Missing Data Imputation for Operation Data of Transformer Based on Functional Principal Component Analysis and Wavelet Transform

摘要

The operation data of transformer plays an important role in assessing the status of electric devices. However, the real operation data often have missing values, which will result in the unreliability of the following data analysis. In view of this, we propose a method based on functional principal component analysis (FPCA) and wavelet transform to impute the missing data. According to the characteristic of the operation data, we separate the daily data curve to low frequency part and high frequency part. We first investigate the fluctuation patterns over the historical data and use FPCA to estimate the low frequency part. We then estimate the residual function and use wavelet transform to estimate the high frequency part. Combining these two parts, we get the approximation to the original data and impute the missing values. Applications of the proposed method to the real operation data show that the method performs very well and it works well whenever the missing points are randomly distributed or in continuous form.