A method and system for predicting earthquake risk based on multi-source data fusion

Abstract

The application discloses a kind of based on multi-source data fusion earthquake risk prediction method and system, including obtaining multi-source monitoring data and extension data, construct precursor monitoring network, with the physical constraint of fault mechanics constitutive relation as earthquake risk characteristic is carried out causal inference, determine precursor characteristic subset and carry out three-level cross correction of space consistency, mechanical coupling and environmental interference, determine local confidence weight, build Seismogenic environment field and precursor response field, determine Seismogenic activity reference and carry out nonlinear probability modulation according to dynamic anomaly information and local confidence weight, get grid earthquake danger probability, carry out criticality classification and combine ground motion attenuation law parameters and hazard-affected body vulnerability parameters, obtain earthquake risk prediction result and different warning strategy.The present application improves the reliability of precursor identification and spatial prediction confidence, realizes the fine early warning for different risk scenarios.

Description

Technical Field

[0001] This invention belongs to the field of earthquake monitoring and early warning technology, and in particular relates to an earthquake risk prediction method and system based on multi-source data fusion. Background Technology

[0002] Earthquake risk prediction is a core means of mitigating earthquake disaster losses. With the development of space-based Earth observation, deep-well observation, and geophysical field monitoring technologies, multi-source monitoring data provides rich information for earthquake precursor identification. Earthquake risk prediction methods based on multi-source data fusion can comprehensively utilize multi-dimensional observation information such as deformation, seismic activity, underground fluids, and geophysical fields. Compared with single monitoring methods, it has the advantages of information complementarity, wide spatial coverage, and strong physical constraints, which has important practical significance and social value for improving regional earthquake risk assessment capabilities and supporting the government's refined disaster prevention and mitigation decision-making.

[0003] However, existing technologies still have significant shortcomings in practical applications: First, multi-source monitoring data are often analyzed in isolation, lacking physical constraints centered on fault mechanical constitutive relations. The causal relationship between precursor features and stress state is weak, leading to low reliability of anomaly identification and a high false alarm rate. Second, existing methods rarely consider the impact of spatial sparsity of monitoring networks and heterogeneity of geological structures on prediction results, making it difficult to conduct spatially adaptive confidence assessment of prediction results, resulting in insufficient prediction credibility in highly complex areas. In addition, traditional early warning strategies typically use uniform thresholds or single-dimensional grading, failing to fully couple seismic motion attenuation laws and vulnerability parameters of disaster-bearing bodies, thus unable to support differentiated disaster prevention decisions for different risk scenarios. To address these issues, this invention proposes a seismic risk prediction method and system based on multi-source data fusion. By introducing key technologies such as mechanically constrained causal inference, three-level cross-correction, local confidence weighting, and a two-dimensional decision matrix, it achieves improved reliability of precursor identification, adaptive spatial confidence quantification, and differentiated customization of early warning strategies, providing a new technical approach for accurate seismic risk prediction and refined disaster prevention and mitigation management. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for earthquake risk prediction based on multi-source data fusion.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Acquire multi-source monitoring data and extended data, construct a precursor monitoring network, extract earthquake risk characteristics, use fault mechanical constitutive relations as physical constraints to make causal inferences on earthquake risk characteristics, and determine a subset of precursor characteristics; Based on multi-source monitoring data, a precursor feature subset is subjected to a three-level cross-correction process considering spatial consistency, mechanical coupling, and environmental interference to obtain precursor corrected data. The sparsity of monitoring data is calculated based on the monitoring network configuration data, and the complexity of geological structure is calculated based on the geological structural background data. The sparsity of monitoring data and the complexity of geological structure are spatially coupled to determine the local confidence weight. Based on historical earthquake catalog data, the background seismogenic activity level is determined, and a seismogenic environment field and a precursor response field are constructed. The seismogenic environment field and the precursor response field are then reorganized differentially by combining local confidence weights to determine the seismogenic activity benchmark. Dynamic anomaly information is extracted from precursor correction data as a correction factor, and local confidence weight is used as a modulation coefficient to perform nonlinear probability modulation on the earthquake activity benchmark to obtain the grid point earthquake occurrence hazard probability. The critical classification of the seismic hazard probability at grid points is carried out. Combined with the seismic motion attenuation law parameters and the vulnerability parameters of the disaster-bearing body, the earthquake risk prediction results are obtained, and differentiated early warning strategies are determined. The monitoring data includes spatial deformation monitoring data, seismic activity monitoring data, underground fluid monitoring data, and geophysical field monitoring data; The extended data includes historical earthquake catalog data, monitoring network configuration data, ground motion attenuation parameters, geological structure background data, vulnerability parameters of disaster-bearing bodies, and environmental disturbance data.

[0006] Furthermore, the method for determining the precursor feature subset includes: Acquire monitoring and extended data, extract active fault segments, and determine the overall activity of each fault segment; Based on the comprehensive activity of fault segments and the location of key structural components, physical monitoring nodes and virtual precursor nodes are deployed. The virtual precursor nodes are equivalent observation points constructed in the blind areas covered by physical points based on the continuity of geological structures and physical field gradient extrapolation. The historical background benchmark of the virtual precursor nodes is determined by physical field gradient interpolation using historical observation data from physical monitoring nodes in the spatial neighborhood. The current observation value of the virtual precursor nodes is determined by physical field gradient extrapolation using current observation data from physical monitoring nodes in the spatial neighborhood. Determine the fault spatial relationship of each node and construct a precursor monitoring network; the fault spatial relationship characterizes the fault segmentation of the node and its relative spatial position with respect to the fault. Historical background benchmarks are determined based on historical observation data from each entity monitoring node. Observational deviations are calculated, and exponential decay of distance from faults and logarithmic enhancement of observation depth are introduced to obtain seismic risk characteristic values. These seismic risk characteristics include deformation field characteristics, seismic activity field characteristics, underground fluid field characteristics, and geophysical field characteristics. Determine the stress state parameters and causal correlation parameters of each feature, use the fault mechanical constitutive relation as a physical constraint, make causal inferences on earthquake risk characteristics, and screen to obtain a candidate set of precursor features; A physical consistency test is performed on the candidate set of precursor features to determine the subset of precursor features; the subset of precursor features represents a set of earthquake-induced anomaly response information that has a mechanical causal relationship with the fault stress state and is self-consistent across physical fields.

[0007] Furthermore, the method for obtaining precursor correction data includes: A spatial influence domain is defined, and each node in the subset of precursor features is extracted as a retrieval center. Similar precursor features are retrieved to obtain a neighborhood similar feature set; the similar precursor features are precursor features belonging to the same physical field type. Heterogeneous precursor features are extracted from the same fault segment to obtain a set of heterogeneous features in the same domain; the heterogeneous precursor features are precursor features belonging to different physical field types; The directional consistency ratio of similar feature sets in the neighborhood is calculated, and the abnormal properties of each feature are determined and corrected according to a preset threshold and the fault spatial relationship of each node; the abnormal properties include regional group anomalies, local structural anomalies and non-structural disturbance anomalies. Based on the rock fracture mechanical response chain, a mechanical coupling test is performed on the heterogeneous feature sets in the same domain, and the mechanical response categories are classified and corrected accordingly. The rock fracture mechanical response chain characterizes the directional sequence relationship of different physical field responses of the fault during stress loading until instability and rupture, according to the mechanical coupling law. This includes the directional sequence of deceleration of cross-fault displacement rate, enhancement of surface strain compression, decrease of georesistivity, and rise of groundwater level under compressive stress. The mechanical response categories include mechanical synergistic features and mechanical anisotropic features. Screening environmentally sensitive features based on the inherent sensitivity of each physical field type to environmental factors; Environmental disturbance data is extracted, an environmental response model is constructed, and the environmental driving components of environmentally sensitive features are separated to obtain residual anomaly components. The environmental response model characterizes the response function relationship between precursor feature observations and environmental disturbance data. By combining the results of anomaly property correction, mechanical coupling correction, and environmental interference stripping, precursor correction data are obtained.

[0008] Furthermore, the method for determining the local confidence weights includes: The effective monitoring capability index is calculated based on the monitoring network configuration data; the effective monitoring capability index is the number of equivalent nodes within the effective monitoring range of each grid point. Determine the ideal network density for the region based on the comprehensive activity level of fault segments, and calculate the sparsity of monitoring data. Geological structural background data is acquired, and a structural heterogeneity index is calculated; the structural heterogeneity index reflects the degree of spatial difference in the physical properties of the geological medium within the monitoring target area. The complexity of geological structures is determined based on the structural heterogeneity index; the complexity of geological structures is divided into three levels according to the structural heterogeneity index. A two-dimensional coupled modulation function is constructed using the sparsity of monitoring data as the supply term and the complexity of geological structure as the loss term. The local confidence weight is calculated and mapped to the local confidence weight level. The local confidence weight is positively correlated with the sparsity of monitoring data and negatively correlated with the complexity of geological structure.

[0009] Furthermore, the method for determining the seismogenic activity benchmark includes: Seismic activity parameters are extracted from each fault segment; these parameters include historical earthquake frequency, magnitude distribution parameters, and cumulative moment release rate. The seismic activity parameters are corrected based on the comprehensive activity level of fault segments to determine the background seismogenic activity level; the background seismogenic activity level characterizes the seismic intensity benchmark of each fault segment under long-term stress background. The seismogenic environment field is constructed by mapping the background seismogenic activity level and the fault segmentation of each grid point; the seismogenic environment field characterizes the static seismogenic hazard level of each grid point under the long-term seismogenic background. Dynamic anomaly information is extracted from the precursor correction data and mapped in combination with the fault spatial relationship of each node to construct a precursor response field; the precursor response field characterizes the dynamic seismic risk increment of each grid point under short-term precursor disturbances. Based on the local confidence weight level of each grid point, the earthquake-generating environment field and the precursor response field are reorganized differentially to determine the baseline of earthquake-generating activity.

[0010] Furthermore, the method for obtaining the grid point seismic hazard probability includes: Extract the corrected feature values ​​of each precursor feature from the precursor correction data and calculate the dynamic anomaly comprehensive index; the dynamic anomaly comprehensive index is the grid-point dimensionless anomaly intensity obtained by aggregated the corrected feature values ​​of each precursor feature through root mean square. The modified channel opening coefficient is determined based on the local confidence weight level of each grid point; the modified channel opening coefficient is positively correlated with the local confidence weight level. The seismogenic activity benchmark is nonlinearly probabilistically modulated based on the dynamic anomaly comprehensive index and the modified channel opening coefficient; the nonlinear probabilistic modulation uses the seismogenic activity benchmark as the base term and the coupling term of the dynamic anomaly comprehensive index and the modified channel opening coefficient as the incremental driving term. The upper limit boundary is defined as the difference between the upper limit constraint value and the earthquake activity benchmark. The probability of earthquake occurrence at grid points is modulated based on the earthquake activity benchmark value to obtain the earthquake occurrence probability at grid points. The upper limit constraint value is not lower than the earthquake activity benchmark value.

[0011] Furthermore, the method for obtaining earthquake risk prediction results includes: A time evolution analysis was performed on the seismic hazard probability of the grid points to calculate the hazard probability growth rate and cumulative hazard index; The criticality level is determined based on the rate of increase in the probability of danger and the cumulative risk index; the criticality level includes four levels: low critical, medium critical, high critical, and extremely high critical. Extract the parameters of the ground motion attenuation law; the parameters of the ground motion attenuation law include the peak ground acceleration attenuation coefficient and the spectral attenuation coefficient. Based on the seismic hazard probability and ground motion attenuation parameters of each grid point, the ground motion intensity of each grid point is predicted; the ground motion intensity includes peak ground acceleration and characteristic period. Extract the vulnerability parameters of the disaster-bearing body; the vulnerability parameters of the disaster-bearing body include the seismic resistance of the building structure, population exposure density and vulnerability of lifeline engineering. The loss rate of the disaster-bearing body is calculated based on the seismic intensity and the vulnerability parameters of the disaster-bearing body. The loss rate of the disaster-bearing body is calculated based on the log-normal vulnerability curve. The median parameter of the log-normal vulnerability curve is determined based on the seismic resistance of the building structure, and the coefficient of variation is determined based on the building structure type. A two-dimensional decision matrix is ​​constructed by combining criticality classification and disaster-bearing body loss rate; the two-dimensional decision matrix has criticality classification as the horizontal axis and disaster-bearing body loss rate as the vertical axis. Differentiated early warning strategies are determined based on a two-dimensional decision matrix, and earthquake risk prediction results are output.

[0012] Secondly, an earthquake risk prediction system based on multi-source data fusion includes: Causal inference module: used to acquire multi-source monitoring data and extended data, construct a precursor monitoring network, perform causal inference on earthquake risk characteristics with fault mechanical constitutive relations as physical constraints, and determine a subset of precursor features; Precursor correction module: used to perform three-level cross correction of spatial consistency, mechanical coupling and environmental interference on a subset of precursor features based on multi-source monitoring data, to obtain precursor correction data; Confidence assessment module: used to calculate the sparsity of monitoring data based on the monitoring network configuration data, calculate the complexity of geological structure based on the geological structure background data, and spatially couple the two to determine the local confidence weight; Earthquake Prediction Baseline Module: Used to determine the background earthquake prediction activity level based on historical earthquake catalog data, construct the earthquake prediction environment field and precursor response field, and determine the earthquake prediction activity baseline by combining local confidence weight nonlinear reorganization; The hazard probability module is used to extract dynamic anomaly information from the precursor correction data as a correction factor, use the local confidence weight as a modulation coefficient, modulate the earthquake activity benchmark, and obtain the grid point earthquake occurrence hazard probability. Early warning strategy module: used to classify the critical probability of earthquake occurrence at grid points, combine the ground motion attenuation law parameters and the vulnerability parameters of the disaster-bearing body, obtain earthquake risk prediction results and determine differentiated early warning strategies.

[0013] The beneficial effects of this invention are: This invention is a method and system for earthquake risk prediction based on multi-source data fusion. Compared with existing technologies, this invention has the following technical advantages: This invention effectively improves the reliability and physical consistency of multi-source precursor feature identification and reduces the risk of false alarms caused by non-tectonic interference by using causal inference with fault mechanical constitutive relations as physical constraints and a three-level cross correction of spatial consistency, mechanical coupling and environmental interference. This invention establishes a local confidence weighting system by monitoring the spatial coupling between data sparsity and geological structure complexity, thereby achieving adaptive confidence quantification of prediction results in the spatial dimension and enhancing the scientific validity and credibility of prediction results for complex geological areas. This invention achieves differentiated early warning strategies for different risk scenarios by differentiating and recombining the earthquake-generating environment field and the precursor response field and nonlinear probability modulation, combined with a two-dimensional decision matrix of criticality classification and vulnerability parameters of the disaster-bearing body, which significantly improves the supporting effectiveness of earthquake risk prediction for disaster prevention and mitigation decision-making. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of an earthquake risk prediction method based on multi-source data fusion according to the present invention. Detailed Implementation

[0015] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0016] The present invention provides a method and system for earthquake risk prediction based on multi-source data fusion, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Acquire multi-source monitoring data and extended data, construct a precursor monitoring network, extract earthquake risk characteristics, use fault mechanical constitutive relations as physical constraints to make causal inferences on earthquake risk characteristics, and determine a subset of precursor characteristics; Based on multi-source monitoring data, a precursor feature subset is subjected to a three-level cross-correction process considering spatial consistency, mechanical coupling, and environmental interference to obtain precursor corrected data. The sparsity of monitoring data is calculated based on the monitoring network configuration data, and the complexity of geological structure is calculated based on the geological structural background data. The sparsity of monitoring data and the complexity of geological structure are spatially coupled to determine the local confidence weight. Based on historical earthquake catalog data, the background seismogenic activity level is determined, and a seismogenic environment field and a precursor response field are constructed. The seismogenic environment field and the precursor response field are then reorganized differentially by combining local confidence weights to determine the seismogenic activity benchmark. Dynamic anomaly information is extracted from precursor correction data as a correction factor, and local confidence weight is used as a modulation coefficient to perform nonlinear probability modulation on the earthquake activity benchmark to obtain the grid point earthquake occurrence hazard probability. The critical classification of the seismic hazard probability at grid points is carried out. Combined with the seismic motion attenuation law parameters and the vulnerability parameters of the disaster-bearing body, the earthquake risk prediction results are obtained, and differentiated early warning strategies are determined. The monitoring data includes spatial deformation monitoring data, seismic activity monitoring data, underground fluid monitoring data, and geophysical field monitoring data; The extended data includes historical earthquake catalog data, monitoring network configuration data, ground motion attenuation parameters, geological structure background data, vulnerability parameters of disaster-bearing bodies, and environmental disturbance data.

[0017] In this embodiment, the method for determining a subset of precursor features includes: Acquire monitoring and extended data, extract active fault segments, and determine the overall activity of each fault segment; Based on the comprehensive activity of fault segments and the location of key structural components, physical monitoring nodes and virtual precursor nodes are deployed. The virtual precursor nodes are equivalent observation points constructed in the blind areas covered by physical points based on the continuity of geological structures and physical field gradient extrapolation. The historical background benchmark of the virtual precursor nodes is determined by physical field gradient interpolation using historical observation data from physical monitoring nodes in the spatial neighborhood. The current observation value of the virtual precursor nodes is determined by physical field gradient extrapolation using current observation data from physical monitoring nodes in the spatial neighborhood. Determine the fault spatial relationship of each node and construct a precursor monitoring network; the fault spatial relationship characterizes the fault segmentation of the node and its relative spatial position with respect to the fault. Historical background benchmarks are determined based on historical observation data from each entity monitoring node. Observational deviations are calculated, and exponential decay of distance from faults and logarithmic enhancement of observation depth are introduced to obtain seismic risk characteristic values. These seismic risk characteristics include deformation field characteristics, seismic activity field characteristics, underground fluid field characteristics, and geophysical field characteristics. Determine the stress state parameters and causal correlation parameters of each feature, use the fault mechanical constitutive relation as a physical constraint, make causal inferences on earthquake risk characteristics, and screen to obtain a candidate set of precursor features; A physical consistency test is performed on the candidate set of precursor features to determine the subset of precursor features; the subset of precursor features represents a set of earthquake-induced anomaly response information that has a mechanical causal relationship with the fault stress state and is self-consistent across physical fields. In actual evaluation, the physical field gradient interpolation of the virtual precursor node adopts the inverse distance weighted interpolation method or the Kriging interpolation method. The interpolation weight is inversely proportional to the distance between the physical node and the virtual node, and decreases exponentially as the distance increases. The expression for the earthquake risk characteristic value is: ; in, For the first The node Earthquake risk feature extraction values ​​based on class characteristics. For the first The node The current observation of the class feature. For the first The node Historical background benchmark values ​​of class features For the first The node Background standard deviation of class features For the first The surface projection distance from each node to the nearest active fault segment The characteristic distance of fault influence, For depth sensitivity coefficient, For the first The observation depth of each node, For depth-sensitive feature scale; The stress state parameters are selected based on the observation conditions in the monitoring network configuration data: When the target fault segment has cross-fault GNSS or InSAR survey lines, and the data integrity rate of the survey lines is not lower than the preset integrity rate threshold and the continuous observation time is not less than the preset duration threshold, the cross-fault deformation rate deviation is selected as the stress state parameter; when the target fault segment has deep well borehole strain observations, and the strain data integrity rate is not lower than the preset integrity rate threshold and the continuous observation time is not less than the preset duration threshold, the borehole strain accumulation is selected as the stress state parameter; when both of the above conditions are met, the cross-fault deformation rate deviation is selected first, and the borehole strain accumulation is used as the verification parameter; when neither of the two observation conditions is met, the fault segment will not perform physical constraint causal inference for the time being, and only statistical causal strength screening is used; The causal correlation parameters include physical response matching degree, statistical causal strength, and temporal consistency. The physical response matching degree is determined based on the consistency between the sign of the change direction of the feature and the sign of the theoretical direction predicted by the fault mechanical constitutive relation. When the signs of the directions are consistent, it is 1; when the signs of the directions are opposite, it is 0; when there is no theoretical response but there is a deviation in reality, the value is the exponent of the product of the negative sensitivity coefficient with the natural constant as the base and the absolute value of the feature deviation. The statistical causal strength is the value of the partial correlation coefficient between the characteristics and stress state parameters after controlling for environmental disturbance factors, after hyperbolic tangent saturation mapping. The time consistency score is determined based on the deviation between the time lead of the characteristic anomaly relative stress state parameter change and the ideal lead time. Based on the elastic rebound theory, during the stress accumulation stage of a fault, the relative motion of the two blocks in the locked segment is suppressed, and the cross-fault displacement rate shows a deceleration or reverse trend. Accordingly, the deviation of the cross-fault displacement rate is marked as a compressive stress state with a negative sign. Based on rock fracture experiments, under increased confining pressure, the closure of microcracks leads to a decrease in resistivity, and the increase in pore pressure drives fluid migration to shallower areas, resulting in a rise in water level. Based on this, the expected direction of the decrease in resistivity and the rise in water level under compressive stress is established. The fault mechanical constitutive relation adopts the Coulomb fracture criterion as a physical constraint, and its expression is: ; in, For shear stress, It is normal stress. Pore ​​pressure, The coefficient of friction, It is cohesive force; During the stress accumulation stage, when As the fracture strength approaches, the characteristics of each physical field should exhibit a change in the same direction as the increase in pore pressure and stress. The causal inference uses a direction sign comparison algorithm to compare the change direction signs of each earthquake risk feature with the theoretical direction signs predicted by the Coulomb rupture criterion, and to screen out a candidate set of precursor features. The physical consistency test includes cross-field consistency constraint test, mechanical coupling consistency constraint test and time scale consistency constraint test; This embodiment uses the northern segment of the Anninghe Fault at the Sichuan-Yunnan Earthquake Scientific Experimental Field as the prediction target area, adjacent to the western segment of the Zemuhe Fault. The prediction time point is 30 days before the occurrence of a strong earthquake. Four physical monitoring nodes and one virtual precursor node are set up in the target area. The monitoring data time span meets the continuous observation duration threshold of 3 years, and the data integrity rate is not lower than the preset integrity rate threshold of 85%. Using historical M7.5 earthquakes in the northern segment of the Anninghe Fault as the back-calculation sample, the parameter search grid is as follows: , , The step sizes are 0.2, 0.1, and 0.2 respectively. The optimization objective is to maximize the agreement between the predicted probability 30 days before the earthquake and the actual earthquake occurrence. This is achieved through a traversal search. , , ; Long-term slip rate, historical maximum earthquake magnitude, and short-term seismic activity indicators were extracted from active fault segmentation and extension data to determine the overall activity level: The northern segment of the Anninghe Fault has a long-term slip rate of 8.5 mm / year, a historical maximum earthquake magnitude of 7.5, and a cumulative frequency of 18 small earthquakes in the past 5 years, classifying it as highly active; the western segment of the Zemuhe Fault has a long-term slip rate of 4.2 mm / year, a historical maximum earthquake magnitude of 6.8, and a cumulative frequency of 9 small earthquakes in the past 5 years, classifying it as moderately active. Based on the comprehensive activity level of the fault segments and the location of key structural components, physical monitoring nodes and virtual precursor nodes were deployed: KMIN, a physical node located in the northern segment of the Anninghe Fault, 0.8 km from the fault, with an observation depth at the surface and a physical field type of spatial deformation; FL-01, a physical node located in the northern segment of the Anninghe Fault, 2.5 km from the fault, with an observation depth of 800 m and a physical field type of underground fluid; RT-02, a physical node located in the northern segment of the Anninghe Fault, 1.5 km from the fault, with an observation depth of 800 m and a physical field type of geophysical field; YB-03, a physical node located in the northern segment of the Anninghe Fault, 1.2 km from the fault, with an observation depth of 1200 m and a physical field type of spatial deformation; VN-01, a virtual node located in the northern segment of the Anninghe Fault, 1.5 km from the fault, is a virtual precursor node for the deformation field, determined by extrapolation of the physical field gradient from neighboring physical deformation monitoring nodes. VN-01 participates in the spatial consistency test only as a neighborhood reference node of the deformation field, and does not enter the mechanical coupling test of the heterogeneous feature set in the same domain; Based on the historical observation data of each entity monitoring node over the past three years, the background baseline value and background standard deviation were calculated, and the current observation value was extracted. The observed deviations were calculated as follows: KMIN cross-fault displacement rate was -1.67, FL-01 groundwater level was +2.75, RT-02 ground resistivity was -1.56, YB-03 borehole strain accumulation was +1.47, and regional b-value scan was -2.13. Introducing exponential decay of distance from faults and logarithmic enhancement of observation depth, , , The calculated values ​​are: KMIN deformation field characteristic value -1.42, FL-01 fluid field characteristic value +2.63, RT-02 geophysical field characteristic value -1.16, YB-03 deformation field characteristic value +2.01, and regional b-value seismic activity field characteristic value, which is taken as -1.85 after regional gridding mapping. According to the monitoring network configuration data, there is a cross-fault GNSS survey line KMIN in the northern section of the Anninghe Fault. The data integrity rate of the survey line is 95%, which is not lower than the preset integrity rate threshold of 85%. The continuous observation period is 3 years, which is not less than the preset duration threshold of 2 years. The deviation of the cross-fault deformation rate of KMIN is selected as the stress state parameter. At the same time, the cumulative strain of the YB-03 borehole is used as the verification parameter. Using fault mechanical constitutive relations as physical constraints, causal inference was performed by a direction sign comparison algorithm. The comparison results of the direction signs of each feature change direction were consistent with the theoretical expected direction signs, and the physical response matching degree was 1. The statistical causal strengths, mapped by the hyperbolic tangent of the partial correlation coefficient, were 0.84, 0.75, 0.78, and 0.81, respectively. The ideal lead time is 10 days, the time tolerance scale is 15 days, and the time consistency scores are 0.72, 0.59, 0.87, and 0.63, respectively. The calculated comprehensive precursor indices were 0.6, 0.44, 0.68, and 0.51, respectively. With a threshold of 0.5, the candidate set of precursor features is KMIN deformation deceleration, RT-02 resistivity decrease, and region b value decrease; Perform a physical consistency check on the candidate set of precursor features: Cross-field consistency constraint test: The candidate set includes three types of physical fields: deformation field, geophysical field, and seismic activity field, and must meet at least two of the requirements; Mechanical coupling consistency constraint test: Under compressive stress, the directional sequence of deformation deceleration, resistivity decrease and b-value decrease conforms to the rock fracture mechanical response chain, and the directions are consistent. Time scale consistency constraint test: Deformation leads by 5 days, geoelectricity leads by 12 days, and b-value leads by 3 days, all within 10. 1 The magnitude is on the order of astronomical levels, but the deviation is less than one order of magnitude. The test passed, and the precursor feature subset was determined to be KMIN deformation deceleration, RT-02 resistivity decrease, and region b value decrease.

[0018] In this embodiment, the method for obtaining precursor correction data includes: A spatial influence domain is defined, and each node in the subset of precursor features is extracted as a retrieval center. Similar precursor features are retrieved to obtain a neighborhood similar feature set; the similar precursor features are precursor features belonging to the same physical field type. Heterogeneous precursor features are extracted from the same fault segment to obtain a set of heterogeneous features in the same domain; the heterogeneous precursor features are precursor features belonging to different physical field types; The directional consistency ratio of similar feature sets in the neighborhood is calculated, and the abnormal properties of each feature are determined and corrected according to a preset threshold and the fault spatial relationship of each node; the abnormal properties include regional group anomalies, local structural anomalies and non-structural disturbance anomalies. Based on the rock fracture mechanical response chain, a mechanical coupling test is performed on the heterogeneous feature sets in the same domain, and the mechanical response categories are classified and corrected accordingly. The rock fracture mechanical response chain characterizes the directional sequence relationship of different physical field responses of the fault during stress loading until instability and rupture, according to the mechanical coupling law. This includes deceleration of cross-fault displacement rate under compressive stress, enhanced surface strain compression, decrease in georesistivity, and rise in groundwater level. The mechanical response categories include mechanical synergistic features and mechanical anisotropic features. Screening environmentally sensitive features based on the inherent sensitivity of each physical field type to environmental factors; Environmental disturbance data is extracted, an environmental response model is constructed, and the environmental driving components of environmentally sensitive features are separated to obtain residual anomaly components. The environmental response model characterizes the response function relationship between precursor feature observations and environmental disturbance data. By combining the results of anomaly property correction, mechanical coupling correction, and environmental disturbance removal, precursor correction data are obtained; In actual assessment, the radius of the spatial influence domain is taken along the direction as the statistical average of the bandwidth of the main fault influence band in the region, and the vertical direction is determined according to the difference in medium wave velocity. The directional consistency ratio is the ratio of the number of nodes with consistent orientations within a spatial neighborhood to the total number of nodes within the spatial neighborhood. In the calculation of the directional consistency ratio, the virtual precursor node determines the differentiated weights based on the interpolation error propagation. The corresponding directional consistency ratio is the ratio of the weighted sum of the directional signs of each node in the spatial neighborhood according to the differentiated weights to the total weight. The preset threshold for the directional consistency ratio is determined by retrospectively analyzing historical earthquake cases in the region and statistically analyzing the minimum consistency ratio of the actual population anomalies before the earthquake. The regional group anomaly is a precursor feature whose directional consistency ratio reaches a preset threshold, and the original feature value is retained; the local structural anomaly is a precursor feature whose directional consistency ratio is lower than the preset threshold but whose node is close to the fault and whose observation depth is large, and the original feature value is retained; the non-structural interference anomaly is a precursor feature whose directional consistency ratio is lower than the preset threshold but whose node is far from the fault and whose observation depth is shallow, and it is downweighted and corrected. The specific directional sequence of the rock fracture mechanical response chain under compressive stress is as follows: deceleration or reversal of the cross-fault displacement rate, enhanced surface strain compression, decrease in ground resistivity, increase in groundwater level, increase in radon concentration, and increase in microseismic frequency. The theoretical lag time ranges exist for the responses of each physical field in the rock fracture mechanics response chain: the cross-fault displacement rate response lags behind the stress change by 0-7 days, the surface strain response lags behind by 0-5 days, the ground resistivity response lags behind by 3-15 days, the groundwater level response lags behind by 7-30 days, the radon concentration response lags behind by 15-60 days, and the microseismic frequency response lags behind by 0-3 days. The allowable deviation window for the mechanical coupling test is that the deviation between the time lead of each precursor feature and the theoretical lag time range does not exceed one order of magnitude, and the amplitude ratio is that the relative amplitude ratio of the corrected feature values ​​of each precursor feature is within the range of the theoretical amplitude ratio calibrated by the rock fracture experiment. The sequence matching criteria for the mechanical coupling test include consistent direction signs, time lead within the allowable deviation window, and relative amplitude ratio within the theoretical range. If all three criteria are met, it is judged as a mechanical synergy feature and is enhanced and corrected. If any criterion is not met, it is judged as a mechanical anisotropy feature and is reduced in weight based on the theoretical response hysteresis characteristics and observation depth of each physical field. The environmental response model is a response function model based on physical mechanisms, including a seepage response function and a thermal expansion response function. The seepage response function adopts a rainfall-water level lag correlation model, and the lag time is calibrated according to the thickness of the vadose zone. The thermal expansion response function adopts a linear thermal expansion coefficient model, and the coefficient is determined according to the regional lithological statistical mean. The neighborhood feature set is centered on KMIN, within a 20km radius along the fault strike and a 10km radius perpendicular to the strike, and retrieves similar deformation field features, including KMIN, VN-01 virtual, and neighboring ZM-01. VN-01 participates in the calculation of the directional consistency ratio as a neighborhood reference node in the deformation field. The differential weight is determined based on the interpolation distance. When the interpolation distance is 1.5 times the average spacing between entity nodes, the differential weight is 0.5 and the entity node weight is 1. The heterogeneous feature set is a set of heterogeneous precursor features extracted within the same fault segment in the northern section of the Anninghe Fault, including KMIN deformation, RT-02 resistivity, and regional b-value. A decrease in the b-value in the region corresponds to an abnormal enhancement, and its direction sign is defined as negative, which is consistent with the direction sign of deformation deceleration and resistivity decrease under compressive stress. Within the deformation field of the neighborhood feature set, the KMIN direction sign is negative, the VN-01 extrapolation direction sign is negative, the ZM-01 direction sign is negative, the number of nodes with consistent directions is 3, the total number of nodes is 3, and the direction consistency ratio is 1. Based on the preset regional group anomaly threshold of 0.60, if the KMIN deformation rate direction consistency ratio is not less than 0.60, it is judged as a regional group anomaly, and the original feature value -1.42 is retained. RT-02 resistivity is unique among similar stations in the neighborhood. Its directional consistency ratio is below the threshold, but the node is close to the fault and the observation depth is large. Based on the spatial relationship of the fault, it is determined to be a local structural anomaly, and the original characteristic value of -1.16 is retained. The value of region b is the result of the region scan and is retained directly. Based on the rock fracture mechanical response chain, the direction sequence of the compressive stress state is deformation deceleration, resistivity decrease, and b-value decrease. Mechanical coupling test is performed on the heterogeneous feature set in the same domain: KMIN deformation deceleration, which is negative and conforms to the direction sequence; RT-02 resistivity decrease, which is negative and conforms to the direction sequence; regional b-value decrease, which is negative and conforms to the direction sequence. If all three features simultaneously satisfy the conditions of consistent direction sign and time lead within the allowable deviation window, they are determined to be mechanically coordinated features and are enhanced and corrected with an enhancement coefficient of 1.15. The environmentally sensitive features selected are: FL-01 groundwater level, which, although not in the precursor feature subset, is verified by environmental stripping of the original fluid features during the correction stage; and RT-02 resistivity, which is significantly affected by temperature changes. For the FL-01 water level, a seepage response function was constructed. The cumulative rainfall within 30 days was 132 mm. After using the rainfall-water level lag correlation model, the lag time was 5 days, calibrated by the thickness of the vadose zone. After removing the rainfall driving component, the standardized value of the residual anomaly component was +1.90. For the resistivity of RT-02, a thermal expansion response function was constructed. The temperature rose by 6.5℃ in the past 30 days. Using a linear thermal expansion coefficient model, the coefficient was calculated based on the statistical average of 8×10⁻⁶ for regional limestone. -6 / ℃ was determined, and after stripping the temperature-driven component, the normalized value of the residual abnormal component was -1.30; Based on the combined results of anomaly correction, mechanical coupling correction, and environmental interference stripping: KMIN deformation, retained, corrected eigenvalue -1.42; RT-02 resistivity, mechanically enhanced and after environmental stripping, corrected eigenvalue -1.50; region b-value, retained, corrected eigenvalue -1.85.

[0019] In this embodiment, the method for determining local confidence weights includes: The effective monitoring capability index is calculated based on the monitoring network configuration data; the effective monitoring capability index is the number of equivalent nodes within the effective monitoring range of each grid point. Determine the ideal network density for the region based on the comprehensive activity level of fault segments, and calculate the sparsity of monitoring data. Geological structural background data is acquired, and a structural heterogeneity index is calculated; the structural heterogeneity index reflects the degree of spatial difference in the physical properties of the geological medium within the monitoring target area. The complexity of geological structures is determined based on the structural heterogeneity index; the complexity of geological structures is divided into three levels according to the structural heterogeneity index. A two-dimensional coupled modulation function is constructed using the sparsity of monitoring data as the supply term and the complexity of geological structure as the loss term. Local confidence weights are calculated and mapped to local confidence weight levels. The local confidence weights are positively correlated with the sparsity of monitoring data and negatively correlated with the complexity of geological structure. In actual assessment, the sparsity of the monitoring data is the ratio of the effective monitoring capability index to the ideal network density of the region; The structural heterogeneity index is calculated by coupling the fault density as the base term, the lithological variation gradient as the exponential enhancement term, and the product of the topographic cutting depth and the distribution density of historical earthquake rupture zones as the product modulation term, after normalization. The expression for the local confidence weight is: ; in, For local confidence weights, To monitor data sparsity, Due to the complexity of the geological structure, As the regional baseline sparsity, the number of effective monitoring nodes corresponding to the ideal network density of the lowest active fault segment in the region is taken. The regional baseline complexity is determined by the average heterogeneity level of the regional geological and structural background data. The minimum confidence level is the lower limit weight. The highest confidence level upper limit weight, The sparsity sensitivity coefficient, The sparsity saturation index is... This is the complexity attenuation coefficient; The local confidence weight levels include four levels: high confidence, medium-high confidence, medium-low confidence, and low confidence. The comprehensive effectiveness of each node was calculated and summed based on the monitoring network configuration data, resulting in an effective monitoring capability index of 5.5 for the target grid points in the northern section of the Anninghe Fault. The ideal grid density for the region is determined to be 5 based on the comprehensive activity level of the fault segments. The calculated sparsity of the monitoring data is 1.1. Geological background data were obtained, with fault density of 0.82, lithological variation gradient of 0.38, topographic cutting depth of 650m, normalized value of 0.65, and distribution density of historical earthquake rupture zones of 0.55, normalized value of 0.55. The calculated heterogeneity index, after normalization, is 0.35, classifying the geological structure complexity as low. The preset minimum confidence threshold weight is 0.3, the maximum confidence threshold weight is 0.95, the regional baseline sparsity is 1, the regional baseline complexity is 0.65, the sparsity sensitivity coefficient is 2.2, the sparsity saturation index is 0.6, and the complexity decay coefficient is 1.6. The calculated local confidence weight is 0.602, which is mapped to a local confidence weight level of high confidence.

[0020] In this embodiment, the method for determining the seismogenic activity benchmark includes: Seismic activity parameters are extracted from each fault segment; these parameters include historical earthquake frequency, magnitude distribution parameters, and cumulative moment release rate. The seismic activity parameters are corrected based on the comprehensive activity level of fault segments to determine the background seismogenic activity level; the background seismogenic activity level characterizes the seismic intensity benchmark of each fault segment under long-term stress background. The seismogenic environment field is constructed by mapping the background seismogenic activity level and the fault segmentation of each grid point; the seismogenic environment field characterizes the static seismogenic hazard level of each grid point under the long-term seismogenic background. Dynamic anomaly information is extracted from the precursor correction data and mapped in combination with the fault spatial relationship of each node to construct a precursor response field; the precursor response field characterizes the dynamic seismic risk increment of each grid point under short-term precursor disturbances. Based on the local confidence weight level of each grid point, the earthquake-pregnancy environment field and the precursor response field are reorganized differentially to determine the earthquake-pregnancy activity benchmark. In actual assessment, the differentiated reorganization includes increasing the contribution ratio of the precursor response field to the earthquake-induced activity benchmark in the high-confidence region, and making the earthquake-induced activity benchmark closer to the background value of the earthquake-induced environmental field in the low-confidence region. The precursor response field value is obtained by normalizing the residual anomaly components with the historical maximum anomaly amplitude. The baseline expression for the seismogenic activity is: ; in, For the first The baseline for seismic activity at each grid point For the first The seismogenic environment field value at each grid point represents the long-term probability background of earthquakes, and the values ​​are taken as follows: , For the first The precursor response field value at each grid point represents the normalized short-term precursor response probability increment, and its value is [value missing]. , For the first Local confidence weights for each grid point The precursor response saturation coefficient is given by the following value: Controlling the maximum allocation ratio of precursor responses within the remaining probability space. The confidence level sensitivity coefficient; Seismic activity parameters were extracted from the northern segment of the Anninghe Fault. The historical earthquake frequency was 15 times per 100 years, the magnitude distribution parameter was 0.82, and the cumulative moment release rate was 1.2 × 10⁻⁶. 17 N∙m / century, after background correction, the background seismogenic activity level is 0.38; Based on the background seismogenic activity level and the fault segmentation mapping of each grid point, a seismogenic environment field is constructed, with a value of 0.38. Dynamic anomaly information was extracted from the precursor correction data, and a precursor response field was constructed by combining it with fault spatial relationship mapping. After normalizing the residual anomaly components with the historical maximum anomaly amplitude, the precursor response field value was 0.32. With a local confidence weight of 0.602, a precursor response saturation coefficient of 0.80, and a confidence sensitivity coefficient of 2.20, the calculated baseline for pre-seismic activity is 0.496.

[0021] In this embodiment, the method for obtaining the grid point seismic hazard probability includes: Extract the corrected feature values ​​of each precursor feature from the precursor correction data and calculate the dynamic anomaly comprehensive index; the dynamic anomaly comprehensive index is the grid-point dimensionless anomaly intensity obtained by aggregated the corrected feature values ​​of each precursor feature through root mean square. The modified channel opening coefficient is determined based on the local confidence weight level of each grid point; the modified channel opening coefficient is positively correlated with the local confidence weight level. The seismogenic activity benchmark is nonlinearly probabilistically modulated based on the dynamic anomaly comprehensive index and the modified channel opening coefficient; the nonlinear probabilistic modulation uses the seismogenic activity benchmark as the base term and the coupling term of the dynamic anomaly comprehensive index and the modified channel opening coefficient as the incremental driving term. The upper limit boundary is defined as the difference between the upper limit constraint value and the earthquake activity benchmark. The probability of earthquake occurrence at grid points is modulated based on the earthquake activity benchmark value to obtain the earthquake occurrence probability at grid points. The upper limit constraint value is not lower than the earthquake activity benchmark value. In practical assessments, the expression for the grid point seismic hazard probability is as follows: ; in, The probability of grid point seismic activity. As a benchmark for seismic activity, This is a probability upper limit constraint value, determined based on the upper limit of the recurrence probability of the largest historical earthquake in the region. It is a dynamic anomaly comprehensive index. To correct the channel openness coefficient, This is the abnormal sensitivity coefficient; The feature values ​​of each feature after correction are extracted from the precursor correction data, and the absolute values ​​are used to participate in the root mean square aggregation. The dynamic anomaly comprehensive index is calculated to be 1.6. Based on the historical maximum anomaly intensity, the value is 0.85 after grid-point dimensionless standardization. The corrected channel opening coefficient is determined to be 1 based on the local confidence level weighting. Based on the upper limit of the recurrence probability of the largest historical earthquake in the region, the upper limit of the probability constraint value is determined to be 0.93, the anomaly sensitivity coefficient is 1.50, and the calculated seismic hazard probability of grid point is 0.81.

[0022] In this embodiment, the method for obtaining earthquake risk prediction results includes: A time evolution analysis was performed on the seismic hazard probability of the grid points to calculate the hazard probability growth rate and cumulative hazard index; The criticality level is determined based on the rate of increase in the probability of danger and the cumulative risk index; the criticality level includes four levels: low critical, medium critical, high critical, and extremely high critical. Extract the parameters of the ground motion attenuation law; the parameters of the ground motion attenuation law include the peak ground acceleration attenuation coefficient and the spectral attenuation coefficient. Based on the seismic hazard probability and ground motion attenuation parameters of each grid point, the ground motion intensity of each grid point is predicted; the ground motion intensity includes peak ground acceleration and characteristic period. Extract the vulnerability parameters of the disaster-bearing body; the vulnerability parameters of the disaster-bearing body include the seismic resistance of the building structure, population exposure density and vulnerability of lifeline engineering. The loss rate of the disaster-bearing body is calculated based on the seismic intensity and the vulnerability parameters of the disaster-bearing body. The loss rate of the disaster-bearing body is calculated based on the log-normal vulnerability curve. The median parameter of the log-normal vulnerability curve is determined based on the seismic resistance of the building structure, and the coefficient of variation is determined based on the building structure type. A two-dimensional decision matrix is ​​constructed by combining criticality classification and disaster-bearing body loss rate; the two-dimensional decision matrix has criticality classification as the horizontal axis and disaster-bearing body loss rate as the vertical axis. Differentiated early warning strategies are determined based on a two-dimensional decision matrix, and earthquake risk prediction results are output. In actual assessments, the time window length for the time evolution analysis is determined based on the regional seismic activity cycle; The rate of increase of the danger probability is calculated using the first difference of the danger probability of earthquake occurrence at grid points within a time window. The cumulative risk index is the time integral of the rate of increase in risk probability within a time window. The criticality classification thresholds are determined based on retrospective analysis of historical earthquake cases in the region: Low criticality is defined as a cumulative hazard index less than 10 and a growth rate less than 0.002 / day; Medium criticality is defined as a cumulative hazard index less than 20 but not less than 10, or a growth rate less than 0.005 / day but not less than 0.002 / day; High criticality is defined as a cumulative hazard index less than 30 but not less than 20, or a growth rate less than 0.008 / day but not less than 0.005 / day; Extremely high criticality is defined as a cumulative hazard index not less than 30 and a growth rate not less than 0.008 / day. The expression for the loss rate of the disaster-bearing body is: ; in, Earthquake intensity The loss rate below The cumulative function of the standard normal distribution. The median seismic resistance capacity of the disaster-bearing structure is taken as the peak ground acceleration corresponding to the seismic fortification intensity of the building structure. The vulnerability dispersion coefficient is determined based on the building structure type. The differentiated early warning strategy includes Level 1 response, Level 2 response, Level 3 response, and routine monitoring; Level 1 response corresponds to extremely high criticality and high disaster-bearing body loss rate areas; Level 2 response corresponds to high criticality or high disaster-bearing body loss rate areas; Level 3 response corresponds to medium criticality and medium disaster-bearing body loss rate areas; and routine monitoring corresponds to low criticality areas. The earthquake risk prediction results include criticality classification, loss rate of disaster-bearing bodies, and differentiated early warning strategies.

[0023] A time evolution analysis of the seismic hazard probability at grid points was performed, and the time window length was determined to be 60 days based on the regional seismic activity cycle. The probability of an earthquake occurring at a grid point 60 days before the earthquake is 0.38, the probability of an earthquake occurring at a grid point 30 days before the earthquake is 0.50, and the probability of an earthquake occurring at a grid point 1 day before the earthquake is 0.81. The rate of increase in the probability of danger is calculated using forward difference with a time step of 1 day. The continuous rate of increase is calculated by linear interpolation to complete the daily data between the three discrete points 60 days, 30 days and 1 day before the earthquake. The cumulative risk index is calculated by numerical integration using the trapezoidal method. The rate of increase in the probability of danger, calculated using the first-order difference, is 0.0072 / day. The cumulative hazard index, calculated using time integration, is 35.7. Based on the rate of increase in the probability of danger and the cumulative danger index, the criticality level is determined to be extremely high criticality. The parameters of the ground motion attenuation law were extracted. The peak ground acceleration attenuation coefficient was 1.35 and the spectral attenuation coefficient was 0.85. The ground motion intensity of the target grid point was predicted based on the grid point earthquake hazard probability of 0.81. The peak ground acceleration was calculated to be 0.42g and the characteristic period was 0.45s by coupling the earthquake probability and the attenuation relationship. The vulnerability parameters of the disaster-bearing body were extracted. The median seismic resistance capacity of the building structure was taken as the peak ground acceleration corresponding to the seismic intensity of VIII, which is 0.25g. The building structure type was a frame structure, and the vulnerability dispersion coefficient was taken as 0.42.

[0024] The loss rate of the disaster-bearing body calculated based on the log-normal vulnerability curve is 0.891, which is a high loss rate. The quantitative boundaries of the dual-dimensional decision matrix are as follows: extremely high criticality and a loss rate of not less than 0.8 correspond to Level 1 response; high criticality or a loss rate of not less than 0.5 correspond to Level 2 response; medium criticality and a loss rate of not less than 0.5 or not less than 0.2 correspond to Level 3 response; low criticality or a loss rate of not less than 0.2 corresponds to conventional monitoring. Based on the comprehensive criticality classification and the loss rate of the disaster-bearing body, a two-dimensional decision matrix is ​​constructed to determine the differentiated early warning strategy for the target grid point as the first-level response, which includes personnel transfer, traffic control, shutdown of lifeline projects, and prepositioning of emergency supplies. The earthquake risk prediction result is extremely high, with a disaster-bearing body loss rate of 0.891, indicating a high loss rate. A Level I response is required.

[0025] Secondly, an earthquake risk prediction system based on multi-source data fusion includes: Causal inference module: used to acquire multi-source monitoring data and extended data, construct a precursor monitoring network, perform causal inference on earthquake risk characteristics with fault mechanical constitutive relations as physical constraints, and determine a subset of precursor features; Precursor correction module: used to perform three-level cross correction of spatial consistency, mechanical coupling and environmental interference on a subset of precursor features based on multi-source monitoring data, to obtain precursor correction data; Confidence assessment module: used to calculate the sparsity of monitoring data based on the monitoring network configuration data, calculate the complexity of geological structure based on the geological structure background data, and spatially couple the two to determine the local confidence weight; Earthquake Prediction Baseline Module: Used to determine the background earthquake prediction activity level based on historical earthquake catalog data, construct the earthquake prediction environment field and precursor response field, and determine the earthquake prediction activity baseline by combining local confidence weight nonlinear reorganization; The hazard probability module is used to extract dynamic anomaly information from the precursor correction data as a correction factor, use the local confidence weight as a modulation coefficient, modulate the earthquake activity benchmark, and obtain the grid point earthquake occurrence hazard probability. Early warning strategy module: used to classify the critical probability of earthquake occurrence at grid points, combine the ground motion attenuation law parameters and the vulnerability parameters of the disaster-bearing body, obtain earthquake risk prediction results and determine differentiated early warning strategies.

[0026] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A seismic risk prediction method based on multi-source data fusion, characterized in that, Includes the following steps: S1. Acquire multi-source monitoring data and extended data, construct a precursor monitoring network, extract earthquake risk characteristics, use fault mechanical constitutive relations as physical constraints to make causal inferences on earthquake risk characteristics, and determine a subset of precursor characteristics; S2. Based on the multi-source monitoring data, the precursor feature subset is subjected to a three-level cross correction of spatial consistency, mechanical coupling and environmental interference to obtain the precursor correction data; S3. Calculate the sparsity of monitoring data based on the monitoring network configuration data, calculate the geological structure complexity based on the geological structure background data, spatially couple the monitoring data sparsity and geological structure complexity, and determine the local confidence weight. S4. Determine the background seismogenic activity level based on historical earthquake catalog data, construct the seismogenic environment field and precursor response field, and differentiate and reorganize the seismogenic environment field and precursor response field by combining local confidence weights to determine the seismogenic activity benchmark. S5. Extract dynamic anomaly information from precursor correction data as correction factor, use local confidence weight as modulation coefficient, and perform nonlinear probability modulation on earthquake activity benchmark to obtain grid point earthquake hazard probability. S6. The critical classification of the seismic hazard probability of grid points is carried out. Combined with the seismic motion attenuation law parameters and the vulnerability parameters of the disaster-bearing body, the earthquake risk prediction results are obtained, and differentiated early warning strategies are determined. The monitoring data includes spatial deformation monitoring data, seismic activity monitoring data, underground fluid monitoring data, and geophysical field monitoring data; The extended data includes historical earthquake catalog data, monitoring network configuration data, ground motion attenuation parameters, geological structure background data, vulnerability parameters of disaster-bearing bodies, and environmental disturbance data.

2. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for determining the precursor feature subset includes: Acquire monitoring and extended data, extract active fault segments, and determine the overall activity of each fault segment; Based on the comprehensive activity of fault segments and the location of key structural components, physical monitoring nodes and virtual precursor nodes are deployed. The virtual precursor nodes are equivalent observation points constructed in the blind areas covered by physical points based on the continuity of geological structures and physical field gradient extrapolation. The historical background benchmark of the virtual precursor nodes is determined by physical field gradient interpolation using historical observation data from physical monitoring nodes in the spatial neighborhood. The current observation value of the virtual precursor nodes is determined by physical field gradient extrapolation using current observation data from physical monitoring nodes in the spatial neighborhood. Determine the fault spatial relationship of each node and construct a precursor monitoring network; the fault spatial relationship characterizes the fault segmentation of the node and its relative spatial position with respect to the fault. Historical background benchmarks are determined based on historical observation data from each entity monitoring node. Observational deviations are calculated, and exponential decay of distance from faults and logarithmic enhancement of observation depth are introduced to obtain seismic risk characteristic values. These seismic risk characteristics include deformation field characteristics, seismic activity field characteristics, underground fluid field characteristics, and geophysical field characteristics. Determine the stress state parameters and causal correlation parameters of each feature, use the fault mechanical constitutive relation as a physical constraint, make causal inferences on earthquake risk characteristics, and screen to obtain a candidate set of precursor features; A physical consistency test is performed on the candidate set of precursor features to determine the subset of precursor features; the subset of precursor features represents a set of earthquake-induced anomaly response information that has a mechanical causal relationship with the fault stress state and is self-consistent across physical fields.

3. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for obtaining precursor correction data includes: A spatial influence domain is defined, and each node in the subset of precursor features is extracted as a retrieval center. Similar precursor features are retrieved to obtain a neighborhood similar feature set; the similar precursor features are precursor features belonging to the same physical field type. Heterogeneous precursor features are extracted from the same fault segment to obtain a set of heterogeneous features in the same domain; the heterogeneous precursor features are precursor features belonging to different physical field types; Calculate the directional consistency ratio of similar feature sets in the neighborhood, and determine the anomalous properties of each feature based on a preset threshold and the fault spatial relationship of each node, and correct them accordingly; the anomalous properties include regional group anomalies, local structural anomalies, and non-structural disturbance anomalies. Based on the rock fracture mechanical response chain, a mechanical coupling test is performed on the heterogeneous feature sets in the same domain, and the mechanical response categories are classified and corrected accordingly. The rock fracture mechanical response chain characterizes the directional sequence relationship of different physical field responses of the fault during stress loading until instability and rupture, according to the mechanical coupling law. This includes the directional sequence of deceleration of cross-fault displacement rate, enhancement of surface strain compression, decrease of earth resistivity, and rise of groundwater level under compressive stress. The mechanical response categories include mechanical synergistic features and mechanical anisotropic features. Screening environmentally sensitive features based on the inherent sensitivity of each physical field type to environmental factors; Environmental disturbance data is extracted, an environmental response model is constructed, and the environmental driving components of environmentally sensitive features are separated to obtain residual anomaly components. The environmental response model characterizes the response function relationship between precursor feature observations and environmental disturbance data. By combining the results of anomaly property correction, mechanical coupling correction, and environmental interference stripping, precursor correction data are obtained.

4. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for determining local confidence weights includes: The effective monitoring capability index is calculated based on the monitoring network configuration data; the effective monitoring capability index is the number of equivalent nodes within the effective monitoring range of each grid point. Determine the ideal network density for the region based on the comprehensive activity level of fault segments, and calculate the sparsity of monitoring data. Geological structural background data is acquired, and a structural heterogeneity index is calculated; the structural heterogeneity index reflects the degree of spatial difference in the physical properties of the geological medium within the monitoring target area. The complexity of geological structures is determined based on the structural heterogeneity index; the complexity of geological structures is divided into three levels according to the structural heterogeneity index. A two-dimensional coupled modulation function is constructed using the sparsity of monitoring data as the supply term and the complexity of geological structure as the loss term. The local confidence weight is calculated and mapped to the local confidence weight level. The local confidence weight is positively correlated with the sparsity of monitoring data and negatively correlated with the complexity of geological structure.

5. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for determining the baseline for seismic activity includes: Seismic activity parameters are extracted from each fault segment; these parameters include historical earthquake frequency, magnitude distribution parameters, and cumulative moment release rate. The seismic activity parameters are corrected for background based on the comprehensive activity level of fault segments to determine the background seismogenic activity level; the background seismogenic activity level characterizes the seismic intensity benchmark of each fault segment under long-term stress background. The seismogenic environment field is constructed by mapping the background seismogenic activity level and the fault segmentation of each grid point; the seismogenic environment field characterizes the static seismogenic hazard level of each grid point under the long-term seismogenic background. Dynamic anomaly information is extracted from the precursor correction data and mapped in combination with the fault spatial relationship of each node to construct a precursor response field; the precursor response field characterizes the dynamic seismogenic risk increment of each grid point under short-term precursor disturbances. Based on the local confidence weight level of each grid point, the earthquake-generating environment field and the precursor response field are reorganized differentially to determine the baseline of earthquake-generating activity.

6. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for obtaining the grid point seismic hazard probability includes: Extract the corrected feature values ​​of each precursor feature from the precursor correction data and calculate the dynamic anomaly comprehensive index; the dynamic anomaly comprehensive index is the grid-point dimensionless anomaly intensity obtained by aggregating the corrected feature values ​​of each precursor feature through root mean square. The modified channel opening coefficient is determined based on the local confidence weight level of each grid point; the modified channel opening coefficient is positively correlated with the local confidence weight level. The seismogenic activity benchmark is nonlinearly probabilistically modulated based on the dynamic anomaly comprehensive index and the modified channel opening coefficient; the nonlinear probabilistic modulation uses the seismogenic activity benchmark as the base term and the coupling term of the dynamic anomaly comprehensive index and the modified channel opening coefficient as the incremental driving term. The upper limit boundary is defined as the difference between the upper limit constraint value and the earthquake activity benchmark. The probability of earthquake occurrence at grid points is modulated based on the earthquake activity benchmark value to obtain the earthquake occurrence probability at grid points. The upper limit constraint value is not lower than the earthquake activity benchmark value.

7. The earthquake risk prediction method based on multi-source data fusion according to claim 1, characterized in that, The method for obtaining earthquake risk prediction results includes: A time evolution analysis was performed on the seismic hazard probability of the grid points to calculate the hazard probability growth rate and cumulative hazard index; The criticality level is determined based on the rate of increase in the probability of danger and the cumulative risk index; the criticality level includes four levels: low critical, medium critical, high critical, and extremely high critical. Extract the parameters of the ground motion attenuation law; the parameters of the ground motion attenuation law include the peak ground acceleration attenuation coefficient and the spectral attenuation coefficient. Based on the seismic hazard probability and ground motion attenuation parameters of each grid point, the ground motion intensity of each grid point is predicted; the ground motion intensity includes peak ground acceleration and characteristic period. Extract the vulnerability parameters of the disaster-bearing body; the vulnerability parameters of the disaster-bearing body include the seismic resistance of the building structure, population exposure density and vulnerability of lifeline engineering. The loss rate of the disaster-bearing body is calculated based on the seismic intensity and the vulnerability parameters of the disaster-bearing body. The loss rate of the disaster-bearing body is calculated based on the log-normal vulnerability curve. The median parameter of the log-normal vulnerability curve is determined based on the seismic resistance of the building structure, and the coefficient of variation is determined based on the building structure type. A two-dimensional decision matrix is ​​constructed by combining criticality classification and disaster-bearing body loss rate; the two-dimensional decision matrix has criticality classification as the horizontal axis and disaster-bearing body loss rate as the vertical axis. Differentiated early warning strategies are determined based on a two-dimensional decision matrix, and earthquake risk prediction results are output.

8. A seismic risk prediction system based on multi-source data fusion, used to perform the method according to any one of claims 1-7, characterized in that, include: Causal inference module: used to acquire multi-source monitoring data and extended data, construct a precursor monitoring network, perform causal inference on earthquake risk characteristics with fault mechanical constitutive relations as physical constraints, and determine a subset of precursor features; Precursor correction module: used to perform three-level cross correction of spatial consistency, mechanical coupling and environmental interference on a subset of precursor features based on multi-source monitoring data, to obtain precursor correction data; Confidence assessment module: used to calculate the sparsity of monitoring data based on the monitoring network configuration data, calculate the complexity of geological structure based on the geological structure background data, and spatially couple the two to determine the local confidence weight; Earthquake Prediction Baseline Module: Used to determine the background earthquake prediction activity level based on historical earthquake catalog data, construct the earthquake prediction environment field and precursor response field, and determine the earthquake prediction activity baseline by combining local confidence weight nonlinear reorganization; The hazard probability module is used to extract dynamic anomaly information from the precursor correction data as a correction factor, use the local confidence weight as a modulation coefficient, modulate the earthquake activity benchmark, and obtain the grid point earthquake occurrence hazard probability. Early warning strategy module: used to classify the critical probability of earthquake occurrence at grid points, combine ground motion attenuation law parameters and disaster-bearing body vulnerability parameters to obtain earthquake risk prediction results and determine differentiated early warning strategies.

Images