Fault gas on-line monitoring system and method for seismological observation

Abstract

The invention discloses a fault gas online monitoring system and method for seismic observation, and relates to the field of seismic monitoring, and the method comprises the steps: deploying a multi-parameter gas monitoring device near a fault zone, collecting gas concentration data at preset time intervals, building a historical gas concentration time sequence, and building a training feature set through preprocessing and feature extraction; an earthquake prediction model is built based on LSTM, model training is carried out with the training feature set as input and the earthquake occurrence probability as output until a loss function converges, and an optimized earthquake prediction model is obtained; acquiring gas concentration data in a certain time period before the current time node, and predicting an earthquake occurrence probability after the current time node through the optimized earthquake prediction model; when the earthquake occurrence probability exceeds a preset threshold value, early warning information is automatically generated and published to the public through multiple channels, and earthquake risks are prompted. According to the invention, early recognition and early warning of earthquake activities can be realized, and powerful support is provided for earthquake monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake monitoring, and more specifically, to an on-line monitoring system and method for fault gases for earthquake observation. Background Art

[0002] An earthquake is a natural phenomenon in which the internal energy of the earth is suddenly released, resulting in violent vibrations of the earth's crust. It is characterized by strong suddenness, great destructive power, and wide spread. The hazards of earthquakes are not limited to instant physical damage. The secondary disasters, social and economic impacts, and long-term environmental effects caused by them are often more profound. As an important means of preventing earthquake disasters, earthquake monitoring observes, records, and analyzes the vibrations on the earth's surface through various instruments and technologies, can predict, give early warnings, and reduce potential damage and casualties, and minimize the impact of geological disasters on human society to achieve sustainable development.

[0003] A fault is a phenomenon in which the underground rock strata are displaced relative to each other on both sides of a fracture surface or fracture zone. Earthquakes are often caused by fault activities and are a manifestation of fault activities. In recent years, with the continuous development of technology, the technologies and means of earthquake monitoring have become increasingly perfect. Researchers have found that there is a close relationship between the occurrence of earthquakes and fault gases. Fault gases are not only by-products during the occurrence of earthquakes, but their dynamic changes may also directly affect the mechanical behavior of faults and even participate in the earthquake triggering mechanism.

[0004] The core of the fault gas monitoring technology lies in capturing the gas components released from the fault zone and their dynamic changes to evaluate the stress state and activity of the fault. It is gradually developing from single gas detection to multi-parameter, intelligent, and networked, but still faces challenges such as low automation level, low detection accuracy, and susceptibility to environmental factors. Therefore, how to accurately monitor the abnormal fault gases, ensure the accuracy and reliability of earthquake early warnings, and improve the detection efficiency are technical problems that need to be urgently solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an on-line monitoring system and method for fault gases for earthquake observation, which solves the problems existing in the background art.

[0006] To achieve the above object, the present invention provides the following technical solutions: An on-line monitoring method for fault gases for earthquake observation, comprising the following steps: Deploy multi-parameter gas monitoring devices near the fault zone, collect gas concentration data at preset time intervals, and establish a historical gas concentration time series; Preprocess and extract features from the historical gas concentration time series to construct a training feature set; Build an earthquake prediction model based on LSTM, and use the training feature set as the input and the earthquake occurrence probability as the output to train the model until the loss function converges, obtaining an optimized earthquake prediction model; Obtain the gas concentration data for a certain period before the current time node, and predict the earthquake occurrence probability after the current time node through the optimized earthquake prediction model; When the earthquake occurrence probability exceeds the preset threshold, automatically generate a warning message and publish it to the public through multiple channels to prompt the earthquake risk.

[0007] Optionally, when deploying multi-parameter gas monitoring devices near the fault zone, adopt a hierarchical deployment and dynamic optimization strategy. Specifically: On the surface layer, install one device every preset distance along the main fault trace to monitor hydrogen and radon; In the middle and deep layers, deploy a cross-fault borehole array to monitor carbon dioxide; During the earthquake active period, use a drone swarm to form a monitoring grid at the fault intersection to collect multi-gas concentration data.

[0008] Optionally, preprocess the historical gas concentration time series. Specifically: Use linear interpolation to fill in the missing data, and use a sliding window statistic to identify abnormal missing points to complete data cleaning; Perform a sliding average filter and wavelet denoising on the cleaned data to complete noise filtering; Perform normalization processing on the filtered data to obtain preprocessed data.

[0009] Optionally, the training feature set includes: statistical features, frequency domain features, and time window features; Calculate the mean, variance, maximum value, minimum value, and hourly / daily concentration change rate through a sliding window to obtain statistical features; Use the fast Fourier transform to analyze the energy distribution in different frequency bands, identify abnormal frequency bands, and obtain frequency domain features; Calculate the mean and variance of the past 24 hours, the gas concentration change trend in the past 7 days, and the concentration mutation amplitude within N hours before the earthquake through a sliding window to obtain time window features.

[0010] Optionally, the earthquake prediction model introduces an Attention mechanism on the basis of the LSTM network, including: an input layer, an LSTM layer, an Attention layer, and an output layer; The input layer is used to receive the training feature set; The LSTM layer contains several LSTM units, and each LSTM unit has an input gate, a forget gate, an output gate, and a cell state; The Attention layer is used to calculate the weights for each time series, and the vectors output at all time series are weighted and summed as the output vector; The output layer adopts a fully connected network, which is used to transform the output vector of the Attention layer in the feature space to obtain the final output result of the model.

[0011] Optionally, to obtain an optimized earthquake prediction model, the following steps are specifically included: Set the time step, and convert the training feature set into a data set that meets the supervised learning task of the input of the earthquake prediction model according to the time step; Randomly initialize the parameters in the earthquake prediction model. Based on the converted data set, use the mini-batch gradient descent algorithm to update the parameters, and use the Adam adaptive optimization algorithm to optimize the model parameters, continuously minimizing the value of the model loss function; Adopt the early stopping method to dynamically adjust the number of training rounds. When the model shows a trend of overfitting, stop training and save the model parameters as the optimized earthquake prediction model.

[0012] Optionally, it further includes: Obtain seismic activity data through seismographs and accelerometers; Use the gas concentration change curve and the seismic activity distribution map to conduct correlation analysis on the gas concentration data and the seismic activity data, identify abnormal patterns, and predict the seismic risk.

[0013] An on-line monitoring system for fault gases used in seismic observation, applying the on-line monitoring method for fault gases used in seismic observation described in any one of the above, includes: The data acquisition module is used to deploy multi-parameter gas monitoring devices near the fault zone, collect gas concentration data at preset time intervals, and establish a historical gas concentration time series; The data processing module is used to preprocess and extract features from the historical gas concentration time series to construct a training feature set; The model construction and training module is used to build an earthquake prediction model based on the LSTM network, and use the training feature set as the input and the earthquake occurrence probability as the output to train the model until the loss function converges, obtaining an optimized earthquake prediction model; The prediction module is used to obtain the gas concentration data for a certain period before the current time node, and predict the earthquake occurrence probability after the current time node through the optimized earthquake prediction model; The early warning module is used to automatically generate early warning information and release it to the public through multiple channels to prompt the seismic risk when the earthquake occurrence probability exceeds the preset threshold.

[0014] Optionally, the multi-parameter gas monitoring device includes: a sampling probe, a filtering component, an air pump, a solenoid valve, a gas chamber, and an exhaust port; The sampling probe is connected to the inlet of the filtering component through a corrosion-resistant flexible hose. The outlet of the filtering component is connected to the inlet of the air pump through a rigid pipeline. The outlet of the air pump is connected to the inlet of the solenoid valve through a hose. The outlet of the solenoid valve is connected to the air chamber inlet through a short straight pipeline. The air chamber outlet is connected to the exhaust port through a flow-limiting valve.

[0015] Optionally, a hydrogen sensor, a carbon dioxide sensor, and a radon sensor are arranged in the air chamber.

[0016] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses an on-line monitoring system and method for fault gases for seismic observation, which has the following beneficial effects: The present invention uses multi-parameter gas monitoring devices such as radon sensors, hydrogen sensors, and carbon dioxide sensors to collect the gas concentrations in the fault zone at fixed time intervals. Based on the LSTM network combined with the Attention mechanism, a seismic prediction model is built to predict the gas concentration changes and detect anomalies in the collected time series, realizing the early identification of seismic activities. When anomalies occur, warning information is automatically generated and released to the public, enabling early warning and providing strong support for seismic monitoring. In addition, it is also possible to further analyze the seismic activity data obtained by combining seismographs, accelerometers, etc., to improve the accuracy and reliability of the warning. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the on-line monitoring method for fault gases for seismic observation provided by the present invention; Figure 2 It is a structural diagram of the on-line monitoring system for fault gases for seismic observation provided by the present invention; Figure 3 It is a structural diagram of the multi-parameter gas monitoring device provided by the present invention. Detailed Embodiments

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] The strong vibrations during an earthquake can directly lead to building collapses, ground fissures, mountain collapses, etc., causing a large number of casualties. Earthquakes in mountainous areas are prone to trigger large-scale landslides, blocking traffic and burying villages and towns. The hazards of earthquakes are not limited to the immediate physical damage. The secondary disasters, socio-economic impacts, and long-term environmental effects triggered by them are often more profound. Therefore, in the face of this global threat of earthquakes, how to predict the occurrence of earthquakes in advance is a major topic for scientific researchers. At present, earthquake monitoring technology has shifted from traditional analog signal processing to digital and intelligent. The relationship between the occurrence of earthquakes and fault gases has become an important direction in earthquake scientific research. However, this technology still faces challenges such as low automation, low detection accuracy, and susceptibility to environmental factors.

[0021] For this reason, the embodiments of the present invention disclose an on-line monitoring method for fault gases for earthquake observation, as Figure 1 shown, including the following steps: Deploy multi-parameter gas monitoring devices near the fault zone, collect gas concentration data at preset time intervals, and establish a historical gas concentration time series; among them, sampling strictly according to the preset time interval (such as 10 minutes) can ensure uniform distribution of timestamps; Preprocess and extract features from the historical gas concentration time series to construct a training feature set; Build an earthquake prediction model based on LSTM, use the training feature set as the input and the earthquake occurrence probability as the output to train the model until the loss function converges to obtain an optimized earthquake prediction model; Obtain the gas concentration data for a certain period before the current time node, and predict the earthquake occurrence probability after the current time node through the optimized earthquake prediction model; When the earthquake occurrence probability exceeds the preset threshold, automatically generate a warning message and publish it to the public through multiple channels to prompt the earthquake risk. For example, a fault activity probability report can be pushed to government agencies for a first-level warning 72 hours in advance; avoidance tips can be published to the public through APPs / broadcasts for a second-level warning 24 hours in advance.

[0022] Based on Figure 1 the shown process, this embodiment conducts multi-parameter collaborative monitoring based on hydrogen sensors, carbon dioxide sensors, radon sensors, etc., builds an earthquake prediction model based on the LSTM network, uses the collected gas concentration time series for model training, can predict the change of gas concentration, realizes the early identification of earthquake activities, and publishes warning messages to the public through multiple channels, solving the challenges in the distribution channels of earthquake warning messages and the public acceptance in remote areas.

[0023] Fault zones are channels for gas migration in the earth's crust, and the main sources are: (1) release from deep within the earth's crust, including magmatic degassing (release of carbon dioxide, sulfur dioxide, etc. from fault zones in volcanic areas) and mantle fluids (anomalous helium isotopes rising from the deep mantle); (2) rock chemical reactions, including water-rock reactions (generation of hydrogen and methane by the interaction of groundwater and rocks) and radioactive decay (release of radon gas by the decay of uranium and thorium); (3) biological activities, where microorganisms decompose organic matter to produce methane and carbon dioxide.

[0024] In traditional monitoring technologies, the concentrations of gases such as radon, carbon dioxide, and hydrogen released from fault zones are monitored by fixed observation stations for a long time. A multi-parameter, intelligent, and networked integrated system is the current development goal, facing huge challenges in terms of sensitivity, interference, etc. Therefore, when designing a multi-parameter gas monitoring device, it is necessary to comprehensively consider gas types, environmental adaptability, data accuracy, and long-term stability. For example, hydrogen can be detected at low cost through an electrochemical sensor, carbon dioxide can be detected with high precision using a non-dispersive infrared sensor, and radon gas can be detected using an alpha spectrometer. By combining electrochemical, spectroscopy, and mass spectrometry technologies, the detection of key gases can be covered.

[0025] When deploying gas monitoring devices in fault zones, it is necessary to scientifically select key positions in combination with the fault structure, gas release pattern, and monitoring objectives. In this embodiment, when deploying multi-parameter gas monitoring devices near fault zones, a hierarchical deployment and dynamic optimization strategy is adopted. Specifically: In the surface layer (0 - 10 meters), one device is deployed at preset intervals (500 meters) along the main fault trace to monitor hydrogen and radon gas; among them, shallow monitoring can select the surface or shallow boreholes, which is suitable for capturing short-term gas anomalies. A portable borehole probe is used to transmit data in real time; In the mid-depth layer (10 - 200 meters), a cross-fault borehole array is deployed to monitor carbon dioxide; among them, mid-depth monitoring can select cross-fault boreholes, which is suitable for obtaining gas information in the locked sections of deep faults and reducing surface interference; specifically, the cross-fault profile array is deployed by arranging multiple monitoring lines perpendicular to the fault strike to cover the hanging wall and footwall of the fault; During the earthquake active period, an unmanned aerial vehicle (UAV) cluster is used to form a 100 m * 100 m monitoring grid at the fault intersection to collect data on the concentrations of various gases.

[0026] In addition, during the actual data collection process, a dynamic sampling mode can be adopted. For example: under normal circumstances, samples are collected once every ten minutes. When a sudden change in gas concentration is detected, the sampling frequency is switched to once per second for ten minutes.

[0027] Furthermore, preprocessing is performed on the historical gas concentration time series, specifically: The linear interpolation method is used to fill in missing data, and a sliding window statistic is used to identify abnormal missing points to complete data cleaning; Perform moving average filtering (to suppress high-frequency noise) and wavelet denoising on the cleaned data to complete noise filtering; Normalize the filtered data to obtain preprocessed data.

[0028] Furthermore, the training feature set includes: statistical features, frequency-domain features, and time-window features; Calculate the mean, variance, maximum value, minimum value, and concentration change rate per hour / day through a sliding window to obtain statistical features for the statistical description of the global data of the entire time series, reflecting the overall characteristics of the data. All data from the start to the end of the sequence are involved in the calculation; Use the fast Fourier transform to analyze the energy distribution in different frequency bands, identify abnormal frequency bands, and obtain frequency-domain features; Statistically calculate the mean and variance of the past 24 hours, the gas concentration change trend in the past 7 days, and the concentration mutation amplitude within N hours before an earthquake through a sliding window to obtain time-window features for the statistical analysis of local segments of the time series, capturing the dynamic changes or short-term patterns of the data. Only the data within the window is involved in the calculation, and the window slides over the entire sequence at a fixed step size.

[0029] Specifically, for the extraction of frequency-domain features of the Fourier transform, the main steps are: calculate the FFT and the frequency axis, plot the frequency-amplitude diagram, identify the main frequency components, find the frequencies with amplitudes significantly higher than the background noise, and identify the main frequency peaks. Among them, the main frequency peaks include: low-frequency peaks corresponding to daily or seasonal cycles; medium-frequency peaks that may be related to fault creep or microseismic activities; high-frequency peaks corresponding to sudden gas releases or instrument noise.

[0030] Furthermore, the earthquake prediction model introduces the Attention mechanism based on the LSTM network, including: an input layer, an LSTM layer, an Attention layer, and an output layer; The input layer is used to receive the training feature set; The LSTM layer contains several LSTM units, and each LSTM unit has an input gate, a forget gate, an output gate, and a cell state. In this embodiment, introducing the Attention mechanism can enhance the model's ability to focus on key time steps or features and improve the prediction accuracy; The Attention layer is used to calculate the weights for each time sequence and perform weighted summation on the vectors output by all time sequences as the output vector; The output layer uses a fully connected network to perform a feature space transformation on the output vector of the Attention layer to obtain the final output result of the model.

[0031] Specifically, the specific design in the LSTM unit is as follows: Input gate: ; where: is the output of the input gate, , , are respectively the weight values of the input gate, is the bias weight vector of the input gate, is the vector of candidate values for the new cell state, is the new cell state, is the cell state at the previous moment; Forget gate: ; where: is the output of the hidden gate, is the activation function Sigmoid, is the input value at the current moment associated with the forget gate weight value, is the weight of the forget gate associated with the output value at the previous moment , is the bias weight vector of the forget gate; Output gate: ; where: is the output of the output gate at the current moment, is the output gate weight value associated with the input value at the current moment , is the weight of the output gate associated with the output value at the previous moment , is the bias weight vector of the output gate, is the final output result at the current moment.

[0032] Furthermore, an optimized earthquake prediction model is obtained, which specifically includes the following steps: Set the time step, and convert the training feature set into a data set that meets the input requirements of the earthquake prediction model for supervised learning tasks according to the time step; Randomly initialize the parameters in the earthquake prediction model. Based on the converted data set, use the mini-batch gradient descent algorithm to update the parameters, and use the Adam adaptive optimization algorithm to optimize the model parameters, continuously minimizing the value of the model loss function; Adopt the early stopping method to dynamically adjust the number of training rounds. When the model shows a trend of overfitting, stop training and save the model parameters as the optimized earthquake prediction model to reduce the risk of overfitting.

[0033] In a further implementation manner, the monitoring method of the present invention further includes: Obtain earthquake activity data through seismographs and accelerometers; use the gas concentration change curve and the earthquake activity distribution map to conduct correlation analysis on the gas concentration data and the earthquake activity data, identify abnormal patterns, and predict earthquake risks.

[0034] In this embodiment, the seismometer can detect the minute vibrations of the earth's crust and capture seismic waves from low frequencies to high frequencies; the accelerometer can detect the high-frequency vibrations during strong earthquakes and is suitable for near-earthquake monitoring. During actual use, high-frequency seismic activity data is collected to ensure capturing the details of seismic waves. Digital signal processing technology is used to process the seismic signals, remove noise, and extract effective information. Based on the time difference of arrival of seismic waves, the earthquake source location can be determined, and the magnitude can be calculated according to the amplitude of seismic waves. By combining gas concentration data and seismic activity data, fault activity can be evaluated, and potential fault slips or earthquake risks can be predicted.

[0035] Corresponding to Figure 1 the method described above, an on-line fault gas monitoring system for seismic observation provided by an embodiment of the present invention is used to Figure 1 For the specific implementation of the method in Figure 2 As shown in a data acquisition module, which is used to deploy a multi-parameter gas monitoring device near the fault zone, collect gas concentration data at preset time intervals, and establish a historical gas concentration time series; a data processing module, which is used to preprocess and extract features from the historical gas concentration time series to construct a training feature set; a model construction and training module, which is used to build an earthquake prediction model based on the LSTM network, take the training feature set as the input and the earthquake occurrence probability as the output for model training until the loss function converges, and obtain an optimized earthquake prediction model; a prediction module, which is used to obtain the gas concentration data for a certain period before the current time node, and predict the earthquake occurrence probability after the current time node through the optimized earthquake prediction model; an early warning module, which is used to automatically generate early warning information and release it to the public through multiple channels to prompt earthquake risks when the earthquake occurrence probability exceeds a preset threshold.

[0036] Furthermore, as Figure 3 shown in the multi-parameter gas monitoring device includes: a collection probe, a filtration component, an air pump, a solenoid valve, an air chamber, and an exhaust port;

[0037] Specifically, a hydrogen sensor, a carbon dioxide sensor, and a radon sensor are arranged in the air chamber.

[0038] In this embodiment, the filtering component adopts multi-stage filtering, which is connected in series in the order of "coarse filtration - fine filtration - molecular sieve". It can be connected to the front and rear pipelines through threaded interfaces or snap structures. The filter element is detachable and replaceable, ensuring the preliminary purification of the inhaled gas and avoiding particle damage to the pump body. The solenoid valve is used to switch the gas path direction or for multi-channel selection. It is centrally installed through an integrated manifold, reducing the pipeline length. An O-ring seal is used between the valve body and the pipeline to ensure airtightness. The short straight pipeline between the gas chamber inlet and the solenoid valve outlet can reduce the gas residence time, and the flow limiting valve at the gas chamber outlet end can maintain the stability of the cavity pressure. In addition, a flow meter can be set between the air pump outlet and the solenoid valve to monitor the actual flow of the pump; a pressure sensor can be set at the gas chamber inlet to detect whether the pressure in the cavity is stable; a temperature and humidity sensor can be set in the gas chamber to measure the gas temperature and humidity for data compensation; a control box is set, and the air pump, solenoid valve, and sensor power supply are connected through a cable harness to control the operation of the air pump and solenoid valve.

[0039] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0040] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault gas online monitoring method for earthquake observation, characterized in that: The following steps are involved: Deploy multi-parameter gas monitoring devices near the fault zone to collect gas concentration data at preset time intervals and establish a historical gas concentration time series; Preprocess and extract features of historical gas concentration time series to construct training feature sets; An earthquake prediction model is built based on LSTM, with the training feature set as input and the probability of earthquake occurrence as output. The model is trained until the loss function converges, and the optimized earthquake prediction model is obtained. Obtain gas concentration data for a certain period of time before the current time node, and use the optimized earthquake prediction model to predict the probability of earthquake occurrence after the current time node; When the probability of an earthquake exceeds a preset threshold, early warning information is automatically generated and released to the public through multiple channels to alert the public to earthquake risks.

2. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: When deploying multi-parameter gas monitoring devices near fault zones, a layered deployment and dynamic optimization strategy is adopted. Specifically: In the surface layer, one detector is deployed at preset intervals along the main fault trace to monitor hydrogen and radon gases; In the mid-depth layer, a cross-fault borehole array is deployed to monitor carbon dioxide; During periods of active earthquakes, drone swarms are used to form a monitoring grid at the intersection of faults to collect data on multiple gas concentrations.

3. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: The historical gas concentration time series is preprocessed as follows: Linear interpolation is used to fill missing data, and sliding window statistics are used to identify abnormal missing points to complete data cleaning; Perform sliding average filtering and wavelet denoising on the cleaned data to complete noise filtering; The filtered data is normalized to obtain preprocessed data.

4. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: The training feature set includes: statistical features, frequency domain features and time window features; The mean, variance, maximum, minimum, hourly / daily concentration change rate are calculated through the sliding window to obtain statistical characteristics; Use fast Fourier transform to analyze the energy distribution of different frequency bands, identify abnormal frequency bands, and obtain frequency domain features; The time window characteristics are obtained by using a sliding window to count the mean and variance of the past 24 hours, the trend of gas concentration changes in the past 7 days, and the concentration mutation amplitude within N hours before the earthquake.

5. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: The earthquake prediction model introduces the Attention mechanism based on the LSTM network, including: input layer, LSTM layer, Attention layer and output layer; Input layer, used to receive the training feature set; The LSTM layer contains several LSTM units, each of which has an input gate, a forget gate, an output gate, and a cell state; The Attention layer is used to calculate the weight of each time series and take the weighted sum of all time series output vectors as the output vector; The output layer uses a fully connected network to transform the output vector of the Attention layer into the feature space to obtain the final output result of the model.

6. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: The optimized earthquake prediction model is obtained, which specifically includes the following steps: Set the time step and convert the training feature set into a data set that meets the supervised learning task of earthquake prediction model input according to the time step; The parameters in the earthquake prediction model are randomly initialized. Based on the converted data set, the small batch gradient descent algorithm is used to update the parameters. The Adam adaptive optimization algorithm is used to optimize the model parameters and continuously minimize the model loss function value. The early stopping method is used to dynamically adjust the training rounds. When the model shows an overfitting trend, the training is stopped and the model parameters are saved as the optimized earthquake prediction model.

7. The method for online monitoring of fault gas for earthquake observation according to claim 1, characterized in that: Also includes: Obtain earthquake activity data through seismometers and accelerometers; Using the gas concentration change curve and seismic activity distribution map, we conduct correlation analysis on gas concentration data and seismic activity data, identify abnormal patterns, and predict earthquake risks.

8. An online fault gas monitoring system for earthquake observation, using the online fault gas monitoring method for earthquake observation as claimed in any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to deploy a multi-parameter gas monitoring device near the fault zone, collect gas concentration data at preset time intervals, and establish a historical gas concentration time series; Data processing module, used to preprocess and extract features of historical gas concentration time series and construct training feature sets; The model building and training module is used to build an earthquake prediction model based on the LSTM network. The model is trained with the training feature set as input and the probability of earthquake occurrence as output until the loss function converges to obtain the optimized earthquake prediction model. The prediction module is used to obtain the gas concentration data for a certain period of time before the current time node, and predict the probability of earthquake occurrence after the current time node through the optimized earthquake prediction model; The early warning module is used to automatically generate early warning information and release it to the public through various channels to alert the public to earthquake risks when the probability of an earthquake exceeds a preset threshold.

9. The fault gas online monitoring system for earthquake observation according to claim 8, characterized in that: The multi-parameter gas monitoring device includes: a collection probe, a filter assembly, an air pump, a solenoid valve, an air chamber, and an exhaust port; The collection probe is connected to the inlet of the filter assembly through a corrosion-resistant flexible hose, the outlet of the filter assembly is connected to the inlet of the air pump through a hard pipeline, the outlet of the air pump is connected to the inlet of the solenoid valve through a hose, the outlet of the solenoid valve is connected to the inlet of the air chamber through a short straight pipeline, and the outlet of the air chamber is connected to the exhaust port through a flow limiting valve.

10. The fault gas online monitoring system for earthquake observation according to claim 9, characterized in that: The gas chamber is equipped with a hydrogen sensor, a carbon dioxide sensor and a radon sensor.

Images