Non-natural seismic event identification and determination method based on deep learning

Abstract

The invention relates to the technical field of seismology and artificial intelligence crossing, and discloses a non-natural seismic event identification and judgment method based on deep learning, and the method comprises the steps: building a seismic data collection module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module, the seismic data analysis module is based on four internal units, and the problems of misjudgment, poor adaptability and low efficiency of a traditional method are solved in a targeted mode. The early warning module performs graded response according to an analysis result, realizes second-level early warning and is linked with an emergency system; the judgment module is used for realizing accurate classification of events by utilizing ensemble learning; the comprehensive optimization module dynamically updates a model and a data strategy based on an early warning and judgment result; the verification module guarantees the reliability of the system through cross verification and blind measurement, and finally achieves the effects of improving the recognition accuracy of the non-natural seismic event, shortening the cross-regional deployment period, reducing the false alarm rate and improving the data utilization rate.

Description

Technical Field

[0001] The present invention relates to the field of intersectional technology of seismology and artificial intelligence, and specifically to a method for identifying and determining non-natural earthquake events based on deep learning. Background Art

[0002] In the field of seismology, accurately identifying and determining non-natural earthquake events is crucial for public safety, resource extraction, environmental monitoring, and national defense security. Non-natural earthquake events, such as explosions, nuclear tests, mining tremors, and engineering collapses, have characteristics that differ from those of natural earthquakes. However, these differences are often difficult for the human eye to discern, and are even more challenging under complex geological conditions. With the development of deep learning technology, deep learning-based methods for identifying and determining non-natural earthquake events have become a research hotspot. These methods can automatically learn features from large amounts of seismic data, improving the accuracy and efficiency of identification and determination.

[0003] However, existing earthquake monitoring systems still face many challenges in identifying non-natural earthquake events. First, because the waveform characteristics of non-natural earthquake events are similar to those of natural earthquakes and are affected by multiple factors such as geological conditions and environmental interference, the accuracy and real-time performance of event identification are difficult to guarantee. Secondly, the geological conditions in different regions vary greatly, and traditional earthquake identification models are difficult to adapt to the needs of cross-regional deployment. Model training and optimization are required based on the geological characteristics of different regions. In addition, the false alarm rate and missed alarm rate of earthquake early warning systems are high, especially in complex environments. How to effectively reduce the false alarm rate and missed alarm rate and improve the accuracy and response speed of early warnings are issues that need to be urgently addressed in the current earthquake early warning field.

[0004] The present invention proposes a method for identifying and judging non-natural earthquake events based on deep learning. By establishing a system including a seismic data acquisition module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module, the entire process from data acquisition to final judgment is realized intelligently. Summary of the Invention

[0005] (0) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a method for identifying and judging non-natural earthquake events based on deep learning, which has the advantages of high accuracy, strong adaptability, low false alarm rate, and real-time and high efficiency. It solves the problems of traditional methods in identifying non-natural earthquake events, such as feature overlap leading to misjudgment and missed judgments, poor adaptability to cross-regional geological conditions, high false alarm and missed alarm rates of early warning systems, and insufficient utilization of multi-source data.

[0006] (2) Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method for identifying and determining non-natural earthquake events based on deep learning, comprising the following steps: Step 1: Establish a system, which includes a seismic data acquisition module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module; Step 2: The earthquake data acquisition module acquires multi-source earthquake related data; Step 3: The seismic data analysis module calculates and analyzes the collected data; Step 4: The early warning module triggers an early warning based on the analysis results; Step 5: The judgment module performs final classification based on the analyzed event type; Step 6: The comprehensive optimization module optimizes the system according to the warning and judgment results of the warning module and the judgment module; Step 7: The verification module verifies the optimized system.

[0007] Preferably, the seismic data acquisition module includes a natural seismic data acquisition unit, a non-natural seismic data acquisition unit, an environmental interference data acquisition unit and a multi-source auxiliary data unit, and the seismic data analysis module includes a feature overlapping interval analysis unit, a geological condition analysis unit, a microseismic prediction unit and a multi-source data accuracy analysis unit.

[0008] Preferably, the natural earthquake data acquisition unit obtains global natural earthquake waveforms, focal depths and magnitude data through a public database of the global seismic network.

[0009] Preferably, the non-natural earthquake data acquisition unit collects waveform and spectrum characteristic data of blasting, nuclear testing, mining earthquakes, and engineering collapse events through multiple channels.

[0010] Preferably, the environmental interference data acquisition unit utilizes a multi-type sensor network to acquire industrial noise, meteorological interference, and geological background noise data.

[0011] Preferably, the multi-source auxiliary data unit obtains geological structure data, satellite remote sensing data, meteorological data and human activity data through cross-domain data fusion.

[0012] Preferably, the feature overlap interval analysis unit calculates the posterior probability based on the collected data , and its calculation formula is: ; In the formula, represents the posterior probability, represents the likelihood function, represents the prior probability, represents marginal probability; The posterior probability Used for Bayesian inference to update the probability of an event occurring "unnaturally".

[0013] Preferably, the geological condition analysis unit calculates the geological similarity index based on the collected data , and its calculation formula is: ; In the formula, represents the geological similarity index, Indicates the current density of underground rock formations, represents the reference underground rock density, Indicates the current propagation speed of the longitudinal wave, represents the propagation velocity of the reference longitudinal wave, Indicates the current propagation speed of the shear wave, represents the propagation velocity of the reference shear wave, represents the current station density, represents the reference station density; The geological similarity index Quantify the similarity of current geological conditions to reference geological conditions.

[0014] Preferably, the microseismic prediction unit calculates the earthquake warning score based on the collected data , and its calculation formula is: ; In the formula, represents the earthquake early warning score, Indicates the current waveform. Indicates the similarity of natural / non-natural earthquake waveform templates, Indicates the current waveform spectrum energy, Indicates the spectrum energy of natural / non-natural earthquakes, represents the geological similarity index, Indicates the factors affecting meteorology and human activities, 、 、 、 Respectively represent weight coefficients; The “ "Combining waveform, spectrum, geological and external environmental factors to calculate earthquake early warning scores .

[0015] Preferably, the multi-source data accuracy analysis unit calculates the cross-modal fusion score based on the collected data , and its calculation formula is: ; In the formula, represents the cross-modal fusion score, Indicates the signal-to-noise ratio, which reflects the quality of the radar signal. Indicates the matching degree between the acoustic signal spectrum and the template spectrum, Indicates the timestamp change value of radar data and sound wave data, represents the time difference threshold, represents the spatial coordinates of the radar data, represents the spatial coordinates of the sound wave data, Indicates the maximum allowed spatial difference; The cross-modal fusion score Assess the reliability and consistency of multi-source data.

[0016] Compared with the existing technology, the present invention provides a method for identifying and judging non-natural earthquake events based on deep learning, which has the following beneficial effects: 1. The present invention calculates the posterior probability ≥0.7, which is used as a quantitative indicator of the possibility of an event being "unnatural". When the posterior probability is , start the multi-model cross-validation mechanism, combine the waveform characteristics and geological conditions to make a secondary judgment; when the posterior probability When ≤0.3, natural earthquakes or environmental interference are given priority, ultimately achieving the effect of reducing human intervention and improving the accuracy of identifying non-natural events.

[0017] 2. The present invention calculates the geological similarity index , as the basis for adaptive adjustment of model parameters, when the geological similarity index When the geological similarity index is 0.6≤ When the regional specific model is automatically enabled, the feature extraction weight is optimized; when the geological similarity index When the model reconstruction process is triggered, the geological characteristics of the new area are combined for transfer learning. This mechanism eliminates the need for retraining when the system is deployed across regions, shortens the adaptation cycle, and maintains stable recognition accuracy.

[0018] 3. The present invention calculates the earthquake early warning score , as the criterion for judging the emergency degree of the event, the earthquake warning score ≥0.8, the system will immediately issue a red alert and automatically link the emergency response system; when 0.6≤earthquake warning score When the earthquake warning score is less than 0.8, a yellow warning will be issued, prompting manual review and enhanced monitoring; When the value is <0.6, the regular monitoring frequency is maintained, but the event characteristics are recorded for model optimization. This hierarchical early warning mechanism will reduce the false alarm rate while ensuring that no true non-natural events are missed, thereby shortening the early warning response time.

[0019] 4. The present invention calculates the cross-modal fusion score , as the core basis for data quality assessment and fusion strategy adjustment, when the cross-modal fusion score When the cross-modal fusion score is ≥0.9, a weighted average fusion strategy is adopted to fully utilize the advantages of each data source; when 0.7≤cross-modal fusion score When the cross-modal fusion score is less than 0.9, the data cleaning process is started to remove outliers and perform interpolation processing; when ... When the error is <0.7, the sensor calibration procedure is triggered and the data source weight is dynamically adjusted. This mechanism improves the system's data utilization and positioning accuracy in complex environments, effectively reducing misjudgments caused by inconsistent data. Through the above optimization, the system realizes full-process intelligence from data collection to final judgment, greatly improving the recognition accuracy, warning speed and system adaptability of non-natural earthquake events, especially showing excellent robustness in areas with complex geological conditions and multi-source data interference environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a system module diagram of the present invention; Figure 2 It is a step diagram of the method of the present invention; Figure 3 This is a diagram of the actual scenario implementation of Example 1 of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] See also Figure 1-Figure 3 , a method for identifying and judging non-natural earthquake events based on deep learning, comprising the following steps: Step 1: Establish a system, which includes a seismic data acquisition module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module; Step 2: The seismic data acquisition module collects multi-source seismic data to provide a data basis for subsequent analysis; Step 3: The seismic data analysis module calculates and analyzes the collected data; Step 4: The early warning module triggers an early warning based on the analysis results; Step 5: The judgment module performs final classification based on the analyzed event type; Step 6: The comprehensive optimization module optimizes the system according to the warning and judgment results of the warning module and the judgment module; Step 7: The verification module verifies the optimized system.

[0023] The advantages are: by establishing a system, the system consists of a seismic data acquisition module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module. The seismic data acquisition module comprehensively covers event characteristics and environmental factors through multi-source data acquisition units (natural seismic data acquisition unit, non-natural seismic data acquisition unit, environmental interference data acquisition unit and multi-source auxiliary data unit) to ensure data integrity; the seismic data analysis module is based on four major units: feature overlapping interval analysis unit, geological condition analysis unit, microseismic prediction unit and multi-source data accuracy analysis unit, to solve the problems of misjudgment, poor adaptability and low efficiency of traditional methods; the early warning module is divided according to the analysis results Level response, achieving second-level early warning and linkage with the emergency system; the judgment module uses integrated learning to achieve accurate classification of events, specifically classified as follows: non-natural earthquakes can be further subdivided into human-operated earthquakes (blasting, mining), industrial-induced earthquakes (oil and gas extraction) and engineering activity earthquakes (reservoir storage, subway construction) subcategories, thereby improving the pertinence of early warnings; the comprehensive optimization module dynamically updates the model and data strategy based on the early warning and judgment results; the verification module ensures system reliability through cross-validation and blind testing, ultimately achieving the effects of improved accuracy in identifying non-natural earthquake events, shortened cross-regional deployment cycles, reduced false alarm rates and increased data utilization, greatly enhancing the intelligence level and practicality of earthquake monitoring in complex environments.

[0024] The seismic data acquisition module includes a natural seismic data acquisition unit, a non-natural seismic data acquisition unit, an environmental interference data acquisition unit and a multi-source auxiliary data unit. The seismic data analysis module includes a feature overlapping interval analysis unit, a geological condition analysis unit, a microseismic prediction unit and a multi-source data accuracy analysis unit.

[0025] The natural earthquake data acquisition unit obtains global natural earthquake waveforms (P waves / S waves / surface waves), focal depths, and magnitude data from public databases of global seismic networks (such as IRIS and the China National Earthquake Networks). The specific data and acquisition requirements are as follows: Collection content (1) Waveform data: original waveforms of P-wave, S-wave, and surface wave (sampling rate ≥ 100 Hz, accuracy ± 0.01 μm / s); (2) Source parameters: focal depth (accuracy ±0.1 km), magnitude (ML, Mw, accuracy ±0.1) (3) Spatiotemporal information: earthquake occurrence time (UTC time, accuracy of ±0.01s) and epicenter location (latitude and longitude, accuracy of ±0.001°).

[0026] Collection requirements (1) Signal integrity detection: ensure that the continuous recording time is ≥300s; (2) Noise level assessment: background noise ≤ 10~ 9 m / s² / √Hz.

[0027] The non-natural earthquake data acquisition unit collects waveform and spectrum characteristic data of blasting, nuclear testing, mining earthquakes, and engineering collapse events through multiple channels. The specific data and data sources are: Specific data (1) Waveform characteristics: high-frequency energy distribution (10-100 Hz energy accounts for ≥70%); (2) Spectral characteristics: dominant frequency (blasting > 20 Hz, natural earthquake < 10 Hz); (3) Time-frequency characteristics: duration (blasting <5s, natural earthquake >10s).

[0028] Data Source (1) Mining sector: blasting operation records (time, charge, location); (2) Construction unit: underground construction log; (3) Nuclear monitoring network: radioactive material monitoring data.

[0029] The environmental interference data acquisition unit uses a multi-type sensor network to obtain industrial noise, meteorological interference (lightning / wind noise), and geological background noise data. The specific data and sensor network are as follows: sensor networks (1) Industrial noise sensors: deployed in mining areas and around factories (monitoring range ≥ 10 km) (2) Meteorological sensors: anemometer, barometer, lightning monitor (3) Geological background noise sensor: long-period seismograph (period ≥ 100s): Specific data (1) Industrial noise: mechanical vibration frequency characteristics (50-500 Hz); (2) Meteorological interference: strong wind (1-5 Hz noise when wind speed > level 6), lightning (electromagnetic pulse interference); (3) Geological noise: tidal effect (period 12 to 24 hours), groundwater fluctuation (0.1 to 1 Hz).

[0030] The multi-source auxiliary data unit obtains geological structure data, satellite remote sensing data, meteorological data and human activity data through cross-domain data fusion. The specific data are: (1) Geological structural data: regional geological map (1:50,000 scale) and fault distribution; (2) Satellite remote sensing data: InSAR deformation data (accuracy ±5 mm) and optical images (resolution ≤ 1 m); (3) Meteorological data: precipitation data (accuracy ±0.1 mm), temperature (accuracy ±0.1°C); (4) Human activity data: mining plans and urban construction plans.

[0031] The feature overlap interval analysis unit calculates the posterior probability based on the collected data , and its calculation formula is: ; In the formula, Represents the posterior probability, that is, given the observed data Under the condition of , the probability of an event occurring "unnaturally" is, Represents the likelihood function, which means that under the condition that the event "is not natural", the data is observed The probability of Represents the prior probability, that is, when no data is observed Previously, estimates of the probability of an event occurring "unnaturally" were Represents the marginal probability, indicating that the observed data The total probability of , obtained by summing or integrating the probabilities of all possible events; Posterior probability Used for Bayesian inference to update the probability of an event occurring "unnaturally".

[0032] The advantage is: by calculating the posterior probability ≥0.7, which is used as a quantitative indicator of the possibility of an event being "unnatural". When the posterior probability is , start the multi-model cross-validation mechanism, combine the waveform characteristics and geological conditions to make a secondary judgment; when the posterior probability When ≤0.3, natural earthquakes or environmental interference are given priority, ultimately achieving the effect of reducing human intervention and improving the accuracy of identifying non-natural events.

[0033] The geological condition analysis unit calculates the geological similarity index based on the collected data , and its calculation formula is: ; In the formula, represents the geological similarity index, Indicates the current density of underground rock formations, represents the reference underground rock density, Indicates the current propagation speed of longitudinal waves (P waves), represents the propagation velocity of the reference longitudinal wave (P wave), Indicates the current propagation speed of shear waves (S waves), represents the propagation velocity of the reference shear wave (S wave), represents the current station density, represents the reference station density; Geological Similarity Index Quantify the similarity of current geological conditions to reference geological conditions.

[0034] Advantages: By calculating the geological similarity index , as the basis for adaptive adjustment of model parameters, when the geological similarity index When the geological similarity index is 0.6≤ When the regional specific model is automatically enabled, the feature extraction weight is optimized; when the geological similarity index When the model reconstruction process is triggered, the geological characteristics of the new area are combined for transfer learning. This mechanism eliminates the need for retraining when the system is deployed across regions, shortens the adaptation cycle, and maintains stable recognition accuracy.

[0035] The microseismic prediction unit calculates the earthquake warning score based on the collected data , and its calculation formula is: ; In the formula, represents the earthquake early warning score, Indicates the current waveform. Indicates the similarity of natural / non-natural earthquake waveform templates, Indicates the current waveform spectrum energy, Indicates the spectrum energy of natural / non-natural earthquakes, It represents the geological similarity index, reflecting the difference between the current geological conditions and the reference conditions. Indicates the factors affecting meteorology and human activities, 、 、 、 They represent the corresponding weight coefficients, which are determined through system model training. The value is determined by training historical earthquake data through machine learning algorithms, adjusting the weights based on expert experience, and determining the optimal weight coefficient with the goal of minimizing the prediction error; “ "Combining waveform, spectrum, geological and external environmental factors to calculate earthquake early warning scores .

[0036] Advantages: By calculating the earthquake early warning score , as the criterion for judging the emergency degree of the event, the earthquake warning score ≥0.8, the system will immediately issue a red alert and automatically link the emergency response system; when 0.6≤earthquake warning score When the earthquake warning score is less than 0.8, a yellow warning will be issued, prompting manual review and enhanced monitoring; When the value is <0.6, the regular monitoring frequency is maintained, but the event characteristics are recorded for model optimization. This hierarchical early warning mechanism will reduce the false alarm rate while ensuring that no true non-natural events are missed, thereby shortening the early warning response time.

[0037] The multi-source data accuracy analysis unit calculates the cross-modal fusion score based on the collected data , and its calculation formula is: ; In the formula, represents the cross-modal fusion score, Indicates the signal-to-noise ratio, which reflects the quality of the radar signal. Indicates the matching degree between the acoustic signal spectrum and the template spectrum, Indicates the timestamp change value of radar data and sound wave data, represents the time difference threshold, represents the spatial coordinates of the radar data, represents the spatial coordinates of the sound wave data, represents the maximum allowed spatial difference, 、 、 、 They represent the corresponding weight coefficients, which are determined by training and validating the multimodal data fusion model. The specific steps include data collection, feature normalization, weight initialization, model optimization, and iterative update under the goal of minimizing cross-modal matching error. Cross-modal fusion score Evaluate the reliability and consistency of multi-source data to support the identification of non-natural seismic events.

[0038] The advantage is: by calculating the cross-modal fusion score , as the core basis for data quality assessment and fusion strategy adjustment, when the cross-modal fusion score When the cross-modal fusion score is ≥0.9, a weighted average fusion strategy is adopted to fully utilize the advantages of each data source; when 0.7≤cross-modal fusion score When the cross-modal fusion score is less than 0.9, the data cleaning process is started to remove outliers and perform interpolation processing; when ... When the error is <0.7, the sensor calibration procedure is triggered and the data source weight is dynamically adjusted. This mechanism improves the system's data utilization and positioning accuracy in complex environments, effectively reducing misjudgments caused by inconsistent data. Through the above optimization, the system realizes full-process intelligence from data collection to final judgment, greatly improving the recognition accuracy, warning speed and system adaptability of non-natural earthquake events, especially showing excellent robustness in areas with complex geological conditions and multi-source data interference environments.

[0039] The following are three scenarios and corresponding examples based on the above method, covering practical applications in different geological conditions, data quality, and event types: Applying the content of this invention to practical scenarios is as follows: Example (Identification of Blasting Events in Mining Areas) 1. Data Collection: (1) Natural earthquake data acquisition unit: obtains natural earthquake waveform data of nearby areas through the global seismic network as a comparison benchmark; (2) Non-natural earthquake data acquisition unit: obtain blasting operation records (time, charge amount, location) from the mining department, and collect high-frequency waveforms of blasting events (main frequency > 20 Hz, duration < 5 s); (3) Environmental interference data acquisition unit: deploy industrial noise sensors (monitoring mechanical vibration frequency 50-500 Hz) and meteorological sensors (monitoring wind speed and lightning); (4) Multi-source auxiliary data unit: obtain geological structure map of the mining area (1:50,000 scale) and satellite remote sensing data (InSAR deformation data).

[0040] 2. Data Analysis: (1) Feature overlap interval analysis unit: the posterior probability is calculated to be 0.85 (≥0.7), and it is marked as a high-confidence non-natural event; (2) Geological condition analysis unit: the geological similarity index was calculated to be 0.95 (≥0.8), and the default model parameters were used; (3) Microseismic prediction unit: The calculated earthquake warning score is 0.9 (≥0.8), triggering a red warning; (4) Multi-source data precision analysis unit: The cross-modal fusion score is calculated to be 0.92 (≥0.9), and a weighted average fusion strategy is adopted.

[0041] 3. Judgment and early warning: (1) The system automatically determines that it is a blasting incident, triggers a red alert, and links the emergency response system; (2) The judgment module classifies it as “non-natural earthquake (blasting)” and the early warning module notifies the mine management personnel.

[0042] 4. Optimization and verification: (1) The comprehensive optimization module adjusts the model weights according to the judgment results to improve the accuracy of blasting event recognition; (2) The verification module was tested with historical data, confirming that the system’s accuracy in high-confidence non-natural events reached 98%.

[0043] The beneficial effects are: the system quickly identifies blasting events, the warning response time is shortened to within 5 seconds, the multi-source data fusion strategy effectively reduces the misjudgment rate, and the positioning accuracy is improved to ±10 meters.

[0044] Example 2 (Identification of mine earthquake events under complex geological conditions) 1. Data Collection: (1) Natural earthquake data acquisition unit: obtains regional natural earthquake waveform data, but the geological conditions are quite different from the current area; (2) Non-natural earthquake data acquisition unit: collects the waveform of mining earthquake events (main frequency <10Hz, duration >10s), but the spectral characteristics partially overlap with natural earthquakes; (3) Environmental interference data acquisition unit: deploy long-period seismometers (period ≥ 100s) to monitor geological background noise (tidal effects, groundwater fluctuations); (4) Multi-source auxiliary data unit: obtain mining area fault distribution data and satellite optical images (resolution ≤ 1m).

[0045] 2. Data Analysis: (1) Feature overlap interval analysis unit: The posterior probability is calculated to be 0.65 (0.3 ≤ posterior probability < 0.7), and multi-model cross-validation is started.

[0046] (2) Geological condition analysis unit: the geological similarity index is calculated to be 0.55 (0.4 ≤ geological similarity index < 0.6), and the regional specificity model is enabled; (3) Microseismic prediction unit: The calculated earthquake warning score is 0.75 (0.6≤earthquake warning score<0.8), triggering a yellow warning; (4) Multi-source data accuracy analysis unit: The cross-modal fusion score is calculated to be 0.85 (0.7≤cross-modal fusion score<0.9), and the data cleaning process is started.

[0047] 3. Judgment and early warning: (1) The system combines regional specific models to determine that it is a mine earthquake event and issues a yellow warning; (2) If the judgment module classifies it as a “non-natural earthquake (mine tremor)”, the early warning module will prompt manual review.

[0048] 4. Optimization and verification: (1) The comprehensive optimization module adjusts the model parameters according to the geological similarity index and optimizes the weight of mine earthquake identification features; (2) The verification module passed simulation tests and confirmed that the system's recognition accuracy under complex geological conditions was improved to 92%.

[0049] The beneficial effects are: the regional specific model effectively adapts to complex geological conditions and reduces misjudgments, the data cleaning process improves the reliability of multi-source data, and the positioning accuracy is improved to ±50 meters.

[0050] Example (Natural Earthquake Misjudgment in Low-Quality Data Environment) 1. Data Collection: (1) Natural earthquake data acquisition unit: Acquires natural earthquake waveform data, but is affected by environmental interference (strong wind, lightning), resulting in increased noise levels; (2) Non-natural earthquake data acquisition unit: no relevant event data was collected; (3) Environmental interference data acquisition unit: Meteorological sensors detect strong winds (wind speed > level 6, generating 1-5 Hz noise) and lightning interference; (4) Multi-source auxiliary data unit: obtain meteorological data (precipitation, temperature) and human activity data (urban construction planning).

[0051] 2. Data Analysis: (1) Characteristic overlapping interval analysis unit: the calculated posterior probability is 0.45 (≤0.5), and natural earthquakes are given priority; (2) Geological condition analysis unit: the geological similarity index is calculated to be 0.75 (≥0.6), and the regional specificity model is enabled; (3) Microseismic prediction unit: The calculated earthquake warning score is 0.55 (<0.6), and routine monitoring is maintained; (4) Multi-source data accuracy analysis unit: The cross-modal fusion score is calculated to be 0.65 (<0.7), triggering the sensor calibration procedure.

[0052] 3. Judgment and early warning: (1) The system initially determined it to be a natural earthquake, but due to low data quality, no warning was triggered; (2) The judgment module marks it as an "uncertain event" and recommends manual review.

[0053] 4. Optimization and verification: (1) The integrated optimization module starts the sensor calibration process and adjusts the parameters of the industrial noise sensor and meteorological sensor; (2) The verification module is tested with historical data to confirm that the system’s misjudgment rate is reduced to below 5% in a low-quality data environment.

[0054] The beneficial effects are: the sensor calibration procedure effectively reduces the impact of environmental interference, improves data quality, the model reconstruction mechanism avoids misjudgment, and enhances the robustness of the system in low-quality data environments.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying and judging non-natural earthquake events based on deep learning, characterized in that: The following steps are involved: Step 1: Establish a system, which includes a seismic data acquisition module, a seismic data analysis module, an early warning module, a judgment module, a comprehensive optimization module and a verification module; Step 2: The earthquake data acquisition module acquires multi-source earthquake related data; Step 3: The seismic data analysis module calculates and analyzes the collected data; Step 4: The early warning module triggers an early warning based on the analysis results; Step 5: The judgment module performs final classification based on the analyzed event type; Step 6: The comprehensive optimization module optimizes the system according to the warning and judgment results of the warning module and the judgment module; Step 7: The verification module verifies the optimized system.

2. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 1, characterized in that: The seismic data acquisition module includes a natural seismic data acquisition unit, a non-natural seismic data acquisition unit, an environmental interference data acquisition unit and a multi-source auxiliary data unit; the seismic data analysis module includes a feature overlap interval analysis unit, a geological condition analysis unit, a microseismic prediction unit and a multi-source data accuracy analysis unit.

3. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The natural earthquake data acquisition unit obtains global natural earthquake waveforms, focal depths and magnitude data through the public database of the global seismic network.

4. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The non-natural earthquake data acquisition unit collects waveform and spectrum characteristic data of blasting, nuclear testing, mining earthquake, and engineering collapse events through multiple channels.

5. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The environmental interference data acquisition unit uses a multi-type sensor network to acquire industrial noise, meteorological interference, and geological background noise data.

6. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The multi-source auxiliary data unit obtains geological structure data, satellite remote sensing data, meteorological data and human activity data through cross-domain data fusion.

7. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The feature overlap interval analysis unit calculates the posterior probability based on the collected data , and its calculation formula is: ； In the formula, represents the posterior probability, represents the likelihood function, represents the prior probability, represents marginal probability; The posterior probability Used for Bayesian inference to update the probability of an event occurring "unnaturally".

8. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The geological condition analysis unit calculates the geological similarity index based on the collected data , and its calculation formula is: ； In the formula, represents the geological similarity index, Indicates the current density of underground rock formations, represents the reference underground rock density, Indicates the current propagation speed of the longitudinal wave, represents the propagation velocity of the reference longitudinal wave, Indicates the current propagation speed of the shear wave, represents the propagation velocity of the reference shear wave, represents the current station density, represents the reference station density; The geological similarity index Quantify the similarity of current geological conditions to reference geological conditions.

9. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The microseismic prediction unit calculates the earthquake early warning score based on the collected data , and its calculation formula is: ； In the formula, represents the earthquake early warning score, Indicates the current waveform. Indicates the similarity of natural / non-natural earthquake waveform templates, Indicates the current waveform spectrum energy, Indicates the spectrum energy of natural / non-natural earthquakes, represents the geological similarity index, Indicates the factors affecting meteorology and human activities, 、 、 、 Respectively represent weight coefficients; The " "Combining waveform, spectrum, geological and external environmental factors to calculate earthquake early warning scores .

10. The method for identifying and determining non-natural earthquake events based on deep learning according to claim 2, characterized in that: The multi-source data accuracy analysis unit calculates the cross-modal fusion score based on the collected data , and its calculation formula is: ； In the formula, represents the cross-modal fusion score, Indicates the signal-to-noise ratio, which reflects the quality of the radar signal. Indicates the matching degree between the acoustic signal spectrum and the template spectrum, Indicates the timestamp change value of radar data and sound wave data, represents the time difference threshold, represents the spatial coordinates of the radar data, represents the spatial coordinates of the sound wave data, Indicates the maximum allowed spatial difference; The cross-modal fusion score Assess the reliability and consistency of multi-source data.

Images