为实现柴油机故障在线诊断，提出了基于增量稀疏核极限学习机(ISKELM)的快速在线诊断方法.针对核在线学习中的样本稀疏化与模型膨胀问题，提出了基于瞬时信息测量的稀疏核函数字典构造策略，根据最小化字典冗余和最大化字典元素自信息量的原则实现样本前向稀疏与后向删减，在最佳阶数内对字典进行在线扩充与修剪，从而建立阶数有限且结构稀疏的诊断模型.针对模型核权重矩阵更新问题，提出了增样学习与改进减样学习算法对核权重矩阵进行在线递推求解，降低了计算复杂度，提高了模型在线更新速度.UCI标准数据与柴油机故障数据分类实验结果表明，与几类现有在线诊断算法相比，ISKELM在保证较高分类精度的同时，极大地提高了在线建模速度，更加快速准确地实现了柴油机故障在线诊断.
In order to realize online fault diagnosis of diesel engine, a fast online diagnosis method based on incremental sparse kernel extreme learning machine (ISKELM) was proposed. Aiming at the problem of sample sparseness and model expansion in kernel online learning process, a sparse kernel function matrix construction strategy based on instantaneous information measurement was proposed. In the strategy, sample forward sparseness and backward deletion are implemented according to the principle of minimizing dictionary redundancy and maximizing the self information of dictionary elements. Then the dictionary is expanded and pruned online under the best order. So a diagnosis model with limited order and sparse structure is established. In order to solve the updating problem of the model kernel weight matrix, a method consisting of sample addition learning algorithm and improved sample reduction learning algorithm was put forward. It reduces the computational complexity and improves the online updating speed of the diagnosis model. UCI standard data and diesel engine fault data classification experiment results indicate that compared with several existing online diagnosis methods, ISKELM not only has high classification accuracy, but also greatly improves the online modeling speed, and it realizes diesel engine fault online diagnosis more quickly and accurately.
［1］尹刚, 张英堂, 李志宁, 等. 运用在线贯序极限学习机的故障诊断方法［J］.振动、测试与诊断, 2013, 33(2): 325-329.
YIN Gang, ZHANG Yingtang, LI Zhining, et al. Fault diagnosis based on online sequential extreme learning machine［J］. Journal of Vibration, Measurement & Diagnosis, 2013, 33(2): 325-329.
［2］MEHDIZADEH M, MACNISH C, KHAN R N, et al. Semi-supervised neighborhood preserving discri-minate embedding: A semi-supervised subspace learning algorithm［J］. Lecture Notes in Computer Science, 2011, 6494: 199-212.
［3］HONEINE P. Analyzing sparse dictionaries for online learning with kernels［J］. IEEE Transactions on Signal Processing, 2015, 63(23): 6343-6353．
［4］LIU W F, PARK I I, PRINCIPE J C. An information theoretic approach of designing sparse kernel adaptive filters［J］. IEEE Transactions on Neural Networks, 2009, 20(12): 1950-1961.
［5］HUANG G B, ZHU Q Y, SIEW C K. Extreme learning machine: Theory and application［J］. Neurocomputing, 2006, 70(1/2/3): 489-501.
［6］SIMONE S, DANILO C, MICHELE S, et al. Online sequential extreme learning machine with kernel［J］. IEEE Transactions on Neural Networks and Learning Systems, 2015, 26(9): 2214-2220.
［7］ZHOU X R, WANG C S. Cholesky factorization based online regularized and kernelized extreme learning machines with forgetting mechanism［J］. Neurocomputing, 2016, 174: 1147-1155.
［8］GUO L, HAO J H, LIU M. An incremental extreme learning machine for online sequential learning problem［J］. Neurocomputing, 2014, 128: 50-58.
［9］JONES M C, HENDERSON D A. Maximum likelihood kernel density estimation: on the potential of convolution sieves［J］. Computational Statistics & Data Analysis, 2009, 53(10): 3726-3733.
［10］尹刚, 张英堂, 李志宁, 等. 基于MSPCA的缸盖振动信号特征增强方法研究［J］. 振动与冲击, 2013, 32(6): 143-147.
YIN Gang, ZHANG Yingtang, LI Zhining, et al. Fault feature enhancement method for cylinder head vibration signal based on multiscale principal component analysis［J］. Journal of Vibration and Shock, 2013, 32(6): 143-147.
