一种基于CNN的地磁异常信息地震预测算法

A Seismic Prediction Algorithm for Geomagnetic Anomaly Information Based on CNN

摘要: 地震是一种难以预测的自然灾害。震前磁异常曾被很多学者观察到过，但是还不能用它进行预测。本文尝试采用多点布设的磁强监测网络，将站点地磁异常以等值线图的形式绘制出来，并基于卷积神经网络模型VGGNet构建了一种地震预测模型VGG-12，该模型可以对地磁异常等值线图进行分析，挖掘其中的地震信息，进而实现地震发生方位和等级预测的目的。将该模型在具有类别的地磁异常等值线图像构成的数据集上训练和预测，结果显示，该模型的预测准确率可以达到75%以上。

Abstract: Earthquake is an unpredictable natural disaster. Magnetic anomalies before earthquakes have been observed by many scholars, but they cannot be used for prediction. This article attempts to use a multi-point magnetic monitoring network to plot the geomagnetic anomalies at the site in the form of contour maps, and constructs an earthquake prediction model VGG-12 based on the convo-lutional neural network model VGGNet. This model can analyze the contour maps of geomagnetic anomalies, mine the seismic information, and achieve the purpose of earthquake prediction. The model was trained and predicted on a dataset composed of geomagnetic anomaly contour images with categories, and the results showed that the prediction accuracy of the model could reach over 75%.