Kalman滤波异常误差检测

为检测动态导航观测异常和动态模型异常误差，本文利用状态方程预测残差二次型构造了整体误差检验法，即观测误差和动力学模型误差整体检验法；讨论了三种观测异常检测法，即以模型为基准的观测异常检验，以当前历元可靠观测为基准的异常检验，以状态Kalman滤波估值为基础的观测异常检验；分析了三种动力模型异常检测法，即状态不符值检验法，以状态参数Kalman滤波估值为基础的动力模型误差检验法，以可靠观测为基础的动力模型误差整体检验法。并对这几种异常检测法进行了简单分析。

Statistics for Outlier Detection in Kalman Filtering

In order to detect outliers of measurements and dynamic models in Kalman filtering, an integral error detection statistic which corresponds to the integral effect of measurement errors and dynamic model errors, is constructed by the quadratic form of predicted residual vector. Three types of statistics for detecting the measurement outliers are studied which are based on dynamic model information, reliable measurements at present epoch, and the estimates of Kalman filtering respectively. Three types of statistics for detecting dynamic model errors correspond to the discrepancies of states, the estimates of Kalman filtering, and the reliable measurements are also established and analyzed.