Imputation Method of Missing Values for Dissolved Gas Analysis Data Based on Iterative KNN and XGBoost

Power transformers are an important part of the power system. Accurate monitoring of its operating status is particularly important for the normal and stable operation of the entire power system and the timely diagnosis of potential faults. Dissolved Gas Analysis (DGA) can detect and judge the oil-immersed power transformer failure by comparing the dissolved gas content of the power transformer in the normal operating state and the oil in the fault state. However, in the operation process of the grid transformer, the detection data is often missing. This paper proposes an effective method based on iterative KNN and XGBoost method for missing values. Firstly, according to the XGBoost integration tree, there are missing values. Information such as the number of attribute divisions obtained by data set training calculates the importance scores of different attributes to determine the priority of the attributes, and then performs interpolation on the missing values ?in an iterative manner. The experimental results in the case of DGA dataset and different missing rate show that the proposed method is superior to the existing similar methods in accuracy, and the dataset after interpolation has a significant improvement on the classification effect of the classifier.