基于半参数均值漂移模型的BDS卫星钟差异常探测与修复

Detection and repair of outliers in BDS satellite clock offset based on semiparametric mean drift model

BDS卫星钟在轨运行过程中,由于空间环境干扰及信号异常等情况,获取的钟差数据中经常会包含异常值。为了有效探测与修复异常值,提出了一种基于半参数均值漂移模型的算法处理钟差数据中的异常。首先,考虑到钟差数据中存在的系统误差,构建了更加符合钟差序列实际变化特点的半参数钟差模型;然后在此基础上引入了半参数均值漂移模型,利用Score统计量对该模型的假设检验进行求解,实现了异常值的定位,对异常定位后的钟差数据进行分组,进而推导出了异常值估值的表达式。最后采用国际GNSS监测评估系统的BDS精密钟差产品,进行单天和多天数据的试验分析,结果表明该算法可以自动准确的探测出钟差数据中的异常并能够精确估计出异常值的大小,异常值估值的平均误差小于0.071 ns,从而实现了适时有效地对异常历元进行修复;同时该算法处理钟差数据异常的效果优于传统MAD法,其预处理后的数据对QP钟差模型和Semi-Com钟差模型的拟合精度分别达到0.612 6和0.034 92 ns。

During the operation of BDS satellite clocks, due to the space environment interference and signal abnormality, the acquired clock offset data often contain abnormal values. In order to effectively detect and repair the outliers, an algorithm based on semiparametric mean drift model is proposed to deal with the outliers in clock offset data. Firstly, considering the systematic errors existing in the clock offset data, a semiparametric clock offset model that is more in line with the actual variation characteristics of the clock offset sequence is constructed. Then, based on this, a semiparametric mean drift model is introduced, the Score statistics is used to solve the hypothesis test and the localization of the outliers is achieved. The clock offset data after outlier localization are grouped, and then the expressions of the outlier estimation are derived. Finally, the BDS precision clock offset products in the international GNSS monitoring and assessment system are used for the test analysis of single-day and multi-day data. The results show that the algorithm can automatically and accurately detect the outliers in clock offset data, and can accurately estimate the size of the outliers. The average error of the outlier estimation is less than 0.071 ns, thus the timely and effective repair of abnormal epochs is realized. At the same time, the algorithm is better than the traditional MAD method in processing the outliers in clock offset data. The fitting accuracies of the preprocessed data to the QP clock offset model and the Semi-Com clock offset model reach 0.612 6 and 0.034 92 ns, respectively.