基于深度置信网络的卡尔曼滤波算法改进

IMPROVEMENT OF KALMAN FILTER ALGORITHM BASED ON DEEP BELIEF NETWORK

卡尔曼滤波是一种应用广泛的基于最小方差的递推式滤波算法,根据一定滤波规则对系统的状态进行估计。采用某种统计量最优方法对噪声和系统模型统计特性的先验知识决定的滤波的性能和估计的准确性进行度量。不精确的先验知识将导致滤波性能的明显下降和发散。采用新息自适应卡尔曼滤波克服标准卡尔曼滤波需要在先验条件下进行估计的缺点,通过深度置信网络对噪声的协方差矩阵做出调整,从而提高滤波性能。

Kalman filtering is a widely used recursive filtering algorithm based on minimum variance,which estimates the state of the system according to certain filtering rules.We used a statistical optimum method to measure the filtering performance and estimation accuracy determined by prior knowledge of the statistical characteristics of noise and system models.The inaccurate prior knowledge could lead to a significant drop and divergence of the filtering performance.The innovation adaptive Kalman filtering overcame the disadvantage of the standard Kalman filtering that needed to be estimated under prior conditions.The covariance matrix of noise was adjusted by the deep belief network to improve the filtering performance.