Massive data-driven method for automatically locating mine microseismic source

Patent metadata

Field	Value
Publication number	US-12536409-B2
Application number	US-202218078708-A
Country	US
Kind code	B2
Filing date	Dec 9, 2022
Priority date	Aug 24, 2022
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

Disclosed is a massive data-driven method for automatically locating a mine microseismic source, including: constructing a microseismic wave calibration data set by using a large-scale seismic data set containing seismic signals and non-seismic signals; constructing a pre-training calibration model based on a full convolution neural network through deep learning of a seismic wave calibration data set; using microseismic data of mine sites for transfer learning of an initial arrival time calibration model to construct an arrival time automatic calibration model suitable for mine microseismic signals; and automatically as well as accurately locating mine microseismic events based on an isokinetic homogeneous isotropic velocity model by using an optimization algorithm to deduce arrival time errors and through repeated iteration and fine-tuning.

First claim

What is claimed is:

1 . A data-driven method for automatically locating a mine microseismic source, comprising:
S1: screening a seismic waveform data from a seismic data set containing seismic signals and non-seismic signals, and preprocessing the seismic waveform data to construct a seismic wave calibration data set; S2: randomly dividing the seismic waveform data in the seismic wave calibration data set into a training set, a verification set and a test set according to a preset proportion; then, establishing a deep neural network model with a U-net structure; and training the deep neural network model by the training set to obtain a pre-training calibration model, wherein the deep neural network model has four down-sampling phases and four up-sampling phases, wherein the four down-sampling phases are used to extract effective wave characters for arrival time calibration from original seismic data, the four up-sampling phases are used to expand the wave characters and convert the wave characters into probability distributions corresponding to P-wave arrival time, S-wave arrival time and noise at corresponding data points, and layers corresponding to the four down-sampling phases and layers corresponding to the four up-sampling phases are connected by jumps; a size of a convolution kernel is set to 7 and a step size of the convolution kernel is set to 4; in a convolution operation, padding is performed before and after each layer, so that input sequences and output sequences have a same length; in a last layer of the deep neural network model, probabilities of the P-wave arrival times, the S-wave arrival times and the noise are output by a Softmax normalized exponential function defined by formula (1), wherein a cross entropy loss function defined by formula (2) is used for training the deep neural network model by the training set, and the cross entropy loss function is ensured to be minimal;

q
i

(
x
)

=

e

z
i

(
x
)

∑

k
=
1

3

⁢

e

z
k

(
x
)

,

(
1
)

wherein Z i (x) is unscaled values of the last layer, and x is data points, wherein i is set to 1, 2, 3, with i=1 representing the noise, i=2 representing the P-wave arrival times, and i=3 representing the S-wave arrival times, and;

H
⁡
(

p
,
q

)

=

-

∑

i
=
1

2

∑
x

p
i

(
x
)

⁢
log
⁢

q
i

(
x
)

,

(
2
)

wherein p i (x) is real probability distribution and q i (x) is predictive probability distribution, wherein i is set to 1, 2, 3, with i=1 representing the noise, i=2 representing the P-wave arrival times, and i=3 representing the S-wave arrival times, and x is data points;
S3: using microseismic data of mine sites for transfer learning of the pre-training calibration model, and constructing an arrival time automatic calibration model suitable for mine microseismic signals to adapt to microseismic wave characters of underground engineering, comprising:
S31: manually labeling a part of samples in the microseismic data of the mine sites, and constructing a microseismic signal sample data set containing the P-wave arrival times and the S-wave arrival times;
S32: constructing a Gaussian distribution mask around manually labeled samples by using a formula (3) to reduce impacts of mislabeling in the microseismic signal sample data set;

P
⁡
(
x
)

=

{

1

σ
⁢

Assignees

• Univ China Mining

Inventors

• Cao Anye
• Wang Changbin
• Yang Xu
• Liu Yaoqi
• Li Sen
• Niu Qiang
• Dou Linming

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/0464 Primary

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• Y02A90/30

  Assessment of water resources · CPC title