基于视神经网络的混沌时间序列奇异信号实时检测算法  

Real-time novelty detector of chaotic time series based on retina neural network

作　　者：刘金海[1,2] 张化光[1] 冯健[1] 

机构地区：[1]东北大学流程工业综合自动化教育部重点实验室,沈阳110004 [2]东北大学信息科学与工程学院,沈阳110004

出　　处：《物理学报》2010年第7期4472-4479,共8页Acta Physica Sinica

基　　金：国家自然科学基金(批准号:60774048;60774093;60728307);国家高技术研究发展计划(批准号:2009AA04Z127);长江学者奖励计划;高等学校博士学科点专项科研基金(批准号:20070145015);国家重点基础研究发展计划(批准号:2009CB320601);新世纪优秀人才支持计划(批准号:NCET-08-0101)资助的课题~~

摘　　要：提出了一种基于视神经网络的实时检测混沌时间序列中的奇异点算法,设计了视神经网络奇异点检测器(RNNND);然后设计了基于反向传播(BP)神经网络和径向基函数(RBF)神经网络的混沌时间序列奇异点检测器.利用Lorenz理论模型产生的时间序列和实测输油管道压力时间序列分别检验了这3个奇异点检测器在抗干扰能力、检测微弱信号能力和运算速度等方面的性能.仿真和分析表明,RNNND具有良好的检测精度和较快检测速度.最后详细分析了3种奇异点检测器优缺点并给出了适用场合.A kind of novelty detection method based on retina neural network is proposed,which could find the novelty in chaotic time series.To demonstrate the capability of the novelty detection method,we designed three novelty detectors,namely the neural network novelty detector（RNNND）,back-propagation（BP） novelty detector（BPND） and radial base function（RBF） novelty detector（RBFND）,which are based on retina neural network,BP neural network and RBF neural network,respectively.Using Lorenz time series and oil pipeline pressure time series,we tested the performance of the three novelty detectors,including performances of anti-jamming,micro-novelty detection and the computing speed.The results show that the three novelty detectors have good precision and fast computing speed.Finally,the merits and shortcomings of the proposed novelty detection method are analyzed based on retina neural network,BP and RBF neural network,and their applicabilities are given.

关 键 词：混沌时间序列 实时 奇异点检测 视神经网络 

分 类 号：O415.5[理学—理论物理]