具有新目标函数的改进型变分贝叶斯  

An Improved Variational Bayesian Method with Novel Target Function

作　　者：姚智颖[1] 刘冬[1] 刘光斌[1] 

机构地区：[1]第二炮兵工程学院

出　　处：《电光与控制》2011年第8期45-46,55,共3页Electronics Optics & Control

基　　金：国家自然科学基金资助(60772151)

摘　　要：针对变分贝叶斯迭代运算中参数集分解所带来的问题,定义表观信息作为新的目标函数,提出了一种改进型的变分贝叶斯方法。该方法应用于线性高斯动态系统时,等价于Kalman滤波方法。推导过程还得出了Kalman增益的简化计算方程。To solve the problems resulted from the decomposition of parameter set during iterated calculation of variational Bayesian methods,apparent information was defined as novel target function.An improved variational Bayesian method based on apparent information was proposed in this paper.The improved method was proved to be equal to Kalman filter when used in linear Gaussian dynamic system,which is optimal.Simplified calculation of Kalman gain was derived additionally.

关 键 词：变分贝叶斯 表现信息 KALMAN滤波 

分 类 号：V247[航空宇航科学与技术—飞行器设计]