← Back to Sessions

Prediction of Seismicity in Sichuan-Yunnan Region based on Several Kinds of Recurrent Neural Networks

Session: New Insights Into the Preparatory Phase of Earthquakes From Tectonic, Field and Lab Experiments [Poster]

Type: Poster

Date: 4/19/2021

Presentation Time: 03:45 PM Pacific

Description: 

With the rapid development of artificial intelligence, a large number of application examples in the field of geophysics and seismology have emerged using the method of machine learning in recent years. Neural network as the commonly used method of machine learning was put forward very early and has now developed into a fairly large, multidisciplinary cross disciplines with the progress of computer performance. Inspired by the central nervous system of animals, neural networks can solve a wide variety of tasks that are difficult to solve with ordinary rule-based programming by adaptively learning large amounts of inputs, and have played a huge role in areas such as computer vision, speech recognition, and weather forecasting. The biggest advantage of artificial neural network is that it can approximate any function, which is very suitable for studying the complex empirical relationship between seismic activity and subsequent strong earthquakes. Meanwhile, the state of seismic activity changes dynamically with time, and the seismic activity in two adjacent periods is related to each other to some extent. This characteristic can be modeled properly using the recursive neural network (RNN) as it has memories of a sequence data and it can better reflect the characteristics of time dependence of seismic activities. Three different kinds of neural networks including Elman RNN, long short-term memory (LSTM) RNN, and gated recurrent unit (GRU) RNN have been applied to predict the seismicity in Sichuan-Yunnan region based on feature values which can represent the spatial and temporal characteristics of previous regional small earthquakes such as G-R relationship, seismic quasi-period, seismic energy, etc. This research provides an alternative approach for evaluating the short to long term seismic hazard potential of a study region and it has a broader space for development compared to traditional empirical earthquake prediction methods.

Presenting Author: Yang Zang

Student Presenter: No

Authors