飞参数据挖掘中的预处理算法研究

Research on the Preprocessing Algorithms of Flight Data Mining

针对飞参数据的特点,从野值剔除和属性选择两个方面对飞参数据预处理进行了研究,提出了基于残差检验的野值剔除方法和基于神经网络的两阶段属性选择方法。在野值剔除方法中,首先计算原始序列和拟合序列的残差,再用门限值对残差进行检验;在属性选择方法中,先用神经网络对属性的相对重要度进行排序,再用神经网络进行属性选择。最后用实验验证了两方法的有效性。

With regard to the characteristics of flight data, outlier elimination and feature selection are studied respectively as the two important sides of flight data preprocessing. Two new methods are proposed, one is the outlier elimination method based on error detection, and the other is the two-phase feature selection method using neural network. The error vector between the initial sequence and the smooth sequence is used to judge outliers in the outlier elimination method, when the error is over the detection threshold. In the two-phase feature selection method, the feature importance is ranked by neural network, and then the important features are selected using neural network. Finally the effectiveness of the two methods is shown by the test data.