验, 结果验证了所提方法的有效性。

关键词: 极限学习机, 互信息, 故障预测, 多元时序, 趋势分析

CLC Number: