基于HMM/SVM的风电设备故障趋势预测方法研究

Wind Power Electricity Generation Facility Fault Trend Prediction Method Study Based on HMM/SVM

由于风力发电设备复杂且积累的资料与故障样本少;传统的诊断方法,例如神经网络,忽视了前与后关系,且需要大量故障训练样本,往往都不能有效的进行故障诊断;结合隐马尔可夫模型(Hidden Markov Model,HMM)有利于处理连续动态信号,以及支持向量机(Support Vector Machine,SVM)分类能力强的优点;提出了基于HMM/SVM串联结构的故障诊断模型;首先通过从风电设备振动信号中有效提取非平稳特征,利用HMM计算未知信号与风力发电设备各状态的匹配程度,形成特征向量提供给SVM最后判别,实验结果表明该方法比单纯HMM和SVM识别率分别提高了9.17%和5.84%。

Due to the complexity of wind power generate electricity facility and less accumulation of data and fault samples. The traditional diagnosis methods such as neural network, can not effectively. Diagnosis is a moment that the result of the information to match the template library, ignoring the relationship between before and after, and need to training a large number of fault samples. Based on Hidden Markov Model （HMM） that conducive to the continuous dynamic characteristics of the signal processing, and the Support Vector Machine （SVM） that classification of the advantages of strong ability. This paper based on the HMM/SVM series structure fault diagnosis model. Firstly from the wind power generate electricity facility effectively extract non--stationary characteristics of vibration signals, the HMM is used to calculate the unknown signal and the matching degree of wind power equipment conditions, the form feature vector for the SVM discriminant finally, experimental of results show that the method can improve9.17% and 5.84% respectively than the pure HMM or SVM diagnosis method.