基于证据理论的迭代多步预测方法研究  

Iterative multi-step prediction model based on theory of evidence

作　　者：洪贝[1] 胡昌华[1] 姜学鹏[2] 

机构地区：[1]解放军第二炮兵工程学院302教研室,陕西西安710025 [2]解放军海军航空工程学院701教研室,山东烟台264001

出　　处：《控制理论与应用》2010年第12期1737-1742,共6页Control Theory & Applications

基　　金：国家自然科学基金资助项目(60736026);国家"863"创新基金资助项目(2008AAJ211);教育部新世纪优秀人才支持计划资助项目(NCET-07-0144)

摘　　要：为了有效利用复杂非线性工程系统中各种不确定信息和主观知识,本文结合相空间重构理论和证据推理理论,建立时序证据的多步预测模型.本文先对时序证据进行相空间重构;针对时序证据预测中信息量大且存在误差累积的不足,采用时序证据滑动窗口处理方法,以降低累积误差;窗口内运用迭代融合规则,以降低信息量的存储.将该模型应用于某型陀螺仪漂移预测,试验结果证明,本文方法能够有效降低迭代时产生的累积误差和信息量,结论更接近实际情况.To make use of the uncertain information and subjective knowledge in complex nonlinear systems,we propose a multi-step prediction method for time series of evidence based on the phase space reconstruction theory and evidence inference theory.In the reconstruction of phase space for time series of evidence,we use the sliding window for time series of evidence to deal with the great amount of information and the shortage of accumulation errors to reduce the accumulated errors.The information storage is reduced by applying the iterative rules to the windows.The proposed methods have been applied to the simulation tests of a missile inertia device for fault prediction.The results show the effectiveness of these methods.

关 键 词：证据理论 迭代多步预测 滑动窗口 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]