众所周知，测量值中含有粗差，则平差所得的结果不是最优的。然而从观测数据自身是无法辩别哪个是粗差的，因此需要采取一定的方法或措施来识别出数据中的粗差位置或在平差过程中减弱、乃至是消除其所带来的影响。目前，对于测量数据中存在粗差的处理大致可分为两类：第一类是把粗差当作函数模型误差的粗差探测法，也称为平均漂移模型探测，比如，巴尔达数据探测法、t法、标准正态检验法和F-T法等将粗差数据识别出来。第二类是把粗差作为随机模型的方差扩大模型，如岭估计法、稳健估计和选权迭代法等减弱或消除粗差在平差中的影响。 应指出上述方法都是基于高斯-马尔科夫模型(G-M模型)取得成效的，但是，随着数据采集的工具越来越先进与观测环境的可塑性，随机误差不再以加法的形式影响测量值而是以乘性或加乘性混合扰动。如果我们仍采用G-M模型发展而来的粗差探测理论与方法来进行粗差识别或减弱其在平差中的影响，能否达到我们预期的效果呢？ 本文的主要内容是介绍并分析上述第一类当中较为常见的几种方法的适用条件及其缺点，然后详细推导出在乘性误差随机模型下的标准正态检验法、t法及F-T法的表达式。再通过实验，验证其可行性。我们发现：当先验单位权中误差已知时，无论样本数据中存在多少个粗差，标准正态检验法均能够检验出粗差的位置，但有时会因为粗差量较大或多个粗差的存在，对参数影响较大，导致其它点产生误判，即弃真错误。当先验单位权中误差未知时，采用偏差改正的t法只能判断一个粗差的位置，当数据中存在多个粗差时，t法只能探测到部分粗差的位置，存在纳伪错误。那么再结合F-T法是能够探测到与实际相符的粗差位置，从而有效的避免弃真和纳伪错误的产生。就实验结果分析可得出的结论是采用上述的标准正态检验法、t法及F-T法来进行乘性误差随机模型的粗差探测是可行的。

It is well known that a outlier in the observed value, the result of the adjustment is not optimal. However, it is unable to identify the gross error from the data itself. So we need to take some methods or measures to identify the position of the outlier data and to weaken or even eliminate the effect in the process of adjustment. At present, the processing of outlier in measurement data can be roughly divided into two categories. The first category is outlier detection method that taking the outlier as a function model error, which is also known as the average drift model detection, such as Barrda Data detection method, t method, standard normal test method and F-T method, of which will identify the outlier data. The second category is the variance expansion model that taking outlier as the stochastic model, such as ridge estimation, Robust estimation and selection iterative method of which will reduce or eliminate the impact of outlier margin in the adjustment. What should be pointed out is that the methods above are all succeed based on the Gauss - Markov model(G-M model). However, as the data acquisition tool becoming more advanced and the observing environment more plasticity, the random error will no longer affect the observed value in the form of addition, but with the multiplicative or additive mixed disturbance. If we still use outlier detection theory and method developed from the G-M model to do the error recognition or reduced its influence in the adjustment, .Can we achieve the effect we expected? The main content of this article is to introduce the applicable condition and shortcomings of the first category methods and to derive the expression of standard normal test method, the t method and the F-T method under the random model of multiplicative error. Then, through the experiment to verify its feasibility. The findings are as follows: when the error in the a priori unit is known, no matter how many outliers exist in the sample data, the standard normal test method can detect the position of the outlier. But sometimes the existing of large a outlier or multiple outliers has a great impact on the parameters that will lead misjudgment of other points. When the error in the a priori unit is unknown, the deviation of the t method can only be used to determine the position of the outlier. When there are multiple outliers in the data, t method can only detect the location of the part of the outliers, there is a pseudo-false. Then combined with the F-T method, it will be able to detect the actual outlier match with the actual location, that will effectively avoid the mistakes happening of true and false error. Based on the experimental results, we can conclude that using the standard normal test method, t method and F-T method to carry out outlier detection of multiplicative error random model is feasible.