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.
展开▼