U-D分解的前向固定区间平滑新算法

本文提出了两种前向固定区间平滑新算法以解决工程问题.为了确保算法的数值稳定性并提高计算效率,两种算法中的协方差矩阵传播均使用了U-D分解形式.计算量分析结果表明,两种新算法与Keigo Watanabe前向平滑算法相比较,计算量减少40%以上;状态维数较高时,计算效率提高3倍以上.

A New U-D Factorization-Based Forward-Pass Fixed-Interval Discrete-Time Smoother

To solve the problem of time-history arrangement of test data, two kinds of forward-passfixed-interval discrete-time smoothers are presented. These smoothers are based on the Kalmanfilter, Rauch-Tung-Streible smoother, and Keigo Watanabe forward-pass smoother. To gethigh numerical stability and computational efficiency, U-D factorizations are used in the propagationsof covariance matrices of these new smoothing algorthms. For different systems, oneof the new smoothers would be chosen easily by the comparison of operation counts. These newsmoothers exhibit excellent numerical accuracy and stability and the number of operations ofboth the filter and smoother are decreased greatly. Comparison of operation numbers showsthat these new smoothers are more than 1.8 times as faster as Keigo. Watanabe's forward-passsmoother.