Artificial intelligence calculation method and apparatus for monitoring earthquake in real time based on edge cloud cooperation, and storage medium

Patent metadata

Field	Value
Publication number	US-11513245-B2
Application number	US-202117507742-A
Country	US
Kind code	B2
Filing date	Oct 21, 2021
Priority date	Dec 9, 2020
Publication date	Nov 29, 2022
Grant date	Nov 29, 2022

Abstract

An artificial intelligence calculation method and apparatus for monitoring an earthquake in real time based on edge cloud cooperation is applied to a micro-earthquake data processing system. The micro-earthquake data processing system includes an edge calculation device and a remote server in communication connection with the edge calculation device. The remote server deploys a micro-earthquake data analyzing model based on an artificial intelligence to the edge calculation device in advance. The method includes steps of receiving, by the remote server, effective event data related to the micro-earthquake from the edge calculation device; performing a transfer training to the micro-earthquake data analyzing model by the remote server according to the effective event data; and updating the model after the micro-earthquake data analyzing model that has been transfer-trained is transmitted to the edge calculation device by the remote server.

First claim

What is claimed is:

1. An artificial intelligence calculation method for monitoring an earthquake in real time based on edge cloud cooperation, which is applied to a micro-earthquake data processing system, the micro-earthquake data processing system comprises an edge calculation device and a remote server that is in communication connection with the edge calculation device, wherein the remote server deploys a micro-earthquake data analyzing model based on an artificial intelligence to the edge calculation device in advance, and the method comprises: comprising:
receiving, by a remote server, effective event data related to a micro-earthquake from an edge calculation device, wherein the effective event data is obtained, by the edge calculation device, by means of calculating and analyzing the micro-earthquake data collected by the edge calculation device by using a micro-earthquake data analyzing model;
performing a transfer training to the micro-earthquake data analyzing model by the remote server according to the effective event data, wherein data is manually checked after the effective event data is received, data for which the current calculation is insufficient is checked, the data is labeled, and when there are enough labeled data, the model is transfer-trained using the method of transfer learning; and
updating the model after the micro-earthquake data analyzing model that has been transfer-trained is transmitted to the edge calculation device by the remote server, wherein the remote server, before the operation of receiving the effective event data related to the micro-earthquake from the edge calculation device, further includes: the edge calculation device collects the micro-earthquake data; the edge calculation device analyzes and calculates the micro-earthquake data using the micro-earthquake data analyzing model and generates the effective event data corresponding to the micro-earthquake data; and the edge calculation device transmits the effective event data to the remote server; and
wherein the method is applied to a micro-earthquake data processing system; the micro-earthquake data processing system comprises an edge calculation device and a remote server in communication connection with the edge calculation device; the remote server deploys the micro-earthquake data analyzing model based on an artificial intelligence to the edge calculation device in advance;
wherein the micro-earthquake data analyzing model is a micro-earthquake depth picking network MSnet trained using standard data sets, the micro-earthquake depth picking network MSnet is based on a convolutional neural network (CNN) and a recurrent neural network (RNN); the micro-earthquake data analyzing model includes a feature extraction layer, a feature output layer and a full connection layer, wherein the feature extraction layer consists of four layers of convolution layers, the convolution kernel of the first layer of convolution layer is a 9 by 9 convolution kernel, the convolution kernel of the fourth layer of convolution layer is a 3 by 3 convolution kernel, there is one pooling layer for which the pooling kernel is 3 between two layers of convolution layers, the feature output layer includes three layers of a long short-term memory (LSTM) structure and one Projection Layer, each LSTM layer contains 632 units, and the Projection Layer contains 600 units, and the full connection layer has two layers; and
the remote server performs an operation of transfer-training the micro-earthquake data analyzing model according to the effective event data, comprising: the remote server trains a mapping relationship between the feature output layer and the full connection layer using the effective event data and standard micro-earthquake data acquired from a preset standard micro-earthquake database, extracts features output from the last layer of the LSTM structure out, fine-tunes the model using the effective event data and standard micro-earthquake data acquired from the preset standard micro-earthquake database, and reestablishes a mapping relationship from the feature output layer to the full connection layer; and
wherein the edge calculation device is a plurality of edge calculation devices, and the plurality of edge calculation devices is divided into a plurality of groups, wherein the plurality of groups is deployed at different locations, at least one edge calculation device is deployed at each location, the micro-earthquake data analyzing models of the edge calculation devices in each group are the same, and the plurality of edge calculation devices is used to collect the effective event data at different locations, respectively; and the remote server performs a transfer training to the micro-earthquake data analyzing model deployed by the grouped edge calculation devices according to the effective event data transmitted by the edge calculation devices in each group, then transmits the micro-earthquake data analyzing model that has been transfer-trained is transmitted to the group, and updates the micro-earthquake data analyzing model of the edge calculation device in the group, and
wherein the edge calculation device, before collecting the micro-earthquake data, further includes: judges whether the grouped edge calculation devices satisfy preset collection conditions; and in a case of satisfying the preset collection conditions, the edge calculation devices in each group collect the micro-earthquake data; and
wherein judging whether the grouped edge calculation devices satisfy preset collection conditions includes: the grouped edge calculation devices intra-group broadcast state information, wherein the state information is used to record monitoring states of the edge calculation devices; calculating a number of edge calculation devices in which the state information in the group satisfies a preset trigger condition; and judging whether the grouped edge calculation devices satisfy the preset collection conditions according to the number of devices and a preset first threshold; and
wherein the state information includes state identification, and before calculating the number of the edge calculation devices in which the state information in the group satisfies the preset trigger condition, further includes: judging whether the state information in the group satisfies a preset trigger condition according to the state identification, wherein pre-triggering and group-broadcasting signal amplitudes in groups of the multi-node seismograph are carried out for the collected earthquake data, group-broadcasting is performed using a Lora ultra-long distance and ultra-low power communication technology, the broadcasting content is a pre-trigger state of each node in the group, which is a string composed of an ID of each node in the group and a trigger state thereof, and the pre-trigger state is a type of bool; after deployment, the node sets a signal minimum trigger threshold by a circuit, after the collected signal voltage exceeds a minimum trigger threshold, the pre-trigger state of the node changes to be True, judging that state information in the group satisfies the preset trigger condition, and meanwhile updating fields corresponding to group-broadcasting immediately.

2. The method according to claim 1 , wherein the method further comprises visually displaying the effective event data by the remote server.

3. A non-transitory storage medium, comprising: a storage program; wherein the method according to claim 1 is executed by a processor when the storage program is running.

Assignees

• Inst Geology & Geophysics Cas

Inventors

• Xue Qingfeng
• Wang Yibo
• Yao Yi

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/08

  Learning methods · CPC title

• G06F16/252

  between a Database Management System and a front-end application · CPC title

• G01V1/282 Primary

  Application of seismic models, synthetic seismograms · CPC title