Earthquake network multi-neighborhood multi-component cooperative clock error early warning system and method

Abstract

The invention discloses a seismic network multi-neighborhood multi-component cooperative clock error early warning system and method, and belongs to the technical field of seismic network clock errors. According to the method, a reference station set is selected for a target station from a station network neighborhood of the target station, the continuous waveform of each station is subjected to standardization processing, and a multi-component analysis sequence is generated. For each station pair, the relative clock offset and the quality index of the current period are estimated by calculating the similarity between the station pair and the long-term reference feature. And fusing the offsets of all station pairs, and obtaining an absolute clock offset estimation value and uncertainty of the target station based on mass weighting. And analyzing a time sequence formed by historical and current offset values, and automatically identifying a clock error state type. According to a preset early warning rule associated with each state, the current offset value, the uncertainty and the state duration are integrated for judgment, early warning information output of the corresponding grade is triggered, and automatic monitoring and graded early warning of the seismic station clock offset are achieved.

Description

Technical Field

[0001] This invention relates to the field of seismic network clock error technology, specifically to a multi-neighborhood multi-component collaborative clock error early warning system and method for seismic networks. Background Technology

[0002] The temporal accuracy of seismic observation data is fundamental for applications such as earthquake location, focal mechanism analysis, structural imaging, and earthquake early warning. Seismic stations typically rely on satellite timing systems to synchronize their internal clocks. However, in actual operation, factors such as equipment aging, power supply anomalies, communication interruptions, environmental interference, or human error can cause varying degrees of clock shifts or drifts, thus affecting the consistency of seismic data at the network scale.

[0003] Currently, there are several main technical approaches to address the clock error problem at seismic stations.

[0004] At the engineering application level, some technologies are primarily geared towards seafloor seismometers or temporary observation equipment. When continuous satellite timing signals are unavailable, methods based on background noise cross-correlation or event waveform similarity are typically used to estimate and correct relative time deviations between stations after data acquisition. While these solutions have practical value in specific offline scenarios, their core objective focuses on post-event correction, and their application targets and operational modes are fundamentally different from the operational and management needs of land-based fixed seismic networks that require long-term online monitoring and real-time early warning.

[0005] At the academic research level, numerous studies have utilized the cross-correlation function of seismic background noise or the waveform similarity of repeated earthquakes, blasting events, etc., to analyze and discuss clock errors between stations. These studies have verified the technical feasibility of identifying sub-second time offsets through waveform comparison. However, firstly, the above analyses are usually based on limited time periods or specific events, constituting post-event case analysis, which cannot meet the operational and maintenance needs of continuous, online monitoring of the network. Secondly, the process heavily relies on the analyst's manual experience in data screening, parameter adjustment, and result interpretation, resulting in low automation.

[0006] Furthermore, from the perspective of seismic network operation monitoring, clock error is output as an indicator of waveform data quality to reflect the station's operational status. However, this approach struggles to automatically distinguish and characterize different types of clock anomalies, such as sudden changes and drifts, and cannot generate clear and timely early warning information.

[0007] In summary, this invention proposes a method that can adapt to the long-term online operation characteristics of seismic networks and realize the automatic identification and early warning of clock errors at seismic stations. Summary of the Invention

[0008] The purpose of this invention is to overcome the problems in the prior art and provide a multi-neighborhood multi-component collaborative clock error early warning system and method for seismic networks.

[0009] A method for early warning of multi-neighborhood, multi-component collaborative clock errors in seismic networks includes: For the target station, at least K stations that meet the preset spatial constraints are selected from the seismic network as its neighborhood reference set. The continuous waveform data of the target station and each station in the reference set are standardized and preprocessed to generate a multi-component standardized analysis sequence for subsequent comparison. For a station pair formed by the target station and each reference station in the neighborhood reference set, based on the multi-component normalized analysis sequence of both stations, the relative clock offset of the station pair in the current analysis period and the corresponding estimated quality index are estimated by calculating the similarity with the long-term reference feature sequence. For the relative clock offsets of all station pairs, a weighted fusion is performed based on their estimated quality indices to finally generate the absolute clock offset estimate of the target station in the current analysis period and the uncertainty characterizing the dispersion of the estimate. A trend analysis is performed on the time series composed of the historical and current absolute clock offset estimates of the target station. Based on the preset morphological classification rules, the current state type of the clock error is automatically identified. The state type includes at least one of abrupt change, linear drift, and nonlinear drift. Based on the identified state type and according to the preset warning rules associated with different state types, the current absolute clock offset estimate, its uncertainty, and the duration of the state are comprehensively judged to trigger the output of warning information of different levels.

[0010] Preferably, by calculating the similarity with the long-term reference characteristic sequence, the relative clock offset of the station pair in the current analysis period and the corresponding estimated quality index are estimated, including: For a pair of stations, calculate the preliminary time delay estimates and corresponding similarity peak values ​​for the combination of multiple seismic waveform components. A consistency check is performed on the preliminary time delay estimates of the multiple component combinations, and inconsistent component combination results are eliminated. Using the square of the peak similarity of each component combination that passed the test as the weight, the preliminary time delay estimate is weighted and averaged to obtain the relative clock offset and the estimated quality index of the station pair.

[0011] Preferably, the preset morphological classification rules include: If there exists an analysis period in which the absolute value of the daily change of the absolute clock offset estimate is greater than the step threshold, and the offset remains the same for at least a preset number of consecutive periods thereafter, it is identified as a sudden change state. If, within a window of a predetermined number of consecutive analysis periods, the absolute value of the linear fitting slope of the absolute clock offset estimate sequence is greater than the drift slope threshold, and the goodness of fit is greater than the goodness of fit threshold, then it is identified as a linear drift state.

[0012] Preferably, the preset warning rules associated with different state types are set with hierarchical triggering conditions, including at least a first-level warning triggering condition and a second-level warning triggering condition; When the absolute value of the current absolute clock offset estimate exceeds the first threshold and its uncertainty is lower than the second threshold, the first-level warning trigger condition is triggered. If the duration of the abnormal state reaches the minimum confirmation time corresponding to the state type after the first-level warning is triggered, the second-level warning trigger condition will be triggered.

[0013] Preferably, the minimum confirmation time is not less than the length of the sliding time window used to generate the long-term reference feature sequence; for the linear drift state, the minimum confirmation time is not less than twice the length of the sliding time window of the long-term reference feature sequence.

[0014] Preferably, the long-term reference feature sequence is constructed and updated in the following manner: during the system initialization phase, the feature sequences of all quality-compliant days within a historical period are superimposed and averaged to generate an initial reference feature sequence; during system operation, the long-term reference feature sequence is updated using a sliding window method with an update period longer than the analysis period.

[0015] Preferably, the standardization preprocessing includes robust amplitude normalization processing, specifically: for each component data in each analysis period, the median of its absolute value is calculated as a robust scale, and the data is normalized using this scale.

[0016] This invention also provides a multi-neighborhood, multi-component collaborative clock error early warning system and method for seismic networks. The system is deployed on the centralized data processing platform of the seismic network and includes: The system includes: a data preprocessing and access module for acquiring and standardizing waveform data from multiple stations; a multi-neighborhood reference construction module for building and maintaining neighborhood reference sets for each target station; a multi-component collaborative analysis module for performing multi-component time delay estimation between station pairs; a clock offset estimation and stability constraint module for generating clock offset estimates and uncertainties for target stations; an automatic clock error state identification module for automatically identifying the state of clock offset time series; and an early warning triggering and result output module for triggering tiered early warnings based on the identification results and preset rules. The modules described above are configured to work together to execute any of the methods described above.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves the following beneficial technical effects in the identification and management of clock errors during the operation of seismic networks by introducing multi-neighborhood redundancy constraints and multi-component collaborative analysis mechanisms: By constructing multiple neighborhood reference stations and applying collaborative constraints, we avoid relying on a single reference station or a single path for judgment, thereby reducing the impact of local anomalies or problems with the reference stations themselves on the identification results and improving the stability and reliability of clock error identification results during long-term operation.

[0020] By using a multi-component collaborative analysis mechanism, the analysis results of different components are comprehensively evaluated, reducing the possibility of misjudgment caused by single component anomalies or noise interference, and enabling the system to adapt to the complex and variable noise conditions in the seismic network.

[0021] By automatically analyzing the time series characteristics of clock errors, the system can automatically identify and classify different clock anomalies, reducing manual intervention and subjective judgment, and improving the automation level of network operation and management.

[0022] By comparing the identification results with preset threshold conditions, early warning information is automatically output when the trigger conditions are met, enabling operation and maintenance personnel to promptly detect and handle clock anomalies, which is conducive to the long-term stable operation of the seismic network. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the data standardization preprocessing process in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the construction and replacement process of multiple neighboring stations according to an embodiment of the present invention.

[0024] Figure 3 This is a flowchart of the multi-component collaborative analysis process at the station according to an embodiment of the present invention.

[0025] Figure 4 This is a flowchart illustrating the clock error estimation process for the target station in an embodiment of the present invention.

[0026] Figure 5 This embodiment of the invention provides clock error estimation using three adjacent pairs of HE.TAG stations.

[0027] Figure 6 This embodiment of the invention provides clock error estimation using five adjacent pairs of HE.TAG stations.

[0028] Figure 7 This embodiment of the invention provides clock error estimation using 7 adjacent pairs of HE.TAG stations.

[0029] Figure 8 For embodiments of the present invention with 3, 5, and 7 adjacent pairs configured, including a human-interfering station pair, the absolute clock error of the target station HE.TAG is compared.

[0030] Figure 9 This invention provides a comparison of the absolute clock error of the target station HE.TAG under the configurations of 3, 5, and 7 adjacent pairs, simultaneously including two man-made interference station pairs of the same magnitude.

[0031] Figure 10 The clock drift results of the HE.TAG station are shown in this embodiment of the invention when the fixed drift of 0.5 seconds lasts for different durations (1, 3, 5, 7, 11, 31 days).

[0032] Figure 11 The clock drift results of the HE.TAG station are shown in this embodiment of the invention when the 2-second fixed drift lasts for different durations (1, 3, 5, 7, 11, 31 days).

[0033] Figure 12 The clock error detection results for the HE.TAG station after applying a linear time drift that uniformly increases from 0 to 2 s over 5, 7, 11, and 31 days, respectively, are shown in this embodiment of the invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0035] The seismic network multi-neighborhood multi-component collaborative clock error early warning method provided by this invention includes: For the target station, at least K stations that meet the preset spatial constraints are selected from the seismic network as its neighborhood reference set. The continuous waveform data of the target station and each station in the reference set are standardized and preprocessed to generate a multi-component standardized analysis sequence for subsequent comparison. For a station pair formed by the target station and each reference station in the neighborhood reference set, based on the multi-component normalized analysis sequence of both stations, the relative clock offset of the station pair in the current analysis period and the corresponding estimated quality index are estimated by calculating the similarity with the long-term reference feature sequence. For the relative clock offsets of all station pairs, a weighted fusion is performed based on their estimated quality indices to finally generate the absolute clock offset estimate of the target station in the current analysis period and the uncertainty characterizing the dispersion of the estimate. A trend analysis is performed on the time series composed of the historical and current absolute clock offset estimates of the target station. Based on the preset morphological classification rules, the current state type of the clock error is automatically identified. The state type includes at least one of abrupt change, linear drift, and nonlinear drift. Based on the identified state type and according to the preset warning rules associated with different state types, the current absolute clock offset estimate, its uncertainty, and the duration of the state are comprehensively judged to trigger the output of warning information of different levels.

[0036] In this embodiment, the data access and preprocessing stages are performed in the data access module and the preprocessing module. The data access module receives continuous seismic waveform data from multiple seismic stations and performs unified time format management, integrity checks, and caching on the received data. This module provides standardized data input for subsequent analysis and can set the data update cycle according to the network operation requirements.

[0037] The purpose of data processing in this embodiment is to receive continuous waveform data from multiple stations, complete unified formatting, quality marking, and standardized preprocessing, and provide comparable input for subsequent time delay estimation.

[0038] The input to the data access module and the preprocessing module is continuous waveform data. ,in i Station numbering c Number the channels / components. The output is the standardized analysis sequence. With quality mark .

[0039] like Figure 1 As shown, the steps of the data preprocessing method in this embodiment are as follows: (1) Time alignment and missing segment handling: Segment by day. Let the first segment be... d The total daily duration is The cumulative duration of valid data is (Effective sampling duration after removing missing and outlier segments). The quality marker is defined as a piecewise function:

[0040] ; in A configurable threshold (preferably 0.8-0.95). When At that time, the data from that channel will not be used for subsequent processing.

[0041] (2) Mean removal / Detrending / Instrument response removal: The effective data is subjected to mean removal, detrending, and instrument response removal, and a unified output bit velocity sequence is generated. .

[0042] (3) Bandwidth limitation: for In the preset frequency band or period Internal filtering is performed to suppress out-of-band interference.

[0043] (4) Robust amplitude normalization: To reduce the impact of amplitude differences between different stations / time periods on similarity calculation, a robust scaling factor is calculated for each daily segment d: ; in To prevent the denominator from being too small and causing numerical instability, a stable term is established. Normalization yields:

[0044] (5) Caching and update cycle: The standardized analysis sequences within the current analysis time window and its corresponding quality mark The data is written to the cache. The analysis time window covers data from multiple consecutive days. The system updates the analysis window on a rolling basis with natural days as the update cycle. That is, each time the analysis window is rolled forward by one natural day, and the intermediate analysis results of the stations for the day are generated based on the superposition of data from multiple days within the window. Data that exceeds the time window will no longer reside in the cache.

[0045] Multi-neighbor reference modeling Multi-neighborhood reference modeling is used to construct multiple sets of neighboring reference stations for a target station based on the spatial distribution relationships of seismic stations. The neighborhoods are constructed based on the spatial distance, azimuth distribution, or preset rules between stations, and are determined during the system initialization phase to ensure the consistency and comparability of reference values ​​subsequently constructed based on long-term stacking results.

[0046] During long-term operation of the station, the neighborhood structure remains relatively fixed, used for continuous monitoring and analysis of the target station's clock error. When a reference station within the neighborhood experiences persistent anomalies, data interruptions, or fails to meet quality requirements, the system can adjust or replace the corresponding neighborhood by reconstructing the reference value, thereby maintaining the stability of the overall analysis results.

[0047] The purpose of multi-neighborhood reference modeling is to construct a set of reference stations with redundant constraints for the target station, thereby reducing the risk of misjudgment caused by anomalies in a single reference station.

[0048] like Figure 2 As shown, for the target station Candidate reference sets are obtained based on spatial distance (or combined distance and orientation distribution constraints). To reduce the impact of strong coherent interference at close range on the stability of characteristic sequences, the distance between the reference station and the target station should preferably not be less than a preset lower limit. D min (e.g., 30km, configurable).

[0049] Select the K nearest reference stations that satisfy the constraints to form a reference set: , ; Redundant lower bound constraints: ,in The lower limit is configurable; for example, the lower limit can be 3, 5, or 7. As a preferred embodiment, the lower limit is preferably 5 to balance robustness and computational complexity.

[0050] Pollution control and degradation output (executable criteria): Target station The number of effective station pairs formed on day d is After quality thresholding and outlier removal, the number of retained cells is: .

[0051] (1) If ,in, If the value is configurable and preferably not less than 3, the result of the target station on that day will be recorded as "unavailable or downgraded". (2) If the elimination ratio If (configurable), the "reference set contamination / anomaly" flag is triggered, and the system enters the delayed confirmation phase. During the delayed confirmation phase, the system continuously monitors the evolution of the anomaly state and does not perform domain replacement or reference feature sequence reconstruction. If the anomaly state disappears during the delayed confirmation phase, the system automatically clears the anomaly flag and resumes normal operation. (3) The system will only trigger the domain replacement process if the abnormal state persists beyond the delayed confirmation stage and causes the reference set to continuously fail to meet the redundancy lower limit requirement over a long time scale. The domain replacement process includes: marking the reference station that is determined to be a long-term abnormality as frozen and stopping its participation in the calculation; selecting a new reference station from the candidate reference station set according to preset spatial distance and distribution constraints; pre-constructing a corresponding reference feature sequence for the newly introduced reference station pair, the reference feature sequence being obtained by statistically superimposing historical intermediate analysis results over a long time scale; after the reference feature sequence is constructed, starting from the next update cycle, incorporating the new reference station into the subsequent clock error estimation and stability analysis process and resuming normal calculation.

[0052] The multi-component collaborative analysis method in this embodiment is used to collaboratively process different component information of seismic waveforms. The aim is to utilize complementary information from multiple components to form a more stable time delay estimate for station pairs, suppressing single-component anomalies or directional interference. By performing quality assessment and fusion of the analysis results for each component, the impact of single-component anomalies or noise interference on the overall judgment is reduced, thereby improving the stability of clock error identification results. More specifically, such as... Figure 3 As shown, for the target station With reference station and component combinations k (For example, a 3×3 combination, with a total of N=9; fewer combinations can also be configured), construct the station pair "target day feature sequence":

[0053] ; in The time delay operator used to estimate the relative time offset between a target station and a reference station is determined based on the following principle: within a preset analysis time window, the similarity of the two input component sequences is measured, and the time delay that causes the similarity to reach its extreme value is output as the relative time delay feature of the corresponding component combination. Under the premise of satisfying the above functional constraints, the operator... Specifically, it can be implemented as a time delay estimation operator based on cross-correlation function, a time delay estimation operator based on matched filtering or template matching, or other matching operators constructed based on similarity kernel function; the present invention does not limit the specific mathematical form of the operator, but limits its functional characteristic of determining the relative time offset by maximizing the similarity of the input sequence.

[0054] in, Here, d represents the component identifier, t represents the date, and t represents the cross-correlation delay time. k correspond( )combination.

[0055] Constructing reference feature sequences It can be obtained by long-term superposition and averaging (see (5) window and stability constraints below).

[0056] Define a similarity function (one of the normalized inner product forms): ; Within the preset search range Time delay estimation corresponding to maximum similarity: And define the quality index (similarity peak) for this component combination: ; Component consistency constraint: For the same station pair ( The set of components of ) Calculate the consistency metric: ; like (Configurable) If the result is marked as low reliability by the station on that day, its weight will be reduced or it will not participate in the fusion. Preferred .

[0057] For each station pair, the 3×3 components can generate N=9 sets (EE, EN, EZ, NE, NN, NZ, ZE, ZN, and ZZ) of daily relative time delays. This reflects the peak shift of the k-th cross-correlation function relative to the reference cross-correlation function. Then, a weighted average of the multi-component time-shift information is applied to enhance the stability and noise resistance of the estimation.

[0058] Define the multi-component joint time delay for each station pair and daily d as: ; And define the weighted quality indicators for station levels: ; Clock error estimation is used to generate continuous clock error time series for target stations based on multi-neighborhood redundancy constraints and multi-component collaborative analysis. It suppresses anomalous fluctuations and outputs estimation results reflecting clock error trends, providing fundamental data for subsequent identification and early warning.

[0059] The core objective of clock error estimation is to convert the relative time delay results obtained from the fusion of multiple station pairs and multiple components into a daily-scale clock offset estimate for target station i. , and output the uncertainty .

[0060] As Figure 4 shown, for the target station i , its reference set is . For each reference station , obtain the station pair result and , specifically as follows:

[0061] (1) Mass threshold rejection: If (configurable threshold), then reject the estimated value of this station for this day. , preferably , Too low usually means that the cross-correlation (or matching) on this day is unstable and noise dominates; setting Qmin at 0.2 - 0.4 can achieve a balance between rejecting obviously unreliable results and retaining enough valid station pairs.

[0062] (2) Robust outlier screening (MAD): Calculate the median for the retained set : ; Calculate MAD: ; Retain those that satisfy: ; where is a configurable parameter (preferably 2.5).

[0063] (3) Weighted fusion of multiple station pairs (weight is ): ； (4) Uncertainty output (weighted standard deviation): where, is used to characterize the dispersion / uncertainty of the estimate for this day and can be used as an important input for subsequent alarm confirmation and downgraded output.

[0064] (5) Window and stability constraints: The system uses a sliding superposition window with a length of W to generate the target day feature sequence or target day estimate. Therefore, when the abnormal duration D < W, the estimated amplitude may be significantly underestimated. For this reason:

[0065]  For suspected short-duration anomalies, the system uses delayed confirmation instead of immediate confirmation;  For slow linear drifts, use a longer confirmation window to reduce missed detections.

[0066] Automatic identification of clock error type Automatic clock error type identification is used to perform feature analysis on clock error time series. Based on the changing trend, duration, and amplitude characteristics, it automatically identifies and classifies the clock status of the station. This module can distinguish different types of abnormal clock states, providing more targeted reference information for station network operation and management.

[0067] The purpose of automatic clock error type identification is to... The morphology is automatically classified, and state labels such as "mutation / linear / nonlinear / composite / other" are output, along with the state confidence level. The specific method is as follows; Define daily increment: ; In the length of the sliding window Perform linear fitting within: ； And calculate the goodness of fit. .

[0068] State criteria: (1) Mutant (Step): If there exists a date d such that And in the following Ls days Maintaining the same sign offset and satisfying stability constraints ( If ), then it is determined to be a mutant type; among them, The threshold for mutation magnitude is 0.3-1.0s, preferably 0.5s; Ls is the duration in days, preferably 7-31 days, preferably 11 days. (2) Linear drift type: If the clock error sequence exhibits a stable linear change with time within a sliding window of continuous length LL days, and its linear fitting slope satisfies And linear fit goodness If so, it is determined to be a linear drift type; among them, the recommended continuous days LL is 11-31 days, preferably 14 days; The linear drift slope threshold (unit: s / day) is defined in the earthquake industry standard (DB / T 22-2020) as 0.0001% (approximately 1 ppm, or 0.0864 s / day), which represents the maximum permissible drift rate of the internal clock in non-clock-setting mode. In this embodiment... The value is 0.09 s / day; the linear fit goodness threshold. The recommended range is 0.7-0.9, with 0.8 being the preferred value.

[0069] (3) Nonlinear drift type: If the clock error sequence within a sliding window of continuous LL days has a monotonic overall trend within the window, but its linear fit is good. Below the threshold If so, it is determined to be a nonlinear drift; (4) Compound: If the above patterns alternate in different time periods (e.g., mutation followed by linearity, or multiple mutations superimposed), it is determined to be a compound.

[0070] (5) Other: When the clock error sequence does not meet the judgment conditions of mutation type, linear drift type, nonlinear drift type or composite type within the current analysis window, or although the abnormal pattern has been identified, the two-level early warning trigger conditions are not met, or the reliable consistency judgment cannot be performed due to insufficient number of effective stations or data quality not meeting the requirements, the system outputs an "uncertain" state and, in accordance with the downgraded output strategy, only gives a suspected or low confidence prompt and a review suggestion, without triggering a confirmation alarm.

[0071] Correspondence between status types and operation and maintenance handling principles Different status types are used to reflect the changing patterns of clock anomalies. The judgment results, combined with the early warning triggering mechanism, are used to guide network operation and maintenance as well as manual review. As a preferred implementation method, the operation and maintenance handling principles corresponding to each status type are as follows:

[0072] (1) Abrupt change: This indicates that the clock error has changed significantly in a short period of time. It is usually related to events such as time synchronization operation, time synchronization interruption or equipment restart. The system will trigger suspected or confirmed alarms first and prompt the operation and maintenance personnel to conduct key checks on the relevant time period data and time synchronization status.

[0073] (2) Linear drift type: This indicates that the clock error changes steadily and linearly over time, usually reflecting long-term performance changes or free drift of the timing module; the system mainly focuses on continuous monitoring and trend evaluation, and prompts maintenance personnel to perform planned maintenance or time calibration after the confirmation conditions are met.

[0074] (3) Nonlinear drift type: This indicates that the clock error changes have a clear trend but do not conform to stable linear characteristics. It may be related to intermittent time synchronization anomalies or environmental disturbances. The system prompts the operation and maintenance personnel to pay close attention to the operating status of the station and to verify it in conjunction with subsequent observation results.

[0075] (4) Composite type: This indicates that the clock error presents multiple abnormal patterns in different time periods; the system recommends that maintenance personnel conduct manual review and comprehensive analysis, and take action in conjunction with historical records of multiple time periods if necessary.

[0076] (5) Other (Uncertain): This indicates that the current data conditions or consistency are insufficient to support a reliable judgment; the system does not trigger a confirmation alarm, but only outputs a low confidence or suspected prompt, and continuously updates the analysis window. When the warning trigger conditions are met, the corresponding status is automatically upgraded.

[0077] Early warning triggering and result output mechanism As a preferred approach, the preset warning rules associated with different state types are configured with tiered triggering conditions, including at least a first-level warning triggering condition and a second-level warning triggering condition. When the absolute value of the current absolute clock offset estimate exceeds the first threshold and its uncertainty is lower than the second threshold, the first-level warning trigger condition is triggered. If the duration of the abnormal state reaches the minimum confirmation time corresponding to the state type after the first-level warning is triggered, the second-level warning trigger condition will be triggered.

[0078] In this embodiment, the warning and result output are used to compare the identified clock anomaly status with preset threshold conditions, and generate warning information when the trigger conditions are met. Warning information can be output through graphical interface display, log recording, or message push, facilitating timely acquisition and processing of relevant information by network maintenance personnel.

[0079] The purpose of early warning triggering and result output is to compare the identified abnormal state with the threshold conditions, output early warning information, and reduce false alarms through duration and consistency rules.

[0080] This embodiment uses a two-level warning trigger, as detailed below: (1) Level 1 Suspect Alarm: Satisfies: and in , This is a configurable threshold. The first threshold is... The first threshold is the abnormal amplitude threshold, used to filter candidate anomalies with practical operational significance. A suggested range is 0.3–1.0 s, preferably 0.5 s; the second threshold... The uncertainty threshold is used to constrain the reliability of anomaly estimation, and a recommended range is 0.2–0.6 s, preferably 0.3–0.5 s.

[0081] (2) Second-level confirmed alarm: meets any of the following conditions: a) Meeting the Level 1 suspected alarm condition for L consecutive days; b) Maintain the same sign offset within a continuous interval of length not less than W and satisfy the consistency of multiple station pairs (e.g.) ); c) The passing ratio of multi - neighborhood / multi - component consistency ≥ (configurable), where the passing ratio of consistency is defined as the ratio of the number of valid components satisfying the consistency constraint to the total number of valid components within the current analysis window, and at the same time, the proportion of valid station pairs reaches a preset threshold; the recommended value range is 0.6 - 0.9, preferably 0.7 - 0.8.

[0082] As another preferred embodiment, the minimum confirmation duration of this embodiment is not less than the length of the sliding time window used to generate the long - term reference feature sequence; for the linear drift - type state, the minimum confirmation duration is not less than twice the length of the sliding time window of the long - term reference feature sequence.

[0083] Since the system uses a sliding superposition window of length W to generate the target - day estimate, the system has the characteristic of smoothing or underestimating short - duration anomalies. To reduce missed detections and misjudgments:

[0084] When a suspected mutation is detected and the estimated duration D < W (where W is the length of the sliding superposition window, the recommended range is 5 - 15 days, preferably 7 - 11 days), it is not immediately confirmed, but enters a delayed confirmation process: it is required that the number of consecutive days L satisfying the first - level alarm ≥ W (preferably) before confirmation; When it is identified as a linear drift, a longer confirmation window is required (for example, L ≥ 2W is preferred) to offset the impact that short - term linear drift is more difficult to detect; When the number of valid station pairs is insufficient or the rejection ratio is too high, only the "suspected / low - confidence" status and recommended review information are output, and no strong alarm is output.

[0085] Output content and interface In this embodiment, the early - warning and result output content includes: abnormal station number, abnormal type, start - end time, sequence summary, uncertainty statistic, confidence level, and recommended operation and maintenance actions (review time synchronization, check recorder status, etc.). The output methods include graphical interface, log, and message push interface.

[0086] This embodiment also proposes a system for implementing the above - mentioned multi - neighborhood multi - component collaborative clock error early - warning method for a seismic network.

[0087] This system is deployed on a centralized data processing platform of a seismic network. Through unified processing of continuous waveform data from multiple stations, it generates clock error status determination results for target stations using multi-neighborhood redundancy constraints and multi-component collaborative judgment mechanisms, without relying on a single reference station. It then outputs early warning information when preset conditions are met. More specifically, the system in this embodiment includes the following functional modules: 1) Data access and preprocessing module; 2) Multi-neighborhood reference construction module; 3) Multi-component collaborative analysis module; 4) Clock offset estimation and stability constraint module; 5) Automatic clock error status identification module; and 6) Early warning triggering and result output module.

[0088] The above modules are connected sequentially or collaborate in parallel through data interfaces to form a complete automatic clock error identification and early warning processing chain.

[0089] During system operation, the data access and preprocessing module continuously receives continuous waveform data from seismic stations and outputs standardized analysis sequences and quality markers; the multi-neighborhood reference construction module constructs and maintains reference sets for the target station; the multi-component collaborative analysis module constructs feature sequences from the multi-component inputs of the target station and reference stations and calculates similarity to obtain station pair delay estimates and quality indices; the clock offset estimation and stability constraint module performs quality threshold control, MAD outlier removal, weighted fusion, and outputs uncertainty for the results of multi-station pairs; the clock error type automatic identification module performs state identification on the output sequence; when the identification result meets the preset triggering conditions, the early warning and result output module outputs early warning information and executes a downgraded output strategy under low confidence conditions.

[0090] Based on the same inventive concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0091] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method.

[0092] To further demonstrate the advantages of this invention, the inventors also conducted the following experiments: Neighborhood-scale robustness and pollution resistance test (3 / 5 / 7 neighborhoods) Experimental setup: For the same target station, reference sets with K=3, 5, and 7 were used; comparisons were made. Consistency Changes and sensitivity to anomalous station pairs. Comparison results are as follows: Figures 5-7 As shown, where, Figure 5 Clock error estimation for HE.TAG stations using 3 adjacent pairs; Figure 6Clock error estimation for HE.TAG stations using 5 adjacent pairs; Figure 7 Clock error estimation for HE.TAG stations using 7 adjacent pairs; based on Figures 5-7 It can be seen that the overall drift pattern remains consistent across different K values; the dispersion is larger when K=3; and the results are similar for K=5 and K=7. Considering computational efficiency, K=5 is the preferred value.

[0093] Contamination injection: Single contamination pair ( Figure 8 ) / Dual pollution pairs ( Figure 9 (Scenario 1) Conclusion: Single contamination pairs can be suppressed by MAD+ weighted suppression, while for dual contamination pairs, K≥5 is significantly better than K=3.

[0094] Window effect synthesis experiment (step vs. linear drift) Experimental setup: The window length W is fixed (W=11 days). The estimated link is injected with step drift and linear drift, and the drift amplitude and duration are combined for testing.

[0095] Experimental results show that when the duration is shorter than W, step jumps are easily underestimated or even undetectable; linear drifts with short durations are even more difficult to detect; therefore, the preferred rule for alarm confirmation is L≥W (step jump) and L≥2W (linear).

[0096] Test 1 (fixed step drift) involved artificially superimposing time shifts lasting 1–31 days with amplitudes of 0.5 s and 2.0 s. The results showed that:

[0097] (1) When the fixed drift duration of 0.5s approaches or exceeds the 11-day superposition window length, the detected time offset curve will exhibit a transitional characteristic similar to linear change during the beginning and end of the drift (Figure 10). This phenomenon is not caused by the actual drift process, but by the 11-day superposition averaging effect;

[0098] (2) When the drift amplitude equals the detection threshold (0.5 s), drifts lasting less than 11 days are significantly weakened by the averaging effect and are difficult to detect reliably, such as... Figure 10 As shown, Figure 10 Clock drift results for the HE.TAG station with a fixed drift of 0.5 seconds for different durations (1, 3, 5, 7, 11, 31 days). The graph shows that drift within a superimposed window of less than 11 days is underestimated.

[0099] (3) For large drifts (2.0 s), stable and consistent detection results can be obtained when the duration exceeds 4–5 days, such as Figure 11 As shown, Figure 11Clock drift results for the HE.TAG station when the 2-second fixed drift lasts for different durations (1, 3, 5, 7, 11, 31 days). A duration exceeding approximately 4-5 days is sufficient for stable detection, but the detected value will be underestimated.

[0100] Figure 12 The clock error detection results are shown for the HE.TAG station after applying a linear time drift that increases uniformly from 0 to 2 seconds over 5, 7, 11, and 31 days, respectively. Compared with fixed drift, linear drift requires a longer duration (≥7 days) to achieve stable detection, and the detection values ​​are also underestimated.

[0101] Computational complexity and scalability testing To evaluate the scalability of the method, configurations were compared where each station was paired with its 3, 5, and 7 nearest neighbor stations (the minimum distance between stations was greater than 30 km). Under these three configurations, the theoretical number of station pairs were 564, 940, and 1316 pairs, respectively; after removing duplicate pairs, the corresponding number of unique station pairs were 366, 579, and 791 pairs, respectively.

[0102] As a test example, after completing calculations of various data scales, performance testing was conducted on the most computationally intensive step—constructing the cross-correlation function from the raw data. The calculations were performed on a Linux virtual server equipped with a Hygon C86 7390 processor (24 cores, approximately 2.7 GHz) and 62 GB of memory, using 16 parallel worker processes for parallel computation of this step. The calculation results are shown in Table 1.

[0103] Table 1 Computational efficiency for different adjacent pair configurations Table 1 presents a comparison of computation time under the three adjacent pair configurations mentioned above. It can be seen that as the number of nearest neighbor stations selected for each station gradually increases from 3 to 7, the corresponding computation time also increases. Taking one year of single-component data as an example, when the number of unique station pairs increases from 366 to 791, the computation time increases from 10.1 hours to 21.7 hours, with the growth rate basically consistent with the growth rate of the number of station pairs. Similarly, with 5 years of data and 9 components processed together, the computation time increases from 18.9 days to 40.7 days, also showing an approximately linear correlation with the number of unique station pairs.

[0104] The results show that, under the condition of fixed data duration and number of components, the computational load is mainly controlled by the number of unique station pairs, and no significant superlinear growth was observed within the range of the tested adjacent pair configurations. This verifies that the proposed method has good scalability in the cross-correlation construction stage with the largest computational load, and is suitable for large-scale station network applications under different adjacent pair configurations and computational resource conditions.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of multi-neighborhood, multi-component collaborative clock errors in seismic networks, characterized in that, include: For the target station, multiple stations that meet the preset spatial constraints are selected from the seismic network as its neighborhood reference set. The continuous waveform data of the target station and each station in the reference set are standardized and preprocessed to generate a multi-component standardized analysis sequence for subsequent comparison. The target station and each reference station in the neighborhood reference set form a station pair. Based on the multi-component standardized analysis sequence of both stations, the relative clock offset of the station pair in the current analysis period and the corresponding estimated quality index are estimated by calculating the similarity with the long-term reference feature sequence. For the relative clock offsets of all station pairs, a weighted fusion is performed based on their estimated quality indices to finally generate the absolute clock offset estimate of the target station in the current analysis period and the uncertainty characterizing the dispersion of the estimate. A trend analysis is performed on the time series composed of the historical and current absolute clock offset estimates of the target station. Based on the preset morphological classification rules, the current state type of the clock error is automatically identified. The state type includes at least one of abrupt change, linear drift, and nonlinear drift. Based on the identified state type and according to the preset warning rules associated with different state types, the current absolute clock offset estimate, its uncertainty, and the duration of the state are comprehensively judged to trigger the output of warning information of different levels.

2. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 1, characterized in that, By calculating the similarity with the long-term reference characteristic sequence, the relative clock offset of the station pair in the current analysis period and the corresponding estimated quality indicators are estimated, including: For a pair of stations, calculate the preliminary time delay estimates and corresponding similarity peak values ​​for the combination of multiple seismic waveform components. A consistency check is performed on the preliminary time delay estimates of the multiple component combinations, and inconsistent component combination results are eliminated. Using the square of the peak similarity of each component combination that passed the test as the weight, the preliminary time delay estimate is weighted and averaged to obtain the relative clock offset and the estimated quality index of the station pair.

3. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 1, characterized in that, The preset morphological classification rules include: If there exists an analysis period in which the absolute value of the daily change of the absolute clock offset estimate is greater than the step threshold, and the offset remains the same for at least a preset number of consecutive periods thereafter, it is identified as a sudden change state. If, within a window of a predetermined number of consecutive analysis periods, the absolute value of the linear fitting slope of the absolute clock offset estimate sequence is greater than the drift slope threshold, and the goodness of fit is greater than the goodness of fit threshold, then it is identified as a linear drift state.

4. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 1, characterized in that, The preset warning rules associated with different state types are set with hierarchical triggering conditions, including at least a first-level warning triggering condition and a second-level warning triggering condition; When the absolute value of the current absolute clock offset estimate exceeds the first threshold and its uncertainty is lower than the second threshold, the first-level warning trigger condition is triggered. After a Level 1 warning is triggered, if the duration of the abnormal state reaches the minimum confirmation time corresponding to the state type, the conditions for triggering a Level 2 warning will be met.

5. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 4, characterized in that, The minimum confirmation time is not less than the length of the sliding time window used to generate the long-term reference feature sequence; for the linear drift state, the minimum confirmation time is not less than twice the length of the sliding time window of the long-term reference feature sequence.

6. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 1, characterized in that, The long-term reference feature sequence is constructed and updated in the following way: During the system initialization phase, the feature sequences of all quality-compliant days within a historical period are superimposed and averaged to generate the initial reference feature sequence. During system operation, the long-term reference feature sequence is updated using a sliding window method at an update cycle longer than the analysis cycle.

7. The seismic network multi-neighborhood multi-component collaborative clock error early warning method as described in claim 1, characterized in that, The standardization preprocessing includes robust amplitude normalization, which involves calculating the median of the absolute value of each component data in each analysis period as a robust scale, and then using this scale to normalize the data.

8. A multi-neighborhood, multi-component collaborative clock error early warning system for a seismic network, characterized in that, The system is deployed on the centralized data processing platform of the seismic network and includes: The system includes: a data preprocessing and access module for acquiring and standardizing waveform data from multiple stations; a multi-neighborhood reference construction module for building and maintaining neighborhood reference sets for each target station; a multi-component collaborative analysis module for performing multi-component time delay estimation between station pairs; a clock offset estimation and stability constraint module for generating clock offset estimates and uncertainties for target stations; an automatic clock error state identification module for automatically identifying the state of clock offset time series; and an early warning triggering and result output module for triggering tiered early warnings based on the identification results and preset rules. The aforementioned modules are configured to work together to perform the method of any one of claims 1 to 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images