Method and system to predict the extent of structural damage

• US 10,740,684 B1
• Filed: 08/25/2016
• Issued: 08/11/2020
• Est. Priority Date: 12/09/2015
• Status: Active Grant

• Current Assignee: One Concern, Inc.
• Original Assignee: One Concern, Inc.
• Inventors: Wani, Ahmad, Hu, Nicole, Frank, Timothy
• Primary Examiner(s): Cole, Brandon S
• Application Number:
  US15/246,919
• Time in Patent Office: 1,447 Days
• Field of Search: 706 1- 62
• CPC Class Codes: G01V 1/01 Measuring or predicting earthquakes
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Methods, systems, and computer programs are presented for predicting the scale and scope of damage after an earthquake. One method includes an operation for identifying a plurality of features, each feature being correlated to an indication of structural damage caused to a structure by an earthquake. The method further includes performing machine learning, using one or more hardware processors, to analyze destruction caused by one or more earthquakes to obtain a damage-estimation algorithm. The machine learning is based on the identified plurality of features. Further, the method includes operations for accessing shaking data for a new earthquake, and for estimating, using the one or more hardware processors, earthquake damage at a block level for a geographical region utilizing the damage-estimation algorithm and the shaking data. Further, the earthquake damage at the block level is presented, on a display screen, in a map of at least part of the geographical region.

Claims

1. A method comprising:

identifying a plurality of features for an earthquake-damage estimation algorithm, the features comprising;

destruction information for structures within a geographical region caused by an earthquake that happened in the past;

earthquake data comprising duration of shaking, epicenter location, and spectral acceleration of the earthquake; and

fragility curves for structures in the geographical region, the fragility curves based on construction material, seismic zone, and seismic design code;

preparing a training set k including values for the identified plurality of features for one or more earthquakes;

training a machine-learning model, using one or more hardware processors, with the training set to obtain the earthquake-damage estimation algorithm, the training of the machine-learning model being based on the identified plurality of features;

accessing new earthquake data for a new earthquake, the new earthquake data comprising duration of shaking of the new earthquake, epicenter location of the new earthquake, and spectral acceleration of the new earthquake;

estimating, using the one or more hardware processors, earthquake damage at a block level for the blocks in the geographical region utilizing the earthquake-damage estimation algorithm and the new earthquake data, the earthquake damage for the block representing a most probable damage state of a building in the block from a plurality of damage states; and

causing presentation, on a display screen, of the earthquake damage at the block level in a map of at least part of the geographical region.

2. The method as recited in claim 1, wherein estimating the earthquake damage at the block level further comprises:

for each block in the geographical region, determining values of the features of the structures in the block; and

estimating the earthquake damage at the block level based on the values of the features of the structures in each block.

3. The method as recited in claim 1, wherein the plurality of features for training the machine-learning model are classified into built environment, natural environment, and instantaneous line, wherein the built environment is for structures that have been built, wherein the natural environment is for structures occurring in nature, wherein the instantaneous line comprises the earthquake data.

4. The method as recited in claim 3, wherein the built environment features for training the machine-learning model comprise structure location, year built, number of stories, commercial or residential structure, and building material of structure.

5. The method as recited in claim 3, wherein the natural environment features for training the machine-learning model comprise soil type, soil density, elevation, and water table.

6. The method as recited in claim 3, wherein the instantaneous line features for training the machine-learning model comprise the duration of shaking, the epicenter location, the spectral acceleration, and spectral displacement.

7. The method as recited in claim 1, further comprising:

determining an accuracy of a plurality of machine-learning models; and

selecting one or more machine-learning models and associated tuning parameters to apply, in parallel or in sequence, based on the machine-learning models respective accuracy.

8. The method as recited in claim 1, wherein accessing the new earthquake data further comprises:

receiving the new earthquake data as ShakeMap information from United States Geological Survey after the new earthquake, wherein the earthquake damage at the block level is based on the received ShakeMap information.

9. The method as recited in claim 8, wherein accessing the new earthquake data further comprises:

accessing damage inspection data after the new earthquake;

inputting the accessed damage inspection data to the earthquake-damage estimation algorithm; and

re-estimating the earthquake damage after inputting the accessed damage inspection data.

10. The method as recited in claim 1, wherein each block corresponds to a census block that has been defined by the United States Census Bureau.

11. The method as recited in claim 1, wherein the earthquake damage at the block level is presented within a user interface of a disaster-response-center module.

12. The method as recited in claim 1, further comprising:

estimating financial loss of a first structure based on estimated earthquake damage to the first structure and a replacement value of the first structure.

13. The method as recited in claim 1, wherein training the machine-learning model further comprises:

applying a transfer-learning operation to apply data from an earthquake event in a first geographical area to an earthquake event in a second geographical area.

14. The method as recited in claim 1, wherein estimating the earthquake damage comprises classifying each block at one of four damage states, each damage state associated with a level of damage caused by the earthquake in the block.

15. A non-transitory machine-readable storage medium including instructions that, when executed by a machine, causes the machine to perform operations comprising:

identifying a plurality of features for an earthquake-damage estimation algorithm, the features comprising;

destruction information for structures within a geographical region caused by an earthquake that happened in the past;

earthquake data comprising duration of shaking, epicenter location, and spectral acceleration of the earthquake; and

fragility curves for structures in the geographical region, the fragility curves based on construction material, seismic zone, and seismic design code;

preparing a training set b including values for the identified plurality of features for one or more earthquakes;

training a machine-learning model with the training set to obtain the earthquake-damage estimation algorithm, the training of the machine-learning model being based on the identified plurality of features;

accessing new earthquake data for a new earthquake, the new earthquake data comprising duration of shaking of the new earthquake, epicenter location of the new earthquake, and spectral acceleration of the new earthquake;

estimating earthquake damage at a block level for the blocks in the geographical region utilizing the earthquake-damage estimation algorithm and the new earthquake data, the earthquake damage for the block representing a most probable damage state of a building in the block from a plurality of damage states; and

causing presentation, on a display screen, of the earthquake damage at the block level in a map of at least part of the geographical region.

16. The machine-readable storage medium as recited in claim 15, wherein estimating earthquake damage at the block level further comprises:

for each block in the geographical region, determining values of the features of the structures in the block; and

estimating the earthquake damage at the block level based on the values of the features of the structures in each block.

17. The machine-readable storage medium as recited in claim 15, wherein the plurality of features for training the machine-learning model are classified into built environment, natural environment, and instantaneous line, wherein the built environment is for structures that have been built, wherein the natural environment is for structures occurring in nature, wherein the instantaneous line comprises the earthquake data.

18. A system, comprising:

a memory comprising instructions; and

one or more computer processors, wherein the instructions, when executed by the one or more computer processors, cause the one or more computer processors to perform operations comprising;

identifying a plurality of features for an earthquake-damage estimation algorithm, the features comprising;

destruction information for structures within a geographical region caused by an earthquake that happened in the past;

earthquake data comprising duration of shaking, epicenter location, and spectral acceleration of the earthquake; and

fragility curves for structures in the geographical region, the fragility curves based on construction material, seismic zone, and seismic design code;

preparing a training set by including values for the identified plurality of features for one or more earthquakes;

training a machine-learning model with the training set to obtain the earthquake-damage estimation algorithm, the training of the machine-learning model being based on the identified plurality of features;

accessing new earthquake data for a new earthquake, the new earthquake data comprising duration of shaking of the new earthquake, epicenter location of the new earthquake, and spectral acceleration of the new earthquake;

estimating earthquake damage at a block level for the blocks in the geographical region utilizing the earthquake-damage estimation algorithm and the new earthquake data, the earthquake damage for the block representing a most probable damage state of a building in the block from a plurality of damage states; and

causing presentation, on a display screen, of the earthquake damage at the block level in a map of at least part of the geographical region.

19. The system as recited in claim 18, wherein estimating earthquake damage at the block level further comprises:

for each block in the geographical region, determining values of the features of the structures in the block; and

estimating the earthquake damage at the block level based on the values of the features of the structures in each block.

20. The system as recited in claim 18, wherein accessing new earthquake data further comprises:

accessing the new earthquake data as ShakeMap information from United States Geological Survey after the new earthquake, wherein the earthquake damage at the block level is based on the received ShakeMap information;

accessing damage inspection data after the new earthquake;

inputting the damage inspection data to the earthquake-damage estimation algorithm; and

re-estimating earthquake damage after inputting the damage inspection data.