Skip to content

Improving EEW system performance with machine learning and convolutional neural network.

Notifications You must be signed in to change notification settings

mhndda/Earthquake-Early-Warning

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Earthquake-Early-Warning

Improving Earthquake early warning (EEW) system performance with machine learning and convolutional neural network.

The goal of an EEW system is to rapidly predict earthquake magnitudes using just a few seconds of ground motion time series, such that warnings can be issued before strong ground motion arrives at densely populated areas. Currently deployed EEW systems around the world, relying on several empirical measurements, suffer from either high false positive rate or underestimating magnitudes of large earthquakes. In this project, machine learning techniques are utilized to build a robust, accurate earthuqake magnitude predictor.

Seismic waveform data have been complied and cleaned for ~10,000 earthquakes over the past three decades in southern California. More than 2000 features have been carefully extracted and engineered from the raw seismic data. Machine learning models including random forest reduce the mean squared error by roughly 50% compared to the baseline, one of the EEW models under testing in California. In addtion, a convolutional neural network is trained with the raw data as input, and achieves slightly better performance. More importantly, the baseline underestimates the large earthquake magnitudes by 0.42 units, equivalent to underestimating ground motion by a factor of 4.3. This has been reduced to 0.13-0.15 magnitude units (a factor of 1.6-1.7 for ground motion). If operating in real time, these new EEW algorithms could provide more reliable earthquake magnitude predictions for the purpose of hazard mitigation.

alt text

Figure 1. The mapview of seismic station (red triangles) and earthquake (blue dots) distribution.

About

Improving EEW system performance with machine learning and convolutional neural network.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.8%
  • Shell 0.2%