相关向量机在光纤预警系统模式识别中的应用

由于传统模式识别方法存在过学习、训练时间长等缺陷,不能满足光纤预警系统实时在线监测的要求.相关向量机能够克服传统方法的缺点,识别精度高,向量机个数需求少,因此,将相关向量机应用于光纤预警系统模式识别中,采用小波能谱和小波信息熵的特征提取方法,在测试阶段采用有向无环图的方法进行多类识别.通过对威胁管道安全的事件进行实验,识别精度达到92.67%,向量机个数只有2个,验证了相关向量机方法应用于光纤预警系统的可行性和有效性.

Application of Relevance Vector Machine in Pattern Recognition of Optical Fiber Pre-Warning System

Because of severe over-fitting and high computation cost, traditional pattem recognition methods cannot meet the demands for the online monitoring of optical fiber pre-warning system. Relevance vector machines （RVMs） can overcome the disadvantages of traditional methods, which have higher identity precision and fewer vector number, therefore, in this paper RVM was used as recognition method in optical fiber pre-warning system. Wavelet energy and wavelet information entropy were also used as feature extraction method, and directed acyclic graph was used to test multi-types of samples. The experiment on pipeline safety events shows that the recognition accuracy of RVM can reach 92.67%, and the vector number is only 2. The method proposed in this paper is proved to be feasible and effective for optical fiber pre-warning system.