遗传算法在特异值识别与处理中的应用

The use of genetic algorithm in the identification and handling of outliers

在空间数据中特异值的识别处理中,个别奇异的数据(特异值)可能会对模型产生影响而使预测值与实测值间产生较大的偏差。因此,本文在分析现有的方法的基础上,考虑识别过程中可能存在的"拖尾"和"掩盖"效应,根据"同步"的思想提出了一种新的特异值识别方法:基于遗传算法的特异值识别,该方法以数据标准化差异作为评价标准,将遗传算法应用到特异值识别处理当中,能够有效地识别出观测值中的异常数据,获得一致性数据。

The individual strange data （outlier） may possibly have the influence on the model and result in a big deviation between the forecast value and the actual value. Therefore, we consider the possibility of smearing and masking in the recognition process, and propose a new method according to simultaneity by analyzing existing method, which is called genetic algorithm-based outlier recognition. This method takes the data standardization difference as evaluation criteria, and introduces the genetic algorithms to outlier recognition processing. It can treat outlier effectively and obtain the uniform data.