基于量子粒子群优化Volterra时域核辨识的隐Markov模型识别方法  

Hidden Markov model recognition method based on Volterra kernel identified with particle swarm optimization

作　　者：李志农[1,2] 蒋静[2] 冯辅周[3] 袁振伟[2] 

机构地区：[1]南昌航空大学无损检测技术教育部重点实验室,南昌330063 [2]郑州大学机械工程学院,郑州450001 [3]装甲兵工程学院机械工程系,北京100072

出　　处：《仪器仪表学报》2011年第12期2693-2698,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(50775208;51075372);湖南省机械设备健康维护重点实验室开放基金(200904);江西省研究生教育创新基地基金资助项目

摘　　要：将量子粒子群优化算法引入Volterra级数模型的非线性辨识中,并结合隐Markov模型(hidden Markov model,HMM),提出了一种基于量子粒子群优化的Volterra时域核特征提取的HMM识别方法,在提出的方法中,利用量子粒子群优化算法辨识得到的前三阶Volterra时域核作为故障特征,输入到各种状态的HMM中,其中,输出概率最大的HMM对应的状态即为设备的当前运行状态。提出的方法克服了传统的基于Volterra模型系统的机械故障诊断要求目标函数连续可导、容易陷入局部最小以及抗干扰能力差等缺陷。最后,将提出的方法应用到旋转机械故障诊断中。实验结果验证了该方法的有效性。Quantum panicle swarm optimization （QPS0） is introduced into nonlinear system identification of Volterra series model and combined with hidden Markov model （HMM） ; a new fault recognition method is proposed. In the proposed method, the first three order Voherra kernel identified with QPSO algorithm is used as the fault feature vector, which is inputted into the HMM of each fault mode for training; and the HMM with maximum probability determines current condition of the set. The proposed method can overcome the deficiencies of traditional fault diagnosis method based on Volterra model, i.e. the object function must be continuous and derivative, easy to fall into local minimum, has poor anti-interference capability and etc. Finally the proposed method was applied in the fault diagnosis of rotating machine. Experiment result shows that the proposed method is very effective.

关 键 词：VOLTERRA级数 隐Markov模型(HMM) 量子粒子群优化(QPSO) 故障诊断 模式识别 

分 类 号：TN17[电子电信—物理电子学] O235[理学—运筹学与控制论]