Short-term comprehensive prediction method for regional earthquakes based on multi-source information fusion

To solve the problem of short-term earthquake prediction, this study is in view of multi-source information fusion, using precursor observation data from 11 measurement items as experimental data. Then the study preprocesses it and takes it as the input of Convolutional neural network (CNN). After the network structure design, three types of CNN are obtained, through which the earthquake location and magnitude can be simultaneously predicted. The results show that the third kind of CNN has a good prediction performance. In Earthquake prediction, time window, network structure, and data preprocessing methods will affect the performance of the algorithm. Only one group of feature extraction layers of CNN has the best prediction effect. The fixed time window is 240 h, with a higher accuracy of 93.5%; Under this window, after principal component processing, its accuracy is 96.4%. Compared with autoencoder and other algorithms, the third CNN has the highest accuracy and recall, 95.0 and 84.1% respectively. Research methods can accurately predict the area and magnitude of earthquakes.