A Feature Extraction Method for Daily-periodic Time Series Based on AETA Electromagnetic Disturbance Data

Abstract

Seismic monitoring data and some geophysical data often show time series data with daily-periodic amplitude jump. In this kind of time series, the sharp change of amplitude will naturally become the focus of attention, but at the same time, it is easy to ignore the information beyond the dramatic change of amplitude. Therefore, this paper proposes a feature extraction method for this kind of time series data, which can help obtain important information from data other than the dramatic changes in magnitude. By comparing the feature importance between the feature obtained by the proposed feature extraction method and the original feature, it shows that the new feature importance is 1.38 times higher than the original feature importance on average, which also proves the effectiveness of the proposed method.