绝对重力测量异常值的局部异常因子检测算法

Local anomaly factor detection algorithm for absolute gravity measurement outliers

自主研发绝对重力仪的测量结果中出现的离群程度不同的异常值会直接影响测量结果的准确度和测量精度。目前一般采用的一元正态分布异常值检测算法漏检率高,容易造成测量结果的偏差和测量精度的下降。利用人工智能算法中的局部异常因子异常值检测算法,可以在线、快速、高效地完成自主研发绝对重力测量数据的异常值检测。首先,根据实测数据构建测试数据集,利用数值模拟确定局部异常因子算法邻域宽度参数的取值;然后,基于实测数据进行异常值检测并进行结果评估。评估结果表明,局部异常因子异常值检测算法对离群程度不同、连续出现异常值等情况检测效果明显优于一元正态分布异常值检测算法,组测量精度平均提高9.37μGal,可以作为自主研发绝对重力仪异常值检测的通用算法完成组测量结果的异常值检测。

The outliers in the measurement results of an absolute gravimeter will directly affect the measurement accuracy,which is generally detected by the one-dimensional normal distribution outliers(NDO)detection algorithm at present.A local outlier factor(LOF)detection algorithm in artificial intelligence technology is introduced to complete the outlier detection of an independently developed absolute gravity measurement data.Firstly,the test data set is constructed according to the measured data,and numerical simulation is made to determine the neighborhood width parameter value of the LOF algorithm.Then,the outliers are detected and evaluated based on the measured data.Test results show that the proposed algorithm can complete the the outlier detection quickly,online and accurately,and is better than the NDO algorithm,especially for the outliers in different levels,continuous outliers,etc.The group measurement precision is increased by 9.37μGal on average,and the LOF algorithm can be used as a general algorithm of outlier detection in the independently developed absolute gravimeter.