Automatic classification and recognition of geomagnetic interference events based on machine learning

Abstract

Geomagnetic interference events seriously affect normal analysis of geomagnetic observation data, and the existing manual identification methods are inefficient. Based on the data of China Geomagnetic Observation Network from 2010 to 2020, a sample data set including high voltage direct current transmission (HVDC) interference events, other interference events and normal events is constructed. By introducing machine learning algorithms, three geomagnetic interference event recognition models GIEC-SVM, GIEC-MLP, GIEC-CNN are designed based on support vector machines (SVM), multi-layer perceptron (MLP) and convolutional neural networks (CNN) respectively. The classification accuracy for each model on the test set reached 76.77%, 84.96% and 94.00%. Two optimal GIEC-MLP and GIEC-CNN are selected and applied to the identification of geomagnetic interference events at stations not participated in training and testing from January, 2019 to June, 2021. The accuracy are 72.11% and 78.24% respectively, while the efficiency is 150 times that of manual identification. It shows that the geomagnetic interference event recognition algorithm based on machine learning algorithm has high recognition accuracy and strong generalization ability, especially the CNN algorithm.