Alarm method and system based on multi-source seismic motion space-time consistency discrimination

Abstract

The application discloses an alarm method and system based on multi-source seismic motion space-time consistency discrimination, and belongs to the technical field of earthquake monitoring and disaster warning. The method obtains multi-node vibration, position, time scale and state data, combines a pickup uncertainty interval and a residual clock difference interval into a correction to time interval, generates a propagation evidence edge associated with a compatible seismic source position constraint when a theoretical and observed time difference interval intersects, and then calculates a connected consistent subgraph of a cross-space neighborhood and a candidate seismic source feasible region. Through four-state evidence classification of an expected node, negative evidence constraint and node-by-node stability discrimination, an earthquake alarm, local interference or insufficient evidence result is output, so as to reduce local vibration false positives, inhibit offline node false negatives and single node domination.

Description

Technical Field

[0001] This invention relates to the field of earthquake monitoring and disaster early warning, and in particular to an alarm method and system based on the spatiotemporal consistency of multi-source ground motion. Background Technology

[0002] Seismic ground motion monitoring networks typically consist of vibration monitoring nodes distributed in different locations. Each node continuously collects vibration signals and triggers event detection based on amplitude, energy, or frequency band characteristics. In existing alarm methods, some schemes still rely primarily on the vibration amplitude of a single node or a preset threshold as the main basis for judgment. However, local vibrations such as vehicle traffic, mechanical construction, blasting, and equipment collisions, due to their sudden energy, may trigger alarms, making it difficult to distinguish between single-node amplitudes and actual earthquake propagation events.

[0003] On the other hand, different monitoring nodes may have sampling clock deviations, aging synchronous records, communication frame drops, data loss, and differences in installation and transmission characteristics, making it difficult to directly compare the arrival time, waveform morphology, and frequency band energy of the same seismic wave at different nodes. If offline nodes, nodes with missing data, or nodes with abnormal self-tests are all taken as evidence of undetected events, it may incorrectly deny real events; if a small number of nodes have a dominant effect on the fusion results, it may lead to unstable alarms.

[0004] Therefore, there is a need for an alarm method and system based on spatiotemporal consistency discrimination of multi-source ground motion that can overcome the shortcomings of the existing technology. Summary of the Invention

[0005] One objective of this invention is to propose an alarm method and system based on spatiotemporal consistency discrimination of multi-source ground motion. Addressing the problems of existing technologies that rely on single-node thresholds, difficulty in directly comparing the time and installation conditions of different nodes, the misuse of missing node states as counter-evidence, and the potential dominance of single nodes in the fusion results, this invention proposes a technical solution based on time interval correction, intersection gating of theoretical and observed propagation time difference intervals, propagation consistency evidence maps, feasible candidate source regions, expected four-state evidence of nodes, and node-by-node stability checks. This allows for the differentiation between local high-energy vibrations and events with spatial propagation continuity under unified spatiotemporal constraints, and generates interpretable three-state alarm results when some nodes are offline, experience packet loss, or exhibit anomalies.

[0006] This invention provides an alarm method based on the spatiotemporal consistency discrimination of multi-source ground motion, including:

[0007] S1. Obtain the vibration sequence, node coordinates and sampling time scale of multiple ground motion monitoring nodes, perform inter-node comparability processing on the vibration sequence, extract candidate first arrival and vibration features, determine the residual clock difference interval based on the node's time synchronization information and clock drift information, and synthesize the node correction to the time interval by combining the candidate first arrival picking uncertainty interval with the residual clock difference interval.

[0008] S2. For node pairs and candidate source locations, determine the theoretical propagation time difference interval based on node coordinates and propagation velocity range, determine the observation time difference interval based on the node correction to time interval of the node pair, and when the theoretical propagation time difference interval intersects with the observation time difference interval, generate propagation evidence edges that are compatible with candidate source location constraints based on vibration characteristics within the intersection range.

[0009] S3. Construct a propagation consistency evidence map based on the ground motion monitoring nodes and the propagation evidence edges. Determine a connected and consistent subgraph from the propagation consistency evidence map that is explained by the feasible domain of the same candidate source and covers multiple spatial neighborhoods. Determine local interference clusters based on the propagation evidence edges that cannot be incorporated into the connected and consistent subgraph.

[0010] S4. Determine the event credibility based on the spatial coverage of the connected and consistent subgraph, the feasible domain of the candidate seismic source, and the consistency of the propagation evidence edges, and output the earthquake alarm, local interference, or insufficient evidence result based on the event credibility and the local interference cluster.

[0011] Optionally, S1 includes: acquiring synchronization records, synchronization age, clock drift upper limit, frame loss records, and installation parameters of various vibration monitoring nodes;

[0012] The residual clock error interval is determined based on the product of the synchronization deviation, synchronization age, and the upper bound of the clock drift, as well as the time stamp gap corresponding to the lost frame.

[0013] Frequency band division and normalization benchmark are determined based on the noise power spectrum during non-event periods. The vibration sequence is rectified according to the installation transfer baseline corresponding to the installation parameters. The envelope change position is extracted in each frequency band as a candidate first arrival. The picking uncertainty interval is determined according to the signal-to-noise ratio distribution in the neighborhood of the candidate first arrival. The normalized waveform segment, frequency band energy sequence and envelope evolution sequence based on the candidate first arrival are extracted as the vibration features.

[0014] Optionally, S2 includes: discretizing the preset monitoring area to obtain a set of candidate earthquake source locations;

[0015] For each candidate earthquake source location in the candidate earthquake source location set, the theoretical propagation time difference interval is determined based on the path length from the candidate earthquake source location to the two nodes in the node pair, the lower and upper bounds of the regional propagation velocity, and the site travel time correction for each node.

[0016] The observation time difference interval is obtained by subtracting the time intervals between the node corrections of the two nodes in the node pair to the time intervals.

[0017] When two time difference intervals intersect, the candidate source location constraints of the intersecting time difference intervals will be associated with the corresponding propagation evidence edges. The normalized waveform correlation coefficient, frequency band energy ratio consistency, and envelope evolution consistency will be calculated only within the time shift corresponding to the intersection range. The edge weights and conflict markers will be determined based on the comparison results of the three with their respective discrimination thresholds. The propagation direction will be determined based on the order of observation arrival.

[0018] When two time difference intervals do not intersect, compatible propagation evidence edges are not generated between the corresponding node pairs.

[0019] Optionally, S3 includes: using ground motion monitoring nodes as vertices and propagation evidence edges as directed edges;

[0020] The feasible region of candidate seismic sources in the subgraph is obtained by intersecting the compatible candidate seismic source location constraints associated with the propagation evidence edges. The conditions for subgraph expansion are that the feasible region of candidate seismic sources is not empty, the node arrival order is consistent with the direction of the directed edge, and the total weight of the propagation evidence edges with conflict markers does not exceed the conflict threshold.

[0021] The connected and consistent subgraphs are selected based on the number of spatial neighbors to which a node belongs and the geometric dispersion between nodes;

[0022] The set of propagation evidence edges that have vibrational energy but do not meet the subgraph expansion condition and whose number of covered spatial neighbors is less than the spatial coverage threshold is determined as the local interference cluster.

[0023] Optionally, S4 includes: determining the credibility of the underlying event based on a weighted combination of the spatial coverage of the connected consistent subgraph, the geometric dispersion of nodes, the volume of the feasible domain of the candidate seismic source, the interval residual between the observation time difference interval and the theoretical propagation time difference interval, the normalized waveform correlation coefficient, the frequency band energy ratio consistency, and the envelope evolution consistency.

[0024] When the credibility of the basic event reaches the credibility alarm threshold, an earthquake alarm is output.

[0025] When the credibility of the basic event does not reach the credibility alarm threshold, and the local interference cluster exists and the total weight of its propagation evidence edge reaches the interference discrimination threshold, local interference is output;

[0026] In other cases, the output will be "insufficient evidence".

[0027] Furthermore, based on the candidate source locations in the feasible domain of the candidate sources, the event energy scale determined by the frequency band energy of the participating nodes, the regional attenuation envelope, the site response baseline and noise floor of each node, the expected signal amplitude and expected arrival time interval of each node are determined, and nodes whose expected signal amplitude reaches the corresponding node's detectable threshold are included in the expected node set.

[0028] Obtain the communication status, self-test status, and data integrity rate of the expected nodes. Determine the expected nodes that are offline or have abnormal self-test status as unusable evidence nodes. Determine the nodes whose data integrity rate does not reach the integrity rate threshold as unknown evidence nodes. Determine the nodes that form compatible propagation evidence edges as positive evidence nodes and the nodes that do not form compatible propagation evidence edges as negative evidence nodes.

[0029] The expected arrival time interval and detectable threshold of the negative evidence node are used to exclude candidate source locations that contradict their lack of response. The credibility of the basic event is updated based on the updated candidate source feasible region and various evidence nodes to obtain the updated event credibility. The updated event credibility is then used as the event credibility for result discrimination in step S4.

[0030] The ratio of the sum of the number of positive evidence nodes and negative evidence nodes to the total number of expected nodes is determined as the available node coverage rate. Unknown evidence nodes and unavailable evidence nodes are not considered as negative evidence nodes and are not included in the numerator of the available node coverage rate.

[0031] Furthermore, the expected node set is generated for each candidate source location in the feasible domain of the candidate source, and the negative evidence weight is determined based on the magnitude by which the expected signal amplitude of each negative evidence node exceeds the detectable threshold for that candidate source location, the data integrity rate within the expected arrival time interval, and the noise floor of the corresponding node.

[0032] Delete candidate earthquake source locations whose negative evidence weight reaches the exclusion threshold, reduce the credibility of the remaining candidate earthquake source locations according to the negative evidence weight, and update the event result to insufficient evidence when the feasible region of the candidate earthquake source is empty.

[0033] Furthermore, before outputting the earthquake alarm, participating nodes in the connected and consistent subgraph are temporarily removed sequentially. For each removal, the compatible propagation evidence edges and various types of evidence nodes are re-determined, and the connected and consistent subgraph and the candidate source feasible region are reconstructed. Based on the reconstruction results, the credibility of the updated event is recalculated to obtain the credibility of the reconstructed event.

[0034] The credibility of each reconstruction event is sorted in ascending order to obtain a credibility sequence. The volume change rate of the candidate source feasible region before and after each reconstruction is sorted in descending order to obtain a change rate sequence. The most unfavorable stability index is determined based on the first term of the credibility sequence and the first term of the change rate sequence. The node dominance index is determined based on the total change rate of the propagation evidence edge weights before and after the removal of a single participating node, the change in the credibility of the update event, and the volume change rate of the candidate source feasible region.

[0035] Furthermore, when the credibility of the update event reaches the credibility alarm threshold, the first item of the credibility sequence sorting corresponding to the most unfavorable stability index reaches the stability alarm threshold and the first item of the change rate sequence sorting does not exceed the change threshold, and the node dominance index is less than the dominance threshold, an earthquake alarm is output.

[0036] Otherwise, the earthquake alarm will be prohibited from being output, and the result of local interference or insufficient evidence will be output based on the existence of local interference clusters and the coverage of the available nodes.

[0037] Participating nodes whose dominance index reaches the dominance threshold are identified as dominating nodes, and endpoint nodes of propagation evidence edges with conflict markers are identified as conflict nodes. The output results include candidate source regions, propagation directions, participating nodes, conflict nodes, dominating nodes, and reconstruction results that lead to downgrade.

[0038] On the other hand, the present invention also provides an alarm system based on spatiotemporal consistency discrimination of multi-source ground motion, including:

[0039] The system comprises the following modules: a multi-node correction and feature generation module, which acquires data from multiple ground motion monitoring nodes, performs comparability processing between nodes, and generates node corrections to time intervals and vibration characteristics; a propagation evidence generation module, which determines the theoretical propagation time difference interval and the observation time difference interval, and generates propagation evidence edges associated with compatible candidate source location constraints; a consistency graph analysis module, which constructs a propagation consistency evidence graph, determines a connected consistency subgraph, a feasible region of candidate sources, and local interference clusters; a basic credibility and alarm module, which determines the credibility of basic events and outputs earthquake alarms, local interference, or insufficient evidence results; a state credibility fusion module, which generates a desired node set, classifies various types of evidence nodes, and uses negative evidence nodes to constrain the feasible region of candidate sources and update the credibility of basic events; and a stability verification module, which temporarily removes participating nodes one by one, reconstructs a connected consistency subgraph, a feasible region of candidate sources, and event credibility, determines the most unfavorable stability index and node dominance index, and limits alarm output accordingly.

[0040] The beneficial effects of this invention are:

[0041] 1. By combining the picking uncertainty interval of candidate first arrivals with the residual clock difference interval into the node correction arrival interval, and calculating propagation evidence only when the theoretical propagation time difference interval intersects with the observed time difference interval, the decidability of multi-node arrival comparison can be maintained in the presence of clock drift and communication gaps.

[0042] 2. By constructing a propagation consistency evidence graph of associated candidate source location constraints and extracting a connected consistency subgraph that is explained by the feasible domain of the same candidate source and spans multiple spatial neighborhoods, local high-energy edge sets that cannot form a continuous spatial propagation relationship can be identified as local interference clusters.

[0043] 3. By dividing the expected nodes into positive, negative, unknown, and unusable evidence nodes, and using only negative evidence that meets the conditions of online, self-checking, and data integrity to constrain the feasible domain of candidate seismic sources, combined with the most unfavorable stability index and node dominance index obtained by temporary node-by-node elimination, it is possible to reduce false denials caused by missing data or node anomalies and suppress false alarms caused by single node dominance. Attached Figure Description

[0044] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0045] Figure 1 This is a flowchart of the alarm method based on the spatiotemporal consistency judgment of multi-source ground motion according to the present invention.

[0046] Figure 2 This is a flowchart of the process for generating propagation evidence edges based on time difference interval intersection gating in the present invention S2. Detailed Implementation

[0047] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0048] refer to Figures 1-2 An alarm method based on the spatiotemporal consistency of multi-source ground motion includes:

[0049] S1. Obtain the vibration sequence, node coordinates and sampling time scale of multiple ground motion monitoring nodes, perform inter-node comparability processing on the vibration sequence, extract candidate first arrival and vibration features, determine the residual clock difference interval based on the node's time synchronization information and clock drift information, and synthesize the node correction to the time interval by combining the candidate first arrival picking uncertainty interval with the residual clock difference interval.

[0050] S2. For node pairs and candidate source locations, determine the theoretical propagation time difference interval based on node coordinates and propagation velocity range, determine the observation time difference interval based on the node correction to time interval of the node pair, and when the theoretical propagation time difference interval intersects with the observation time difference interval, generate propagation evidence edges that are compatible with candidate source location constraints based on vibration characteristics within the intersection range.

[0051] S3. Construct a propagation consistency evidence map based on the ground motion monitoring nodes and the propagation evidence edges. Determine a connected and consistent subgraph from the propagation consistency evidence map that is explained by the feasible domain of the same candidate source and covers multiple spatial neighborhoods. Determine local interference clusters based on the propagation evidence edges that cannot be incorporated into the connected and consistent subgraph.

[0052] S4. Determine the event credibility based on the spatial coverage of the connected and consistent subgraph, the feasible domain of the candidate seismic source, and the consistency of the propagation evidence edges, and output the earthquake alarm, local interference, or insufficient evidence result based on the event credibility and the local interference cluster.

[0053] In this specific embodiment, S1 includes:

[0054] This specific implementation method is executed by twelve ground motion monitoring nodes and one edge alarm server deployed in a preset monitoring area. Each node is equipped with a three-component accelerometer, a temperature-controlled crystal oscillator, a satellite time receiver, and a status acquisition unit. It continuously generates original vibration frames at a sampling frequency of 200 times per second, which include node number, sampling sequence number, node local time stamp, three-axis acceleration value, synchronization batch number, frame loss bitmap, self-test code, and installation parameter version number. The edge alarm server receives the original vibration frames with an analysis frame length of 20 milliseconds and writes them into a circular buffer with a length of 60 seconds according to the node number and sampling sequence number. When the sampling sequence number is not continuous, the frame loss record is registered with the start and end timestamp of the gap without interpolation to forge vibration samples.

[0055] The multi-node correction and feature generation module reads the synchronization deviation, synchronization completion time, and drift upper bound from the most recent valid synchronization record of each node, and then... Determine the residual clock bias half-width, where the sign is... Indicates the first The residual clock half-width of each node, in milliseconds, is denoted by [symbol missing]. Represents the node index, with values ​​from one to twelve, symbol Indicates the first The synchronization deviation recorded in the most recent synchronization of each node, in milliseconds, with the sign... This indicates the synchronization age of the node, in seconds, with the symbol... This represents the upper bound of the clock drift obtained by calibrating the node using seven consecutive days of synchronization logs, in milliseconds per second. The symbol is... The cumulative time stamp gap corresponding to the frame loss within the analysis window is represented in milliseconds. The residual clock difference interval is recorded accordingly as a closed interval centered at zero, with the upper and lower bounds being the negative and positive values ​​of the residual clock difference half-width, respectively. When the synchronization record is missing or the synchronization age exceeds 3,600 seconds, the node is marked as having an abnormal time state and is stopped from entering the candidate first arrival association in this round.

[0056] Before deployment, a noise baseline is established using at least 30 minutes of event-free data near the node installation location. The module removes DC from each three-component sequence and applies a zero-phase fourth-order Butterworth filter, dividing it into three frequency bands: 0.5 to 2 Hz, 2 to 8 Hz, and 8 to 20 Hz. For each frequency band, the median, absolute median difference, and 95th percentile value of the noise power spectrum are saved. The baseline record uses the node number, installation parameter version number, frequency band number, and effective time as the joint key. The record is reconstructed when the node is reinstalled, the sensor gain change is confirmed to exceed 5% after self-testing, or the noise median shift exceeds three times the absolute median difference for 24 consecutive hours.

[0057] The installation parameter record saves the sensor sensitivity, three-axis direction matrix, mounting base type, and installation transmission baseline for each frequency band. The module first converts the original count value into meters per second squared according to the sensor sensitivity, then projects it onto the unified east-west, north-south, and vertical coordinates using the direction matrix, and completes the response correction in each frequency band by dividing the measured amplitude-frequency response by the corresponding installation transmission baseline. Frequency points with a denominator lower than the deployment calibration lower limit are set as invalid frequency points and do not participate in subsequent energy accumulation. The corrected samples are written into the comparable vibration buffer according to the event candidate number, node number, unified frame sequence number, and frequency band number.

[0058] The module performs a Hilbert transform on the correction sequence of each frequency band to obtain the envelope. The square root of the 95th percentile of the power spectrum of the event-free noise is used as the amplitude benchmark for that frequency band. Positions where the mean of the envelope of three consecutive analysis frames exceeds four times the benchmark and the first-order difference of the envelope of the first frame is positive are registered as candidate mutation positions. All candidate mutation positions of the three frequency bands are then combined into a candidate set. Adjacent positions are examined in ascending order of time scale. The first group of positions with a time difference of no more than 120 milliseconds and at least two frequency bands appearing together are output as a common mutation group. The earliest position of this group is written into the candidate first arrival field. Isolated positions that do not form a common mutation of multiple frequency bands are only retained as noise diagnosis records.

[0059] For each candidate's initial arrival, the module calculates the signal-to-noise ratio within the range of the first two seconds and the last two seconds. , where the symbol Indicates the first The signal-to-noise ratio of each candidate node's initial arrival neighborhood, in decibels, with the sign... This represents the sum of the energy of the three frequency band correction signals within 0.5 seconds after the candidate arrives, expressed in meters squared per fourth power second, with the symbol [symbol missing]. It represents the sum of the three-band noise energy calculated based on the same number of samples within one second before the candidate's first arrival. The unit is the same as the signal energy. When the noise energy is lower than the minimum energy of the same unit in the baseline record, the minimum energy is used as the denominator. Candidates with a signal-to-noise ratio lower than six dB are marked as low-confidence picks and do not participate in the generation of evidence for this round of propagation.

[0060] The uncertain half-width of the pickup is pre-calibrated by manually verified event samples of the same type of node. The calibration record saves the 95th percentile value of the first arrival absolute error according to the signal-to-noise ratio range. In this specific embodiment, 6 to 10 decibels, 10 to 20 decibels and not less than 20 decibels correspond to 80 milliseconds, 40 milliseconds and 20 milliseconds respectively. The module finds the half-width according to the candidate first arrival signal-to-noise ratio and forms the uncertain pickup range. If the signal-to-noise ratio falls on the record boundary, the maximum value of the half-width on both sides of the boundary is taken. The calibration record is updated with the sensor model and sampling rate as keys and after adding 100 manual verification events.

[0061] Module by Synthetic node correction to time interval, where, symbol Indicates the first The node correction of each node to the time interval, in milliseconds, is indicated by the symbol. This represents the center time of the candidate node's initial arrival after conversion to the server's unified time base, in milliseconds, with the symbol... This represents the pickup uncertainty half-width obtained based on the signal-to-noise ratio distribution, in milliseconds, with the symbol... The formula represents the aforementioned residual clock error half-width. This formula is equivalent to performing a Minkowski sum operation on the picking uncertainty interval and the residual clock error interval, thus preserving the upper and lower bounds of the two types of uncertainties.

[0062] The edge alarm server writes the initial arrival records of each node into the queue to be associated according to the lower bound of the corrected time interval. The maximum propagation delay obtained by dividing the maximum path length between any two nodes in the monitoring area by the lower bound of the area speed, plus 200 milliseconds, is used as the rolling association window. The record at the head of the queue triggers a new event candidate and generates an event candidate number consisting of the server start time, an incrementing sequence number, and a check bit. Other node records that intersect with the association window in the corrected time interval are merged into the same candidate. When there are multiple records for the same node, only the one with the highest signal-to-noise ratio is merged, and the remaining records are reserved for the next candidate. If a new record falls into two candidate windows at the same time, it is assigned to the one with the smallest absolute distance from the center of the corrected time interval to the center of each candidate. If the distances are the same, it is assigned to the one with the earlier number. The candidate is closed after no new records are added for a consecutive maximum propagation delay. Candidates with fewer than two valid nodes expire and are only written into the noise diagnosis record. Two closed candidates are no longer merged, thus forming a list of candidate records with unique event candidate numbers.

[0063] The module extracts the three-component corrected waveform from 0.5 seconds before to 2.5 seconds after the candidate arrival time of each node in the candidate record list. Each component is divided by its window root mean square and limited to -6 to +6. The energy proportion sequence and envelope median sequence of the three frequency bands are calculated in 20-millisecond frames to form a vibration feature record. The record includes the event candidate number, node number, correction to time interval, normalized waveform segment, frequency band energy sequence, envelope evolution sequence, signal-to-noise ratio, baseline version, and installation parameter version. When the completeness rate of any effective frequency band sample is less than 80%, the feature record is set to the data incomplete state and is only used by S4 to divide unknown evidence nodes. Other feature records are submitted to S2. The candidate record list is the only input for S2 to read the node records of this round of events.

[0064] In this specific embodiment, S2 includes:

[0065] The propagation evidence generation module reads the valid vibration feature records and node coordinates output by S1, discretizes the preset monitoring area on the horizontal plane at 1 km intervals, and discretizes it at 2 km intervals in the range of 0 to 30 km below the ground surface, to obtain a set of candidate source locations with grid number, three-dimensional coordinates, stratigraphic partition number and valid marker. Grids located outside the boundary of the monitoring area or falling into pre-registered invalid stratigraphic units are not included in the candidate set. The set version is bound to the regional velocity model version for storage.

[0066] For candidate epicenter locations and nodes in the node pair ,node The module reads the regional velocity model and site travel time correction records obtained from historical event calibration, calculates the path length through layered straight-line ray tracing, and then... Forming a theoretical propagation time difference interval, where the symbol Indicates the location of the candidate epicenter Pair of nodes The theoretical propagation time difference interval, in seconds, symbol Represents the candidate source grid index, symbol and symbols Represents two distinct node indices, symbol ,symbol These represent the ray path lengths from the candidate epicenter location to the two nodes, in kilometers, with the symbol [symbol missing]. ,symbol These represent the lower and upper bounds of the effective propagation speed given by the regional velocity model, respectively, in kilometers per second, with the symbol... ,symbol These represent the site travel time corrections for the two nodes as defined by the stratigraphic zoning, in seconds. The quotient of the path length divided by the propagation speed is converted to seconds and then added to or subtracted from the site travel time correction. If the ray path does not satisfy the effective layer constraints, the grid is marked as infeasible and deleted.

[0067] The upper and lower bounds of the regional velocity model are obtained by travel time inversion of regional seismic events that have been manually confirmed in the last three years. The lower bound, upper bound, sample number and version number are stored in each stratigraphic partition. In this specific implementation, the 5% and 95% quantile values ​​are used as velocity limits. When a ray path passes through multiple partitions, the path length in each partition is divided by the corresponding velocity limit and then summed. If the path corresponding to the grid passes through an unlabeled partition, the grid does not generate a theoretical interval for the current node pair and registers the model gap.

[0068] The site travel time correction is obtained by subtracting the travel time calculated by the velocity model according to the propagation formula from the measured travel time of historical confirmed regional events. It is divided into eight azimuth sectors according to the node number and the main propagation incident azimuth. The median of the residuals of each sector is taken as the calibration coefficient, and the absolute median of the residuals and the number of valid samples are saved. When the number of samples is less than 20, the median of the residuals of all azimuths of the node is used. The calibration is recalibrated when the node migrates or the velocity model version is updated. Sectors with an absolute median of residuals exceeding 0.5 seconds are marked as not meeting the usage constraints and deleted. S2 saves the azimuth sector used and the correction version in the edge record so that the same parameters can be reused when S4 is reconstructed.

[0069] The module performs interval subtraction on the correction intervals of the two nodes in S1, according to... The observation time difference interval is obtained, where the sign is... Represents a node relative nodes The observation time difference interval, in seconds, symbol and symbols Representing nodes respectively The correction is applied to the lower and upper bounds of the time interval, in seconds, with the symbol... and symbols Representing nodes respectively The correction is applied to the lower and upper bounds of the time interval, in seconds. All boundaries are converted from milliseconds to seconds before calculation.

[0070] For each candidate source grid, the module calculates the intersection of the theoretical propagation time difference interval and the observation time difference interval. Only when the intersection is not empty is the grid number written into the compatible grid bitmap of the node pair, and an allowable time shift mapping is established in the edge record with the candidate grid number as the key and the lower and upper bounds of the intersection as the values. When the intersection is empty, the grid is not written into the compatible bitmap. When the compatible grid bitmap of a node pair is empty, the propagation evidence edge of the node pair is not generated. The intersection intervals of different grids must not be merged into a single allowable time shift range across grids.

[0071] For each candidate grid in the allowed time-shift map, the module translates a normalized waveform segment of two nodes within the allowed time-shift interval corresponding to that grid, with a step size of 20 milliseconds. It then calculates the mean-free normalized cross-correlation of the overlapping samples after splicing the three components, and selects the time shift with the largest absolute value as the observation time shift for that grid. The selected correlation coefficient is denoted as... ,symbol Represents node pairs In the candidate grid The maximum normalized waveform correlation coefficient within the allowed time shift interval is a dimensionless quantity, ranging from -1 to 1. The larger the correlation coefficient, the stronger the waveform in-phase consistency. When the overlapping samples are less than 70% of the complete window, this item of the grid is set to invalid instead of being replaced by a zero value. Subsequently, the energy ratio consistency, envelope evolution consistency, edge weights, and conflict markers are all calculated according to the observation time shift corresponding to the same grid and written into the grid feature map.

[0072] Module by Calculate the frequency band energy ratio consistency, where the symbol Represents node pairs In the candidate grid The frequency band energy ratio uniformity under corresponding observation time shift is a dimensionless quantity ranging from zero to one, with a larger value indicating stronger uniformity. The sign is... Represents the candidate source grid index, symbol Indicates the index of three frequency bands, symbol ,symbol These represent two nodes arranged in a grid. After the observation time shift is aligned, the first The energy percentage of each frequency band within the characteristic window is a dimensionless quantity ranging from zero to one, with the sign... This represents a smoothing quantity to prevent the energy percentage from being zero, and in this specific implementation, it is taken as 10 to the power of negative 6, which is a dimensionless quantity.

[0073] The module processes the envelope evolution sequence of two nodes according to the candidate grid. The selected observations are time-shifted and aligned. The cosine similarity of the first-order differences of the sequences is calculated and linearly mapped to zero to one as the envelope evolution consistency of the grid. ,symbol Represents node pairs In the candidate grid The envelope evolution consistency under the corresponding observation time shift is a dimensionless quantity ranging from zero to one. If the first difference norm of any sequence is less than the minimum norm specified by the noise baseline, this item is set to invalid. The threshold configuration record saves the waveform correlation threshold of 0.65, the frequency band energy ratio consistency threshold of 0.70, and the envelope evolution consistency threshold of 0.60. The thresholds are selected from the validation set of confirmed earthquake events and vehicle, construction, and blasting interference samples with a false alarm rate not exceeding one percent, and are updated as fifty new events are added to the validation set.

[0074] Module by Determine the edge weights of the propagation evidence, where the symbol... Represents node pairs In the candidate grid The propagation evidence edge weights are dimensionless quantities ranging from zero to one, with larger values ​​indicating greater consistency of evidence. The sign is... Indicates candidate grid The waveform consistency component, after being cropped to zero to one from the maximum normalized waveform correlation coefficient within the allowable time shift interval, is a dimensionless quantity, with the sign... This indicates the aforementioned grid-level band power ratio consistency, denoted by [symbol]. The above-mentioned grid-level envelope evolution consistency is represented by the three weights, which are derived from the constrained grid search on the above-mentioned validation set and are summed to one. When any consistency component is invalid, it is renormalized according to the original weights of the remaining valid components. When all components are invalid, the edge features corresponding to the grid are not generated.

[0075] When at least two of the three valid consistency components corresponding to a certain grid reach their respective thresholds and the edge weight of the grid is not less than 0.65, the module sets the conflict flag of the grid to no; otherwise, it sets it to yes and retains it for S3 to calculate the total conflict. The module determines the direction of the directed edge based on the order in which the two nodes are corrected to the center value of the time interval. If the absolute value of the difference between the center values ​​does not exceed the sum of the half widths of the two intervals, the direction is determined by the positive or negative of the observed time shift of the corresponding grid. If it is still impossible to distinguish, it is registered as a bidirectional undetermined edge. The final edge record includes the grid feature mapping and model version of the starting node, ending node, compatible grid bitmap, grid number to the allowed time shift interval, observed time shift, three consistency degrees, edge weight, and conflict flag, and is submitted to S3. After S3 finds the intersection of the compatible grid bitmap of the edge, it only reads the time shift and features of the corresponding grid within the intersection and prohibits the use of features obtained outside the allowed interval of a certain grid to support the grid.

[0076] In this specific embodiment, S3 includes:

[0077] The consistency graph analysis module uses nodes with valid vibration feature records in this round as vertices and propagation evidence edges output by S2 as directed edges to construct a propagation consistency evidence graph. Vertex records store node number, spatial neighborhood number, node coordinates, correction to time interval, and data status. Edge records use the compatible grid bitmap and grid feature mapping. For the current candidate source feasible region of an edge, the module reads the direction, edge weight, and conflict flag of each grid in ascending order of grid number. The minimum edge weight is used as the edge-level weight. If any grid is in conflict or the directions of the grids are inconsistent, the edge-level conflict flag is set to one. If all directions are consistent, the consistent direction is taken; otherwise, it is registered as a bidirectional pending edge. When there are multiple edges for the same node pair, only the edge with the largest edge-level weight obtained according to the above rules and which is not empty in the compatible grid bitmap is retained, and the replaced edge is written into the audit chain.

[0078] The preset monitoring area is divided into spatial neighborhoods using hexagonal units with a side length of 5 kilometers. Boundary nodes are uniquely assigned to units according to their coordinates. The module selects unprocessed edges as seeds based on their edge weights from high to low, and establishes candidate subgraphs by their two endpoints and a compatible mesh bitmap. Then, it checks each adjacent edge that shares at least one vertex with the candidate subgraph. The adjacent edges intersect with the compatible mesh bitmaps of all edges in the existing subgraph bitmap bitmap, and only when the intersection is not empty is it retained as an expansion candidate.

[0079] For the edge set The candidate subgraphs are constructed, and the modules are arranged according to... Calculate the feasible region of candidate seismic sources, where the symbol Representing candidate subgraphs The feasible region of candidate seismic sources is represented by a set of candidate grid numbers, with symbols... Represents the current candidate subgraph, symbol Let the set of propagation evidence edges of the candidate subgraph be denoted by the symbol . The symbol represents the edge index in the edge set. Indicates the spread of evidence Only extended results of the associated compatible candidate source grid set, where the candidate source feasible region is not empty, proceed to the next check.

[0080] The module checks whether the direction of each directed edge in the candidate subgraph is consistent with the non-decreasing order of the node correction to the center value of the time interval. Bidirectional undetermined edges are not rejected for expansion separately but are included in the conflict amount, and are processed according to... Calculate the conflict weight percentage, where the symbol The conflict weights represent the proportion of candidate subgraphs, and are dimensionless quantities ranging from zero to one. A larger value indicates a stronger conflict. The symbol is... Representing an edge The conflict flag is set to 1 for conflicts and 0 for non-conflicts. Representing an edge The edge weights are dimensionless. When the sum of the edge weights of all propagated evidence edges is zero, expansion is rejected. In this specific implementation, the conflict threshold is set to 0.25.

[0081] Only when the feasible region of a candidate seismic source is not empty, the arrival order check passes, and the proportion of conflict weights does not exceed the conflict threshold, will the module add the extended edge and its new vertex to the candidate subgraph. After each bitmap intersection causes the feasible region of the current candidate seismic source to shrink, the module immediately recalculates the edge-level weights, edge-level conflict markers, and directions of all edges in the subgraph according to the shrunken grid set, and recalculates the seed sorting value, the proportion of conflict weights, and the average non-conflict edge weights accordingly. The total weight of local interference clusters also uses the same edge-level weights. If multiple extended edges have the same weight, they are sorted in ascending order by the number of feasible region grids, the truth value priority of whether a new spatial neighborhood is introduced, and the ascending order of endpoint node numbers. The first edge in the sorting is added to the candidate subgraph and written to the extension log. If an edge causes any extension condition to fail, it is not merged into the current subgraph and the failure condition is recorded, until there are no qualified adjacent edges.

[0082] Module by Calculate the geometric dispersion of nodes, where the symbol is... The geometric dispersion of nodes in a candidate subgraph is a dimensionless quantity between zero and one, with a larger value indicating a more dispersed spatial distribution of nodes. The sign is... The number of nodes in the candidate subgraph is represented by the symbol. The set of nodes in a candidate subgraph is represented by the symbol. ,symbol Representing nodes respectively and nodes The planar coordinate vector, in kilometers, with the symbol . This represents the maximum planar distance between any two nodes within the monitoring area, in kilometers. If the calculated result exceeds one, it is clipped to one; if the number of nodes is less than two, it is set to zero.

[0083] Before the subgraph filtering begins, the module performs consistency cleaning on vertex and edge records: it deletes abnormal edges with non-existent node numbers, identical start and end nodes, inconsistent lengths of compatible mesh bitmaps with candidate set versions, or edge weights exceeding the range of zero to one; it removes duplicate edges with the same start and end nodes, the same direction, and the same compatible mesh bitmap by event candidate number; vertices with missing node coordinates or whose spatial neighborhood numbers cannot be recalculated from coordinates are marked as invalid vertices and removed from the current graph along with their associated edges; if the number of cleaned vertices is less than four, the search for a connected and consistent subgraph stops and the current event is marked as insufficient evidence.

[0084] A candidate subgraph is registered as a qualified subgraph and written into the candidate subgraph set when it covers no less than three spatial neighborhoods, has no less than four nodes, a node geometric dispersion of no less than 0.15, and an average non-conflict edge weight of no less than 0.65. When there are multiple qualified subgraphs, they are sorted in descending order according to the number of covered neighborhoods, node geometric dispersion, and average non-conflict edge weight. When all indicators are the same, the subgraph with the fewest grids of candidate seismic source feasible domains is output as a connected and consistent subgraph. Subgraphs that are not output but share the same candidate seismic source feasible domain with the main subgraph are retained as auxiliary consistent subgraphs.

[0085] For two qualified subgraphs that share some nodes but whose candidate seismic source feasible regions do not intersect, the module does not force a merger. Instead, it compares the total weights of their non-conflicting edges, retains the one with the highest total weight as the main subgraph, and marks the other subgraph as a competing event branch. If the difference in the total weights of the two branches does not exceed 10%, both are written into the competing branch list and the main event of this round is set to a pending review state. S4 must not directly output an earthquake alarm, thereby avoiding the erroneous compression of two events that are close in time but have different propagation sources into one feasible region.

[0086] The module extracts edge sets from edges that have effective vibration energy at their endpoints, have edge weights of not less than 0.4, and are interconnected from edges that have never been incorporated into any connected and consistent subgraph. If the edge set covers fewer than three spatial neighborhoods, or if its compatible grid intersection is empty and the edge endpoints are concentrated in the same spatial neighborhood, it is identified as a local disturbance cluster. The local disturbance cluster record includes the edge set, endpoint nodes, covered neighborhoods, total weight, conflict cause, and earliest arrival time. Isolated edges with weights below the lower limit are only registered as insufficient evidence noise.

[0087] When the candidate subgraph set is empty, the module still performs local interference cluster determination on all unincorporated edges according to the aforementioned rules, and forms an event graph record where the main subgraph is empty and the candidate seismic source feasible domain is empty; if there is a local interference cluster with a total weight of not less than 1.5, the local interference pending judgment flag is recorded; otherwise, the insufficient evidence flag is recorded. S4 does not calculate the basic event credibility, update event credibility, or representative grid available node coverage rate in this branch; for the local interference pending judgment flag, the module divides the number of nodes with effective vibration characteristics, uninterrupted heartbeat, normal self-check code, and sample integrity rate of not less than 80% in the endpoints of the interference cluster by the total number of endpoint nodes of the interference cluster to obtain the local interference endpoint availability rate. When the endpoint set is empty, it is zero; when the endpoint availability rate is not less than 0.5, local interference is output; otherwise, insufficient evidence is output.

[0088] The consistency graph analysis module writes the vertex set, edge set, candidate seismic source feasible region, coverage space neighborhood set, node geometric dispersion, conflict weight ratio, and auxiliary consistency subgraph of the selected connected consistency subgraph into the event graph record. When the main subgraph is empty, the corresponding fields above are written with null values ​​and an empty main subgraph flag. At the same time, each local disturbance cluster is written into the disturbance cluster list. The event graph record shares the event candidate number with the vibration feature record of S1 and the edge record of S2. Then it is submitted to S4 to calculate the confidence level or execute the empty main subgraph branch.

[0089] In this specific embodiment, S4 includes:

[0090] The basic credibility and alarm module reads the event graph records output by S3 and normalizes the spatial coverage, node geometric dispersion, candidate source feasible domain volume, propagation time difference interval residual, and three types of vibration consistency. The numerator of the spatial coverage is the number of deduplicated spatial neighbors covered by the connected and consistent subgraph, and the denominator is the number of deduplicated spatial neighbors covered by nodes that have at least one non-temporally abnormal and non-incomplete data vibration feature record within the event association window when the S1 candidate record list is closed. The denominator node set and neighborhood number are written into the event candidate record along with the S1 candidate record list. When the denominator is zero, the event is downgraded to insufficient evidence. This spatial coverage only reflects the coverage of the effective participating neighborhoods of S1 and is different from the event-level available node coverage formed according to the expected arrival time four states. The candidate source feasible domain volume is obtained by summing the effective grid volumes and dividing by the total volume of the candidate grids in the monitoring area, then subtracting this ratio to make the more concentrated the feasible domain, the higher the score.

[0091] For each non-conflict edge in the selected connected and consistent subgraph and its candidate grids within the final feasible region, the module takes the absolute value of the difference between the observed time shift and the center value of the theoretical propagation time difference interval in the feature map of the grid, divides it by the sum of the half-width of the theoretical interval and 0.02 seconds, and then trims it to zero or one to obtain the normalized time shift deviation. When there is boundary contact, the deviation of the observed time shift relative to the center of the interval is still calculated. When the theoretical interval width is zero, 0.02 seconds is used as the lower limit of the denominator. The module uses the weighted average of the edge weights of all normalized time shift deviations minus one to form the time difference consistency score, thereby distinguishing boundary contact, narrow amplitude overlap and center overlap and ensuring that the score decreases monotonically when the deviation increases. The three types of vibration consistency are respectively taken as the weighted average of the edge weights with corresponding effective values ​​in the connected and consistent subgraph.

[0092] Module by Calculate the confidence level of the underlying events, where the symbol The confidence level of a basic event is represented by a dimensionless quantity ranging from zero to one, with larger values ​​indicating greater confidence. The symbol is... Indicates spatial coverage, symbol S3 indicates that a connected and consistent subgraph has been selected. The geometric dispersion of nodes, sign The score represents the feasible region concentration of candidate seismic sources, with the symbol... Indicates time zone consistency score, symbol The score represents the consistency score of the normalized waveform, with the symbol... Indicates the frequency band energy ratio consistency score, symbol The envelope evolution consistency score is represented by the following components, all of which are dimensionless quantities ranging from zero to one. The weights are determined by minimizing the weighted loss of false alarm rate and false alarm rate on the verification set of historical confirmed events and interference events, and the sum is one. Each type of vibration consistency score requires at least three non-conflict edges with valid values. Components lacking valid values ​​are not counted as zero, but their original weights are removed from the denominator and normalized again by the sum of the original weights of the remaining valid components. If there are fewer than five valid components among the seven components or the time difference consistency score has no valid edges, the event state is downgraded to insufficient evidence.

[0093] The state credibility fusion module estimates the desired nodes for each grid in the feasible domain of the candidate seismic source; for each grid... First, subtract the propagation time interval from the grid to the node and the site travel time correction interval from the correction time interval of each participating node to obtain the earthquake occurrence time interval inverted by the node. Then, find the intersection of the earthquake occurrence time intervals of all participating nodes and use the resulting absolute earthquake occurrence time interval as the time anchor point. If the intersection is empty, delete the grid. If the intersection is not empty, add it to the propagation time interval from the grid to the target node, the target node site correction interval, the picking uncertainty interval, and the residual clock difference interval to form the absolute estimated arrival time interval under the unified time base of the server. Subsequent cache readings and 200-millisecond delays are all calculated from the upper bound of this interval.

[0094] The module uses the median of the three-band corrected energy of the participating nodes acquired in real time to back-calculate the event energy scale according to the following attenuation formula. Then, it reads the geometric diffusion index, quality factor, and effective frequency band stored in the regional attenuation record obtained from historical event regression, which is divided by stratigraphic partition, as well as the site response baseline and noise floor of each node, and calculates the expected signal amplitude from the grid to each node. When the path length is within the applicable distance range of the attenuation record, it is used directly. When it exceeds the upper limit of the applicable distance but does not exceed 10% of the upper limit, it is calculated using the boundary distance and the uncertainty of the expected amplitude is increased by a residual standard deviation. When it is lower than the lower limit of the applicable distance or exceeds 10% of the upper limit, the grid is marked as infeasible and deleted. Subsequent expected nodes and negative evidence processes only process the grids retained in this order.

[0095] Expected amplitude based on Calculation, where the symbol Represents candidate source grid At the node The expected acceleration amplitude, in meters per second squared, sign This represents the event reference amplitude at the reference distance, derived from the median energy of the participating node bands, in meters per second squared. Represents a node The field response baseline ratio is a dimensionless quantity, with the symbol [symbol missing]. Represents a grid To node The path length, in kilometers, is used in this formula only if the path length is not lower than the applicable distance lower limit greater than zero in the regional attenuation record. If the original path length is zero or lower than the lower limit, the grid is marked as infeasible and deleted according to the aforementioned rules, without positive lower limit backoff. (Symbol) This indicates the reference distance used for regional attenuation recording. In this specific implementation, it is taken as one kilometer and is a positive value. The unit is kilometers, and the symbol is... The geometric diffusion index in the regional decay record is a dimensionless quantity, with the symbol [symbol missing]. The effective frequency band, representing the highest energy content, is measured in Hertz (Hz) and has the symbol [symbol missing]. The quality factor representing this stratigraphic zone is a positive dimensionless quantity, with the symbol […]. The effective velocity of the frequency band in the regional velocity model is represented by a positive value in kilometers per second, with the symbol [symbol missing]. Pi is a dimensionless mathematical constant; the distance ratio and the exponent are both dimensionless quantities, so the expected amplitude and the reference amplitude have the same unit.

[0096] The inversion of the event energy scale uses participating nodes in the main connected consistent subgraph as samples. The measured peak frequency of each node is corrected according to the site response baseline, and the reference amplitude is calculated by substituting the path length corresponding to the geometric center of the feasible domain of the candidate source into the regional attenuation relationship. After removing the highest and lowest values, the median is taken as the reference amplitude. When there are fewer than four participating nodes, the interquartile range of the inverted reference amplitude exceeds 1.5 times the median, or there is a lack of valid versions of the regional attenuation record, the event energy scale is marked as unpredictable. In this branch, the module no longer selects nodes based on the predicted amplitude, but instead selects nodes with effective propagation time. The set of online target nodes for the grid consists of all nodes in the interval, the site travel time correction interval, and the picking uncertainty interval. The absolute estimated arrival time interval of each node is generated using the aforementioned time anchor point. All nodes in the set are classified as unknown evidence nodes, and no positive or negative evidence nodes are established. When the set of expected time nodes is empty, the grid is deleted. When it is not empty, it is used as the set of expected nodes for this branch and the grid-level available node coverage is set to zero. This ensures that subsequent representative grid selection, coverage recording, and confidence updates have definite inputs and will not cause erroneous rejection due to incomplete amplitude estimation.

[0097] The regional attenuation record is robustly regressed from no less than one hundred confirmed events according to stratigraphic zones and main frequency bands. The fields include geometric diffusion index, quality factor, effective velocity, lower limit of applicable distance, upper limit of applicable distance, residual standard deviation, sample size, and version number. The path retained according to the aforementioned 10% boundary extrapolation rule increases the uncertainty of the expected amplitude by one residual standard deviation. Only nodes whose expected amplitude still reaches the detectable threshold after deducting this uncertainty are included in the expected node set. When the model version is updated, the detectable threshold record is used together with the historical event playback verification.

[0098] When the expected signal amplitude reaches the node's detectable threshold, the module includes the node in the expected node set of the corresponding grid. The detectable threshold is three times the 99th percentile of the node's amplitude without event noise and no less than five times the sensor resolution. The recorded fields include node number, expected amplitude, detectable threshold, expected arrival time interval, and baseline version used. Nodes with expected amplitudes below the threshold do not provide positive or negative evidence. When a candidate grid lacks attenuation parameters, the default value is not used for inference, but the grid is marked as model incomplete.

[0099] The module reads the communication heartbeat, self-check code, and data integrity rate within the expected arrival time interval of each expected node. Nodes with heartbeat interruptions exceeding 30 seconds or abnormal self-check codes are classified as unusable evidence nodes. Among other nodes, those with a ratio of valid samples to expected samples of less than 80% are classified as unknown evidence nodes. Among the remaining nodes, those that form compatible propagation evidence edges with the connected consistent subgraph are classified as positive evidence nodes, and those that do not form compatible propagation evidence edges are classified as negative evidence nodes. The four types of states are mutually exclusive. Unknown evidence nodes and unusable evidence nodes are not used as evidence of no response.

[0100] For grid Negative evidence nodes, module according to Calculate the relative amplitude, where the sign Represents a node Candidate source grid The expected amplitude exceeding the limit is a dimensionless quantity ranging from zero to one, with the sign... Indicates the aforementioned expected amplitude, symbol Represents a node The detectable gate limit, in meters squared seconds, symbol This represents the normalized positive lower limit of amplitude, which is five times the sensor resolution, and the unit is the same as the detectable threshold; the relative amplitude is zero when the expected amplitude equals the threshold.

[0101] Module press again Calculate the negative evidence weight, where the sign... Represents a node Candidate source grid The negative evidence weight is a dimensionless quantity ranging from zero to one, with a larger value indicating a stronger contradiction and no response. The sign... This represents the expected data completeness rate within the specified time interval, a dimensionless quantity between zero and one, with the symbol […]. The reliability coefficient is determined based on the noise floor stability of the node and is a dimensionless quantity ranging from zero to one. When the absolute deviation of the median noise in the most recent 24 hours from the median of the deployment baseline does not exceed one absolute median difference, the reliability coefficient is taken as one; when it is greater than one but not more than two, it is taken as 0.75; when it is greater than two but not more than three, it is taken as 0.5; when it is greater than three, it is taken as 0.25. When it is exactly at the boundary of one, two, or three times, it is classified into the interval of reliability coefficient of one, 0.75, or 0.5, respectively. When the baseline is missing, the absolute median difference is zero, or the noise statistical sample is less than one thousand, negative evidence is not calculated and the node is classified as an unknown evidence node.

[0102] The initial support value of each candidate grid is 1. Negative evidence nodes are processed in ascending order of node number. When the negative evidence weight reaches 0.8, the grid is deleted. When it is below the exclusion threshold, the current support is multiplied by 1 minus a reduction factor of 0.5 times the negative evidence weight. Grids with incomplete models do not participate in negative evidence inference and are directly deleted, and model gaps are recorded. After processing, the module forms mutually exclusive sets of positive evidence, negative evidence, unknown evidence, and unusable evidence for each retained grid. The grid-level usable node coverage is obtained by dividing the sum of the number of positive evidence and negative evidence nodes in the grid by the total number of expected nodes in the grid. Grids with empty expected node sets are deleted.

[0103] The module selects the grid with the highest support value as the representative grid for the event. When the support values ​​are the same, it selects the grid with the lowest volume value of the connected components of the candidate feasible region and the grid with the lowest grid number value in turn. The four-state set of the representative grid and the grid-level available node coverage are used as the event-level node state and available node coverage, and the support of the representative grid is used as the negative evidence retention coefficient. When there are no grids to retain after deletion, the event state is set to insufficient evidence. When there are still grids, the candidate source feasible region set score is recalculated based on the retained grids. The recalculated set score is then substituted back into the aforementioned basic event credibility calculation formula, keeping the other effective components and weight normalization rules unchanged, to obtain the basic event credibility after grid deletion. Subsequent updates to event credibility only use the basic event credibility after grid deletion, thus prohibiting the direct mixing of the pre-deletion score or the state of the same node under different grids.

[0104] Module by Update event credibility, where the symbol Represents the confidence level of an updated event, a dimensionless quantity between zero and one, with the symbol […]. This represents the confidence level of the basic events after grid deletion, obtained by recalculating the feasible region set score of candidate seismic sources using the retained grid and then back to the aforementioned calculation formula. The symbol is... Represents the event-level available node coverage of the mesh, a dimensionless quantity between zero and one, with the symbol […]. This represents the negative evidence retention coefficient for the grid, a dimensionless quantity ranging from zero to one; unavailable evidence nodes and unknown evidence nodes are not included in the numerator of available node coverage.

[0105] The module uses the credibility of the updated event as the credibility of the event for result discrimination in step S4, and no longer directly triggers alarms based on the credibility of the base event of the unfused node state; when only one discrete grid remains after the negative evidence grid is deleted and there is still incomplete data transmission between the grid and any negative evidence node in the expected arrival time interval, the result discrimination is delayed until 200 milliseconds after the end of the expected arrival time interval. During the delay period, the newly arrived data is re-classified into four states, and nodes that are still incomplete after the delay time limit remain in the unknown evidence state;

[0106] When the confidence level of the updated event reaches the confidence level alarm threshold of 0.78, the stability verification module starts the node-by-node elimination. It temporarily deletes one participating node and all propagation evidence edges connected to it in turn according to the participating node number. It redetermines the compatible edges, connected consistent subgraphs, candidate source feasible regions and updated event confidence levels using the same model version, threshold and four-state classification rules as S2, S3 and this step. Each reconstruction does not inherit the previous temporary deletion status. When the reconstruction fails or the feasible region is empty, the confidence level of the reconstruction is recorded as zero.

[0107] For the removed nodes Modules are Calculate the rate of change of the feasible region volume, where the symbol is... Indicates the removal of nodes The subsequent volume change rate of the feasible region of the candidate seismic source is a non-negative dimensionless quantity, with the sign... Indicates the removal of nodes The volume of the feasible domain of the candidate seismic source is given in cubic kilometers, with the symbol [symbol missing]. This represents the volume of the feasible domain of candidate seismic sources before elimination, in cubic kilometers, with the symbol... This represents the volume of a single candidate source grid, in cubic kilometers. Using the maximum value in the same unit as the denominator can prevent the original feasible region volume from being zero.

[0108] The module sorts the confidence of each reconstruction event in ascending order to obtain a confidence sequence, and sorts the volume change rate of the feasible region of the candidate source before and after each reconstruction in descending order to obtain a change rate sequence. When the confidence is the same, the order is maintained by the number of the removed node in ascending order. When the change rate is the same, the same rule is used. The first item of the confidence sequence, i.e. the confidence of the lowest reconstruction event, and the first item of the change rate sequence, i.e. the maximum volume change rate, together constitute the most unfavorable stability index. The sequence and its node index are written into the stability record.

[0109] Module by Calculate the rate of change of the total weights of the propagation evidence, where the sign... Indicates the removal of nodes The relative rate of change of the total edge weights of the propagation evidence is a dimensionless quantity ranging from zero to one, with the sign... Indicates the removal of nodes The total weight of the edges that are retained afterward, sign This represents the total weight of the edges retained before removal; both are dimensionless quantities, and the symbol is... This represents the positive lower limit of the total weight and is taken as 10 to the power of negative 6. When the total weight before elimination is zero, this positive lower limit is used as the denominator and the absolute difference is retained, thus obtaining a unique backoff result.

[0110] Module press again Calculate the node dominance index, where the symbol Represents a node The node dominance index is a dimensionless quantity ranging from zero to one, with a larger value indicating stronger dominance of the node over the fusion result. The sign is... This represents the absolute change in the credibility of the update event before and after removing the node, pruned to zero or one. (Symbol: ...) The symbol represents the rate of change of the volume of the aforementioned feasible region. This represents the rate of change of the total weights of the aforementioned propagation evidence. The three weights are determined by removing historical nodes one by one from the verification results, and their sum is one. The symbol representing the upper bound of clock drift in S1 They are independent of each other and not reused;

[0111] The module outputs an earthquake alarm only when the update event confidence is not less than 0.78, the minimum reconstruction event confidence is not less than 0.68, the maximum volume change rate does not exceed 0.6, and the node dominance index of all participating nodes is less than 0.55; otherwise, the earthquake alarm is prohibited. In the branch where the alarm is prohibited, if a local interference cluster exists, its total edge weight is not less than 1.5, and the available node coverage is not less than 0.5, then local interference is output; otherwise, the result of insufficient evidence is output. The threshold record is jointly calibrated by historical confirmed events, local construction interference, and node fault playback and is fixed with the alarm strategy version.

[0112] As a result, the state machine only allows three mutually exclusive states for the same candidate event number: insufficient evidence, local interference, and earthquake alarm. Insufficient evidence can be converted to local interference or earthquake alarm after adding a valid node. Local interference can only be converted to earthquake alarm after a connected and consistent subgraph spanning three spatial neighborhoods is re-formed and passes the stability check. Once an earthquake alarm is issued, it is not automatically withdrawn but an evidence update record is added. If the competing event branch is in a pending review state, the model version used in the reconstruction process is inconsistent, or any threshold configuration check fails, the state machine will force the output of insufficient evidence and record the reason for the downgrade.

[0113] The output records include the event candidate number, result type, update time, candidate source region outer polygon, main propagation direction, participating nodes, positive and negative unknown unusable evidence nodes, conflict nodes, dominant nodes, local interference clusters, basic event credibility, updated event credibility, minimum reconstruction event credibility, maximum volume change rate, and reconstruction results that lead to downgrade. Nodes with a node dominance index of 0.55 are listed as dominant nodes, and the two endpoints of the edge with conflict-marked propagation evidence are listed as conflict nodes. Earthquake alarm records are sent to the early warning terminal through a dual-channel message queue. Local interference and insufficient evidence records are written into the event audit database and used for threshold review.

[0114] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0115] This invention uses interval propagation constraints as a pre-gating control for waveform correlation, so that waveform correlation, frequency band energy and envelope evolution characteristics form evidence only between node pairs with physical propagation feasibility, and the continuity of event propagation is verified by intersection of the common feasible domains of the evidence graph and spatial neighborhood coverage, thereby distinguishing local vibration from multi-node consistent propagation events.

[0116] This invention implements mutually exclusive evidence state classification for detectable expected nodes and removes participating nodes one by one for reconstruction before alarming, constraining alarm output with the most unfavorable result; this improvement changes the processing structure of negating all missing nodes and directly alarming with single fusion, so that nodes that are offline, have lost packets or are abnormal participate in the judgment only according to their true evidence ability.

Claims

1. An alarm method based on spatiotemporal consistency discrimination of multi-source ground motion, characterized in that, include: S1. Obtain vibration sequences, node coordinates and sampling timestamps of multiple ground motion monitoring nodes, perform inter-node comparability processing on the vibration sequences, extract candidate first arrivals and vibration features, determine the residual clock difference interval based on the node's time synchronization information and clock drift information, and synthesize the candidate first arrival picking uncertainty interval and the residual clock difference interval to correct the node to the time interval. S2. For node pairs and candidate source locations, determine the theoretical propagation time difference interval based on node coordinates and propagation velocity range, determine the observation time difference interval based on the node correction to time interval of the node pair, and when the theoretical propagation time difference interval intersects with the observation time difference interval, generate propagation evidence edges that are compatible with candidate source location constraints based on vibration characteristics within the intersection range. S3. Construct a propagation consistency evidence map based on local earthquake monitoring nodes and propagation evidence edges. From the propagation consistency evidence map, identify connected and consistent subgraphs that are explained by the feasible domain of the same candidate source and cover multiple spatial neighborhoods. Then, identify local interference clusters based on propagation evidence edges that cannot be incorporated into the connected and consistent subgraphs. S4. Determine the event credibility based on the spatial coverage of the connected and consistent subgraph, the feasible domain of the candidate seismic source, and the consistency of the propagation evidence edges. Output the earthquake alarm, local interference, or insufficient evidence result based on the event credibility and the local interference cluster.

2. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 1, characterized in that, S1 includes: Acquire synchronization records, synchronization age, clock drift upper limit, frame loss records, and installation parameters of various seismic monitoring nodes; The residual clock error interval is determined based on the product of the synchronization deviation, synchronization age, and the upper bound of the clock drift, as well as the time stamp gap corresponding to the lost frame. Frequency band division and normalization benchmark are determined based on the noise power spectrum during non-event periods. The vibration sequence is rectified according to the installation transfer baseline corresponding to the installation parameters. The envelope change position is extracted in each frequency band as a candidate first arrival. The picking uncertainty interval is determined according to the signal-to-noise ratio distribution in the neighborhood of the candidate first arrival. The normalized waveform segment, frequency band energy sequence and envelope evolution sequence based on the candidate first arrival are extracted as the vibration features.

3. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 2, characterized in that, S2 includes: Discretize the preset monitoring area to obtain a set of candidate earthquake source locations; For each candidate source location in the candidate source location set, the theoretical propagation time difference interval is determined based on the path length from the candidate source location to the two nodes in the node pair, the lower and upper bounds of the regional propagation velocity, and the site travel time correction of each node. The node corrections of the two nodes in the node pair are subtracted from the time interval to obtain the observation time difference interval. When two time difference intervals intersect, the candidate source location constraint that causes the two time difference intervals to intersect is associated with the corresponding propagation evidence edge, and the normalized waveform correlation coefficient, frequency band energy ratio consistency, and envelope evolution consistency are calculated only within the time shift corresponding to the intersection range. The edge weight and conflict marker are determined based on the comparison results of the three with their respective discrimination thresholds, and the propagation direction is determined based on the order of observation arrival. When two time difference intervals do not intersect, compatible propagation evidence edges are not generated between the corresponding node pairs.

4. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 3, characterized in that, S3 includes: taking ground motion monitoring nodes as vertices and propagation evidence edges as directed edges; obtaining the candidate source feasible region of the subgraph by intersecting the compatible candidate source location constraints associated with the propagation evidence edges; and using the following conditions for subgraph expansion: the candidate source feasible region is non-empty, the node arrival order is consistent with the direction of the directed edge, and the total weight of propagation evidence edges with conflict markers does not exceed the conflict threshold; filtering the connected and consistent subgraph according to the number of spatial neighbors to which the nodes belong and the geometric dispersion between nodes; and determining the set of propagation evidence edges that have vibration energy but do not meet the subgraph expansion conditions and whose covered spatial neighbors are less than the spatial coverage threshold as the local interference cluster.

5. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion according to claim 4, characterized in that, S4 includes: determining the credibility of a basic event based on a weighted combination of spatial coverage of the connected consistent subgraph, node geometric dispersion, candidate source feasible domain volume, interval residuals between the observation time difference interval and the theoretical propagation time difference interval, normalized waveform correlation coefficient, frequency band energy ratio consistency, and envelope evolution consistency; outputting an earthquake alarm when the credibility of the basic event reaches the credibility alarm threshold; outputting local interference when the credibility of the basic event does not reach the credibility alarm threshold, and the local interference cluster exists and the total weight of its propagation evidence edges reaches the interference discrimination threshold; and outputting insufficient evidence results in other cases.

6. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 5, characterized in that, Based on the candidate source locations within the feasible domain of candidate seismic sources, the event energy scale determined by the frequency band energy of participating nodes, the regional attenuation envelope, the site response baseline of each node, and the noise floor, the expected signal amplitude and expected arrival time interval of each node are determined. Nodes whose expected signal amplitude reaches the corresponding node's detectable threshold are included in the expected node set. The communication status, self-checking status, and data integrity rate of the expected nodes are obtained. Expected nodes that are offline or have abnormal self-checking are identified as unusable evidence nodes. Among the remaining expected nodes, nodes whose data integrity rate does not reach the integrity rate threshold are identified as unknown evidence nodes. Nodes among the remaining expected nodes that form compatible propagation evidence edges are identified as positive evidence. Nodes that do not form compatible propagation evidence edges are identified as negative evidence nodes. The expected arrival time interval and detectable threshold of negative evidence nodes are used to exclude candidate source locations that contradict their lack of response. The credibility of the basic event is updated based on the updated feasible domain of candidate sources and various types of evidence nodes to obtain the updated event credibility. This updated event credibility is then used as the event credibility for result judgment in step S4. The ratio of the sum of the number of positive evidence nodes and negative evidence nodes to the total number of expected nodes is determined as the available node coverage rate. Unknown evidence nodes and unusable evidence nodes are not considered negative evidence nodes and are not included in the numerator of the available node coverage rate.

7. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 6, characterized in that, For each candidate source location in the feasible domain of the candidate source, the expected node set is generated. The negative evidence weight is determined based on the magnitude by which the expected signal amplitude of each negative evidence node for the candidate source location exceeds the detectable threshold, the data integrity rate within the expected time interval, and the noise floor of the corresponding node. Candidate source locations whose negative evidence weight reaches the exclusion threshold are deleted. The credibility of the remaining candidate source locations is reduced according to the negative evidence weight. When the feasible domain of the candidate source is empty, the event result is updated to insufficient evidence.

8. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion according to claim 6, characterized in that, Before outputting an earthquake alarm, participating nodes in the connected and consistent subgraph are temporarily removed sequentially. For each removal, compatible propagation evidence edges and various types of evidence nodes are re-determined, and the connected and consistent subgraph and candidate source feasible regions are reconstructed. The credibility of the updated event is recalculated based on the reconstruction results to obtain the credibility of the reconstructed event. The credibility of each reconstructed event is arranged in ascending order to obtain a credibility sequence, and the volume change rate of the candidate source feasible region before and after each reconstruction is arranged in descending order to obtain a change rate sequence. The most unfavorable stability index is determined based on the first term of the credibility sequence and the first term of the change rate sequence. The node dominance index is determined based on the total change rate of propagation evidence edge weights before and after the removal of a single participating node, the change in the credibility of the updated event, and the volume change rate of the candidate source feasible region.

9. The alarm method based on spatiotemporal consistency discrimination of multi-source ground motion as described in claim 8, characterized in that, When the credibility of the update event reaches the credibility alarm threshold, the first item of the credibility sequence sorting corresponding to the most unfavorable stability index reaches the stability alarm threshold and the first item of the change rate sequence sorting does not exceed the change threshold, and the node dominance index is less than the dominance threshold, an earthquake alarm is output. Otherwise, the earthquake alarm will be prohibited from being output, and the result of local interference or insufficient evidence will be output based on the existence of local interference clusters and the coverage of the available nodes. Participating nodes whose dominance index reaches the dominance threshold are identified as dominating nodes, and endpoint nodes of propagation evidence edges with conflict markers are identified as conflict nodes. The output results include candidate source regions, propagation directions, participating nodes, conflict nodes, dominating nodes, and reconstruction results that lead to downgrade.

10. An alarm system based on spatiotemporal consistency determination of multi-source ground motion, used to execute the alarm method based on spatiotemporal consistency determination of multi-source ground motion as described in any one of claims 1 to 9, characterized in that, include: The multi-node correction and feature generation module is used to acquire data from multiple ground motion monitoring nodes, perform inter-node comparability processing, and generate node correction to time intervals and vibration characteristics. The propagation evidence generation module is used to determine the theoretical propagation time difference interval and the observation time difference interval, and generate propagation evidence edges associated with compatible candidate source location constraints; the consistency graph analysis module is used to construct a propagation consistency evidence graph, determine the connected consistency subgraph, the feasible region of candidate sources, and local interference clusters; the basic credibility and alarm module is used to determine the credibility of basic events and output earthquake alarm, local interference, or insufficient evidence results; the state credibility fusion module is used to generate the expected node set, classify various types of evidence nodes, and use negative evidence nodes to constrain the feasible region of candidate sources and update the credibility of basic events. The stability verification module is used to temporarily remove participating nodes one by one, reconstruct the connected and consistent subgraph, the feasible domain of candidate seismic sources and the credibility of events, determine the most unfavorable stability index and node dominance index, and limit alarm output accordingly.

Images