一种增量式贝叶斯算法及篦冷机故障诊断

摘要/Abstract

摘要： 针对批量式算法增量维护性能差的缺点，提出了一种贝叶斯增量学习算法(ILA)。检测到新数据集时，构造WTUN函数来判断结构是否需要更新，若结构需要更新，则构建影响度(Affect)函数，得到结构中需要修正的节点集，在其马尔可夫范围内利用爬山算子修改得到候选结构，利用改进的评分函数选择评分最大的结构作为最优结构。无论结构是否更新，都将原参数作为先验参数，利用EM算法更新参数。将该算法与批量爬山（HC）算法、增量爬山（IHC）算法、增量遗传算法（IGA）对比，ILA算法可以对网络进行增量维护，一定程度上节省了空间和时间。利用该算法建立篦冷机工艺参数的故障诊断模型，该模型能较为准确地实现对二次风温的故障诊断。

关键词: 贝叶斯增量学习, 结构学习, 篦冷机, 水泥熟料换热, 二次风温, 故障诊断

Abstract: Aiming at the shortcomings of incremental maintenance performance of batch algorithm, an incremental learning algorithm(ILA) of Bayesian network was proposed. When the new data sets were detected, the WTUN function was constructed to determine whether to update the structure. If the structure was updated, the Affect function was constructed to obtain the set of nodes that needed to be modified in the structure. The candidate structure was obtained by hill climbing operator correction in the Markov range of these nodes. The improved scoring function was used to select the largest structure as the optimal structure. Whether the structure was updated or not, the original parameters were used as the prior parameters, and the parameters were updated with the expectation maximization(EM) algorithm.The algorithm may incrementally maintain the network and save space and time to a certain extent compared with hill climbing(HC),incremental hill climbing(IHC) and incremental genetic algorithm(IGA). The fault diagnosis model of the technological parameters of grate coolers was established by ILA algorithm, which may be used to realize the accurate fault diagnosis of the secondary air temperature.

Key words: Bayesian incremental learning, structure learning, grate cooler, clinker heat transfer, secondary air temperature, fault diagnosis

中图分类号:

TH165.3

刘兆伦1,2;张春兰2;武尤2;王海羽2;刘彬1,2. 一种增量式贝叶斯算法及篦冷机故障诊断[J]. 中国机械工程.

LIU Zhaolun1,2;ZHANG Chunlan2;WU You2;WANG Haiyu2;LIU Bin1,2. An Incremental Bayesian Algorithm on Fault Diagnosis of Grate Coolers[J]. China Mechanical Engineering.

参考文献

[1]杨昌昊，竺长安，胡小建. 基于贝叶斯网的复杂系统故障诊断方法[J].中国机械工程,2009, 20(22):2726-2732.
YANG Changhao,ZHU Changan,HU Xiaojian.Methods on Complex System Fault Diagnosis Based on Bayesian Network[J].China Mechanical Engineering,2009, 20(22): 2726-2732.
[2]SAMET S, MIN A, GRANGER E, et al. Incremental Learning of Privacy-preserving Bayesian Networks [J].Appl. Soft Comput., 2013, 13(8): 3657-3667.
[3]刘彬, 刘永记, 刘浩然, 等. 基于贝叶斯改进结构算法的回转窑故障诊断模型研究[J].中国机械工程, 2017, 28(18): 2143-2151.
LIU Bin, LIU Yongji, LIU Haoran, et al.Research on Fault Diagnosis Model of Rotary Kiln Based on Improved Bayesian Structure Algorithm[J].China Mechanical Engineering, 2017, 28(18): 2143-2151.
[4]ZHU Y, LIMIN H, LU J. Bayesian Networks-based Approach for Power Systems Fault Diagnosis[J]. IEEE Transactions on Power Delivery, 2006, 21(2): 634-639.
[5]田凤占, 黄丽, 于剑, 等. 包含隐变量的贝叶斯网络增量学习方法[J]. 电子学报, 2005, 33(11): 1925-1928.
TIAN Fengzhan, HUANG Li, YU Jian, et al. Incremental Learning Method for Bayesian Networks with Hidden Variables[J]. Journal of Electronics, 2005, 33(11): 1925-1928.
[6]ALCOBE J R. An Incremental Algorithm for Tree-shaped Bayesian Network Learning[C]// Eureopean Conference on Artificial Intelligence.Lyon, 2002:350-354.
[7]王飞, 刘大有, 王淞昕. 基于遗传算法的Bayesian网结构增量学习的研究[J]. 计算机研究与发展, 2005, 42(9): 1461-1466.
WANG Fei, LIU Dayou, WANG Songxin.Research on Incremental Learning of Bayesian Nets Based on Genetic Algorithm[J].Computer Research and Development, 2005, 42(9): 1461-1466.
[8]ZHU Y, LIU D, CHEN G, et al. Mathematical Modeling for Active and Dynamic Diagnosis of Crop Diseases Based on Bayesian Networks and Incremental Learning [J]. Mathematical & Computer Modelling, 2013, 58(3/4): 514-523.
[9]YASIN A, LERAY P. Incremental Bayesian Network Structure Learning in High Dimensional Domains[C]// International Conference on Modeling, Simulation and Applied Optimization. Hammamet, 2013:1-6.
[10]慕春棣, 戴剑彬, 叶俊. 用于数据挖掘的贝叶斯网络[J].软件学报, 2000, 11(5): 660-666.
MU Chunli, DAI Jianbin, YE Jun.Bayesian Networks for Data Mining[J].Journal of Software, 2000, 11(5): 660-666.
[11]LI S H, ZHANG J, SUN B L, et al. An Incremental Structure Learning Approach for Bayesian Network[C]// Control and Decision Conference.Changsha, 2014:4817-4822.
[12]MADDEN M G. Evaluation of the Performance of the Markov Blanket Bayesian Classifier Algorithm[R]. Galway: National University of Ireland, 2002.
[13]王宝帅, 杜兰, 和华,等. 基于复高斯模型的样本缺失窄带雷达信号重构算法[J]. 电子与信息学报, 2015, 37(5): 1065-1070.
WANG Baoshuai, DU Lan, HE Hua, et al.Reconstruction Algorithm of Narrowband Radar Signals with Missing Samples Based on Complex Gauss Model [J]. Journal of Electronics and Information, 2015, 37(5): 1065-1070.
[14]YUE K, FANG Q, WANG X, et al. A Parallel and Incremental Approach for Data-intensive Learning of Bayesian Networks[J]. IEEE Transactions on Cybernetics, 2015, 45(12): 2890-2904.
[15]王美琪, 刘彬, 闻岩,等. 篦冷机内水泥熟料温度的软测量[J].机械工程学报, 2016, 52(6): 159-165.
WANG Meiqi, LIU Bin, WEN Yan, et al. Soft Measurement of Cement Clinker Temperature in Grate Cooler[J]. Journal of Mechanical Engineering, 2016, 52(6): 159-165.
[16]刘浩然, 吕晓贺, 李轩,等. 基于Bayesian改进算法的回转窑故障诊断模型研究[J].仪器仪表学报, 2015, 36(7): 1554-1561.
LIU Haoran, LYU Xiaohe, LI Xuan, et al. An effective Inference Algorithm for Fault Diagnosis[J].Control and Decision, 2015, 30(11): 2033-2040.