Real-time neural network earthquake profile predictor

• US 5,490,062 A
• Filed: 05/11/1994
• Issued: 02/06/1996
• Est. Priority Date: 05/11/1994
• Status: Expired due to Term

• Current Assignee: Lawrence Livermore National Security LLC (Government of the United States of America), US Department of Energy (Government of the United States of America)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Leach, Richard R., Dowla, Farid U.
• Primary Examiner(s): McElheny, Jr., Donald E.
• Application Number:
  US08/241,060
• Time in Patent Office: 636 Days
• Field of Search: 364/421, 395/929
• US Class Current: 702/15
• CPC Class Codes: G01V 1/01 Measuring or predicting earthquakes

Abstract

A neural network has been developed that uses first-arrival energy to predict the characteristics of impending earthquake seismograph signals. The propagation of ground motion energy through the earth is a highly nonlinear function. This is due to different forms of ground motion as well as to changes in the elastic properties of the media throughout the propagation path. The neural network is trained using seismogram data from earthquakes. Presented with a previously unseen earthquake, the neural network produces a profile of the complete earthquake signal using data from the first seconds of the signal. This offers a significant advance in the real-time monitoring, warning, and subsequent hazard minimization of catastrophic ground motion.

Claims

1. A method for predicting an earthquake profile, comprising:

training an artificial neural network (ANN), comprising;

inputting at least one seismogram into the ANN to produce an output;

calculating an error between the ANN output and the seismogram;

adjusting the internal weights of the ANN to reduce the error in the calculating step;

repeating the preceding steps until the error is reduced to a level where the ANN can generalize sufficiently to produce an accurate output for a previously unseen input;

detecting an earthquake with three mutually orthogonal ground motion detectors which will produce real-time earthquake data;

transmitting into the ANN the real-time earthquake data from the three mutually orthogonal ground motion detectors; and

feedforwarding the real-time earthquake data in the ANN to produce a predicted earthquake profile.

2. The method of claim 1, wherein the inputting step includes producing the seismogram by filtering and envelope detecting a seismic signal to provide time series and frequency information.

3. The method of claim 1, wherein the step of inputting at least one seismogram into the ANN to produce an output includes inputting a plurality of seismograms into an ANN selected from a group consisting of a self-organizing Kohonen network and a backpropagation network.

4. The method of claim 1, wherein the step of inputting at least one seismogram includes inputting at least one seismogram stored in a United States Geological Society database.

5. The method of claim 1, wherein the step of detecting an earthquake includes tightly coupling to the earth the three mutually orthogonal ground motion detectors.

6. The method of claim 1, wherein the step of detecting an earthquake includes configuring the three mutually orthogonal ground motion detectors to monitor ground motion in the range of plus or minus 2 g.

7. The method of claim 1, wherein the step of detecting an earthquake includes powering the detectors by battery.

8. The method of claim 1, wherein the step of detecting an earthquake includes power conditioning the real-time earthquake data to provide a linear representation of ground acceleration.

9. The method of claim 1, wherein the step of transmitting earthquake data includes transmitting earthquake data when ground acceleration exceeds a preset threshold.

10. The method of claim 1, wherein the step of transmitting earthquake data includes transmitting earthquake data when ground acceleration exceeds a preset threshold of 0.5 g.

11. The method of claim 1, wherein the step of transmitting earthquake data includes transmitting earthquake data for a preset time period when ground acceleration exceeds a preset threshold.

12. The method of claim 1, wherein the step of transmitting earthquake data includes transmitting earthquake data for a preset time period of 5 minutes when ground acceleration exceeds a preset threshold.

13. The method of claim 1, wherein the step of transmitting earthquake data includes feeding the real-time earthquake data through an anti-aliasing filter to produce a filtered signal, said filtered signal then being fed into an analog-to-digital converter which digitizes the filtered signal to produce a digitized signal, said digitized signal then being fed into the ANN which resides in a workstation computer.

14. The method of claim 1, wherein the step of transmitting earthquake data includes pre-processing the real-time earthquake data in a computer to produce pre-processed data, said pre-processed data then being fed into the ANN which resides in the computer.

15. The method of claim 1, wherein the step of transmitting earthquake data includes filtering the real-time earthquake data with a third-order Butterworth bandpass filter having corner frequencies at 0.5 Hz and 3 Hz to produce filtered preprocessed data, performing an envelope function by using a Hilbert transform on said filtered pre-processed data to produce enveloped data, and feeding the enveloped data into the ANN which resides in a computer.

16. The method of claim 1, wherein the step of feedforwarding the real-time earthquake data in the ANN includes continuously updating the profile as the signal continues to be detected.

17. The method of claim 1, further comprising the step of activating a warning system if the magnitude of the profile produced in the feedforwarding step exceeds a preset threshold.

18. The method of claim 1, further comprising the step of activating a warning system if the magnitude of the profile produced in the feedforwarding step exceeds a preset threshold of 5.0 on the Richter scale.

19. The method of claim 1, further comprising the step of activating automatic shutoff valves on gas lines if the magnitude of the profile produced in the feedforwarding step exceeds a preset threshold.

20. The method of claim 1, further comprising the step of activating automatic shutoff of electrical lines if the magnitude of the profile produced in the feedforwarding step exceeds a preset threshold.

21. The method of claim 1, further comprising the step of activating audible warnings if the magnitude of the profile produced in the feedforwarding step exceeds a preset threshold.

22. The method of claim 1, wherein the step of inputting at least one seismogram into the ANN to produce an output includes inputting a plurality of seismograms into a self-organizing Kohonen network which clusters the seismograms into groups and orders the seismograms according to an unsupervised feature extraction rule.

23. The method of claim 1, wherein the step of detecting an earthquake with three mutually orthogonal ground motion detectors includes detecting an earthquake with an east-west detector, a north-south detector and an up-down detector.

24. The method of claim 1, wherein the step of training an ANN includes performing the inputting step with the same seismogram at least 10 times before inputting a new seismogram.

25. The method of claim 1, wherein the step of training an ANN includes training the ANN with backpropagation learning for a multilayer feed-forward neural network using a conjugate gradient method.

26. The method of claim 1, wherein the step of feedforwarding the real-time earthquake data includes calculating the integral of a portion of the predicted earthquake profile to produce a magnitude estimate.

27. An apparatus for predicting an earthquake profile, comprising:

at least one ground motion detector, comprising three mutually orthogonal components, wherein each component is identical and is selected from a group consisting of seismometers, accelerometers and geophones, said detector being battery powered and producing an output signal;

an amplifier to receive said output signal and produce a three channel amplified analog signal;

an analog-to-digital (A-D) converter board comprising at least three channels electrically configured to receive said three channel amplified analog signal;

a computer to house said A-D converter board, said computer comprising an event detection algorithm, wherein an incoming seismic signal is preprocessed to produce a seismogram having time series and frequency information; and

an artificial neural network which feeds forward said seismogram to produce a complete earthquake profile.