A Machine Learning-Based Intensity Monitoring Method and System

By using machine learning methods for earthquake intensity monitoring, the problems of relying on human experience and non-closed zoning areas in existing technologies have been solved. This has enabled quantitative discrimination and automated document generation, improving the accuracy and efficiency of the assessment.

CN123115705APending Publication Date: 2026-11-17四川省地震应急服务中心
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611619558.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Current earthquake intensity assessments rely on human experience, lack quantitative standards, have incomplete regional area statistics, and result in inefficient and untraceable documentation.

Method used

A machine learning-based intensity monitoring method is adopted, which realizes fully automated intensity monitoring through confidence quantification, automatic separation of abnormal samples, cumulative deviation allocation and secondary verification, and generates multi-granularity documents.

Benefits of technology

It achieves quantitative determination of intensity, avoids errors from human experience, ensures closure of zoning areas, and improves the efficiency and traceability of the compilation of results documents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115705A_ABST
    Figure CN123115705A_ABST
Patent Text Reader

Abstract

This invention discloses a machine learning-based intensity monitoring method and system, relating to the field of earthquake disaster prevention and mitigation information technology. The method includes: collecting a basic set of earthquake event parameters, a set of regional records, a set of survey point records, a set of instrument intensity records, and the total area of ​​the entire region; after data preprocessing and feature encoding, inputting hyperparameters to an intensity discrimination model optimized by an improved Golden Cow algorithm, outputting the confidence distribution and scores of each survey point; separating the effective sample set and the abnormal sample set according to a differential threshold; performing area grouping and summarizing and five-level stability sorting based on the effective sample set; identifying the cumulative deviation of the ordered regional area set and allocating it according to the regional area ratio, performing secondary verification after abnormal sub-area splitting, and outputting the balanced and verified regional area set and the area of ​​the abnormal area; and adaptively generating a multi-granularity standardized intensity description document. This invention achieves full-process automation of intensity discrimination quantification, area closure verification, and automatic generation of result documents.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of earthquake disaster prevention and mitigation information technology, and in particular to an intensity monitoring method and system based on machine learning. Background Technology

[0002] Seismic intensity is a physical quantity that characterizes the strength of an earthquake's impact on the Earth's surface and engineering structures at a specific location. It is a core basis for rapid earthquake disaster assessment, emergency response deployment, and post-disaster recovery decision-making. Current intensity assessment and result presentation methods mainly suffer from the following problems: First, intensity assessment relies on human experience. The intensity assessment at the field investigation points depends on experts' manual judgment based on factors such as damage phenomena and site environment. The assessment criteria vary from person to person, and the assessment results lack quantitative confidence measures, making it impossible to automatically screen suspicious assessment results.

[0003] Secondly, the statistical caliber of intensity zoning areas is not closed. During the measurement and hierarchical aggregation of intensity map zoning areas, the sum of zoning areas often deviates from the total area of ​​the entire region due to the rounding and differences in registration caliber between upstream data sources. There is also a risk of inconsistency between the registration caliber of abnormal areas corresponding to abnormal survey points and regular zoning areas, and the existing technology lacks a systematic balancing verification mechanism.

[0004] Third, the compilation of the results documents is inefficient. The intensity description document needs to organize the data into zones at multiple levels of granularity, such as intensity level, province, city, county, and township. Manual compilation is time-consuming and prone to errors, and the data references are not traceable, resulting in inconsistent results from multiple compilations.

[0005] Therefore, there is an urgent need for an automated intensity monitoring method that allows for quantifiable judgments, area balancing, and traceable results. Summary of the Invention

[0006] This invention provides a machine learning-based intensity monitoring method and system. Through confidence quantification, automatic separation of abnormal samples, cumulative deviation proportion allocation and secondary verification, automatic generation of multi-granularity documents and numerical consistency verification, the system achieves fully automated production of intensity monitoring results and ensures numerical determinism.

[0007] In a first aspect, embodiments of the present invention provide a machine learning-based intensity monitoring method, the method comprising: Obtain the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset; Perform data preprocessing on the original dataset to output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; The feature vector set is input into the intensity discrimination model, which outputs the discrimination confidence distribution of each survey point at each intensity level, and the maximum confidence score is taken as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; Based on the confidence score and the preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an abnormal sample set. Based on the effective sample set, the area grouping and summarization and five-level stable sorting are performed on the partition record set to output an ordered set of partition areas; The summation of the areas of each partition with an intensity value higher than the preset lower limit in the ordered partition area set is compared with the total area of ​​the entire domain to identify the cumulative deviation and eliminate it according to the partition area ratio; the abnormal area obtained by performing spatial aggregation on the abnormal sample set is split into abnormal sub-areas falling into each cross-partition according to the partition affiliation, and the regular partition net area after deducting the abnormal sub-areas on the partition side is compared with the benchmark of the net area of ​​the entire domain after deducting the registered area of ​​the abnormal area on the entire domain side to perform a second verification, and output the partition area set and abnormal area after balancing verification; The balanced and verified set of zonal areas, the area of ​​the abnormal zone, the statistical results of the confidence score, and the set of basic parameters are input into the multi-granularity document generation module. Based on the magnitude of the earthquake in the set of basic parameters and the number of county-level administrative regions involved, the upper limit of the number of township-level listings is adaptively determined, a standardized intensity description document is generated, exported, and automatically opened and presented.

[0008] The technical solution provided in this application has at least the following beneficial effects: Intensity discrimination is transformed from manual experience to quantitative confidence output, and the confidence threshold is set differently according to the actual discrimination accuracy of each intensity level. Abnormal samples are automatically separated and independently organized to avoid local intensity distortion caused by site effects from contaminating the zonal statistics. A three-level area protection mechanism of "source suppression - end leveling - caliber verification" is formed by high-precision decimal accumulation, cumulative deviation ratio amortization and secondary verification based on abnormal sub-area splitting to ensure that the zonal area system and the total area of ​​the whole domain are strictly closed. Five-level stable sorting, inference result caching and document numerical consistency verification jointly ensure that the output results are consistent when running the same dataset multiple times. Each value in the document has a data source label for traceability. The improved Golden Jackal optimization algorithm improves the global search capability and convergence accuracy of hyperparameter optimization through chaotic initialization, dynamic refraction back learning, three-layer collaborative position update and golden sine perturbation.

[0009] In one optional implementation, the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the seismic event are obtained, and the raw dataset is output, including: Read the epicenter records from the epicenter data table and extract the fields of earthquake time, epicenter location, epicenter latitude and longitude, magnitude, major axis direction of isoseismal lines, major and minor axis lengths, and surface rupture zone length. Fill missing fields with standard placeholders and output the basic parameter set. Read the intensity zoning records from the intensity map statistics table. Each intensity zoning record should contain at least the fields of intensity level, zoning area, province, city, county, township name and township area. Output the zoning record set. Read the survey point records from the survey point statistics table. Each survey point record should contain at least the fields of survey point number, latitude and longitude, name of the fourth-level administrative region, on-site intensity assessment, description of the damage phenomenon and site environment category. Output the survey point record set. Read the instrument intensity records from the instrument intensity statistics table, group and count them according to the preset standard intensity interval, and output the instrument intensity record set; Read the total area of ​​the earthquake zone from the global basic geographic data table. The total area is the sum of the land areas within the scope of this earthquake intensity assessment. Output the total area. Perform cross-table consistency checks on the basic parameter set, the partition record set, the survey point record set, the instrument intensity record set, and the total area of ​​the entire region. Construct a regional affiliation index with the survey point number as the primary key and the name of the fourth-level administrative region as the foreign key. Construct a spatial association index with the latitude and longitude of the survey points and the latitude and longitude of the epicenter. Output the original dataset with the association index.

[0010] In one optional implementation, data preprocessing is performed on the original dataset to output a normalized dataset, and feature encoding is performed on the survey point data in the normalized dataset to construct a feature vector set, including: Perform format validation on each field of the original dataset, validate the numeric type and valid range for numeric fields, mark records that exceed the range as invalid records, and fill missing fields with standard placeholders. Output the format-validated intermediate dataset, the list of invalid records, and the records filled with placeholders. Perform whitespace stripping, invisible character removal, and garbled character replacement on the intermediate dataset after format verification, and output the cleaned intermediate dataset; The cleaned intermediate dataset is normalized by administrative region names. Duplicate provincial and municipal district names are removed from the municipalities directly under the central government. The county-level divisions under the provincial government are attached according to the hierarchical relationship between the province and the county-level city. Standardized labels are generated for towns, townships and streets at the township level. The normalized intermediate dataset is then output. For the normalized intermediate dataset, duplicate detection is performed on the survey point records using the combination of survey point number and latitude and longitude as the key, and duplicate deduplication is performed on the partition records using the combination of intensity level, fourth-level administrative region name and township name as the key, eliminating duplicate registration of the same township between adjacent intensity partitions, and outputting a normalized dataset, effective sample size and duplicate record list. For each survey point record in the standardized dataset, feature encoding is performed to construct a feature vector that includes at least epicentral distance features, magnitude features, site shear wave velocity features, instrument intensity features, damage phenomenon codes, site environment category codes, and administrative division codes, forming a feature vector set and a corresponding intensity label vector; wherein, for survey points without matching instrument intensity records but which need to be estimated and filled, the estimated intensity calculated by the empirical attenuation formula is used to fill the estimated intensity and an estimated label is added. The invalid record list, the placeholder-filled records, the duplicate record list, and the estimated markers are summarized to generate a data quality report that includes the original sample size, the number of removed samples, the number of valid samples, and the distribution of removal reasons at each stage.

[0011] In one optional implementation, the training method of the intensity discrimination model includes: The feature vector set and the corresponding intensity label vector are divided into a training set, a validation set and a test set according to a preset ratio; The sample size of each intensity level in the statistical training set is calculated. For intensity levels with a sample size lower than the preset lower limit, the relationship between epicentral distance and intensity attenuation is used as a priori. Interpolation and oversampling are performed on the samples of the intensity level with a sample size lower than the preset lower limit to generate synthetic samples and add synthetic labels. The enhanced training set is then output. Using the hyperparameters of the machine learning model as decision variables, and the weighted combination of the intensity prediction bias and partition assignment cross-entropy of the candidate model trained on the enhanced training set on the validation set as the fitness function, the improved Golden Jackal optimization algorithm is executed to iteratively find the optimal hyperparameter vector. The model structure is instantiated using the optimal hyperparameter vector, and an intensity discrimination model is fitted on the enhanced training set. The intensity discrimination accuracy and partition assignment accuracy are evaluated on the test set, and a model evaluation report is output.

[0012] In one optional implementation, the iterative optimization of the improved golden jackal optimization algorithm includes: The decision variables are encoded as position vectors of the improved Golden Jackal optimization algorithm, and a chaotic sequence is generated by the Logistic mapping and mapped to the hyperparameter search space to generate candidate initial populations. For each candidate individual in the candidate initial population, a refracted reverse individual is generated with a dynamic scale factor that decays with iteration. The refracted reverse individual with the better fitness value among the original candidate individuals is retained to form the initial population. Based on the fitness function, calculate and sort the fitness values ​​of all individuals in the initial population or the population updated in the last iteration, and select the top three individuals as the first leader, the second leader, and the third leader, respectively. In each iteration, the first leader, the second leader, and the third leader estimate the prey's location. Combining the prey escape mechanism and the gray wolf cooperation mechanism, they enter the exploration or development phase. The three-layer cooperative location update is performed with the prey's location as the convergence guide to obtain the temporary individual location. The position of the temporary individual is perturbed by a combination of sine and cosine using the golden ratio constant to obtain the perturbed position; The refracted inverse solution is generated for the perturbed position using the current iterative dynamic scale factor. The refracted inverse solution with the better fitness value among the perturbed positions is retained as the new generation position of the temporary individual, thus obtaining an updated population. The fitness values ​​of all individuals in the updated population are recalculated and the first leader, the second leader, and the third leader are updated. The results are then fed back to the prey location estimation stage to form an iterative closed loop. The iteration terminates when the maximum number of iterations is reached or the change in the optimal fitness value for a consecutive preset number of generations is less than the convergence threshold. The first leader individual is output as the global optimal individual, and the global optimal individual is decoded to obtain the optimal hyperparameter vector.

[0013] In one optional implementation, the feature vector set is input into the intensity discrimination model, which outputs the discrimination confidence distribution of each survey point at each intensity level, and takes the maximum confidence score as the confidence score of that survey point, including: Verify that the intensity discrimination model is loaded and the version tag is consistent with the current configuration; verify that the sample size of the feature vector set is consistent with the effective sample size output by the data preprocessing stage, and that the dimension of each feature vector is consistent with the preset feature definition; if any verification fails, terminate the inference and output the corresponding error message. After the verification is passed, the feature vector set of all survey points is constructed as an inference input matrix and input into the intensity discrimination model in batches at one time. The original output of the model is normalized to obtain the discrimination confidence distribution of each survey point at each intensity level after normalization, which sums to 1. For each survey point, the maximum value in the discrimination confidence distribution is extracted as the confidence score of that survey point, the intensity level corresponding to the maximum value is recorded as the discrimination intensity, and the second highest confidence score is extracted to calculate the difference between the confidence score and the second highest confidence score as the discrimination divergence degree.

[0014] In one optional implementation, based on the confidence score and a preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an outlier sample set, including: Based on the discrimination accuracy rate of each intensity level in the model evaluation report, a confidence threshold is configured for each intensity level. Intensity levels with a discrimination accuracy rate not lower than the average discrimination accuracy rate of each intensity level correspond to a confidence threshold not lower than a preset benchmark threshold. Intensity levels with a discrimination accuracy rate lower than the average value correspond to a confidence threshold lower than the preset benchmark threshold, thus forming a differentiated threshold configuration table. For each survey point, the corresponding threshold is retrieved from the differential threshold configuration table using the discrimination intensity level of the survey point as the index key. The confidence score of the survey point is compared with the corresponding threshold, and all survey points in the feature vector set are separated into a valid sample set and an abnormal sample set. The sum of the sample sizes of the valid sample set and the abnormal sample set is compared with the valid sample size output from the data preprocessing stage. If the comparison is successful, the partition area aggregation is initiated. If the comparison fails, the process is terminated and a location prompt for the uncovered survey points is output. Summarize the differentiated threshold configuration table, each anomaly separation marker, and the quantity verification results to generate anomaly separation files; perform spatial aggregation on the anomaly sample set, and based on the administrative division association index and the spatial association index, aggregate anomaly survey points that are spatially less than the preset aggregation radius and belong to the same county-level administrative region into the same anomaly area, count the number of anomaly points within the area, calculate the registered area of ​​the anomaly area and record the background intensity level, and output the anomaly area list and the anomaly area area; when the anomaly sample set is empty, output an empty anomaly area marker, and the anomaly area is set to zero.

[0015] In one optional implementation, based on the valid sample set, area grouping and summarizing and a five-level stable sort are performed on the partitioned record set to output an ordered set of partition areas, including: The verification of the sum of the sample sizes of the valid sample set and the abnormal sample set has been passed, and the verification of each survey point in the valid sample set can be matched with at least one intensity partition record in the partition record set according to the zoning association index. The survey points that cannot be matched are marked as isolated points. For each intensity zone, the number of valid survey points falling into that zone from the valid sample set is counted to obtain a zone coverage count table, and zero-coverage zones with a coverage count of zero are identified. Using the intensity level as the first-level grouping key and the names of the four-level administrative regions (province, city, county, and township) as the hierarchical grouping keys, the partition record set is grouped and summarized. Multiple partition records of the same intensity level and the same township are merged and accumulated, and high-precision decimal arithmetic is used to perform surface accumulation to obtain the total area of ​​each intensity level and township combination, as well as the total area of ​​the county, city, and province at each level upwards. A stable sorting algorithm is used to perform a five-level stable sort on the grouped summary results, outputting an ordered set of partition areas and a sorting process archive. The partition coverage count table is output along with the ordered set of partition areas to the balancing verification stage and the document generation stage. The list of zero-coverage partitions is output to the balancing verification stage to perform upper limit constraints on the area of ​​zero-coverage partitions and is output to the document generation stage for zero-coverage prompts.

[0016] In one optional implementation, the areas of each partition in the ordered partition area set whose intensity value is higher than a preset lower limit are summed and compared with the total area of ​​the entire region. Accumulated deviations are identified and eliminated by apportioning them according to the partition area ratio. After spatial aggregation of the abnormal sample set is performed, the abnormal areas are split into abnormal sub-areas falling into each cross-boundary partition according to their partition affiliation. The regular partition net area (after deducting the abnormal sub-areas) is then compared with the baseline of the total net area (after deducting the registered area of ​​the abnormal areas) on the entire region side for secondary verification. The result is the output of the balanced and verified partition area set and the abnormal area area, including: Confirm the range of zones to be included in the summation, and only include zones with intensity values ​​higher than the preset lower limit level. The area of ​​zones with intensity values ​​not higher than the preset lower limit level is separately recorded as the area of ​​the unincluded zone. Establish a calculation identity as the calculation basis for the first verification. The areas of each partition involved in the summation are summed, and the first verification difference between the total area of ​​the whole region minus the area of ​​the non-included area and the sum of the areas of the partitions is calculated. If the absolute value of the first verification difference is not greater than the first error tolerance, it is determined that no fine-tuning is required. If it is greater than the first error tolerance, it is determined that there is a cumulative deviation of the partition area system relative to the whole region benchmark. In cases where there is a cumulative deviation, the first verification difference is distributed to each partition according to the proportion of each partition area to the total area of ​​the partitions, thus obtaining the fine-tuned partition area. Each anomalous area is divided into anomalous sub-areas falling into each intensity zone it crosses, according to its zoning affiliation. The sum of the anomalous sub-areas of each anomalous area is not greater than the registered area of ​​the anomalous area, and no anomalous sub-area is greater than the fine-tuned area of ​​its respective zone. The anomalous sub-areas falling into each zone are subtracted from the anomalous area in the fine-tuned area, and then summed to obtain the net area of ​​the regular zones. The registered area of ​​the anomalous area is subtracted from the total area of ​​the entire region to obtain the net area benchmark of the entire region. The second-order verification difference between the net area benchmark of the entire region (after deducting the area of ​​the unincluded area) and the net area of ​​the regular zones is calculated. The second-order verification difference consists of a primary verification residual term and a caliber term, where the caliber term is the difference between the sum of the anomalous sub-areas of each anomalous area and the registered area of ​​the anomalous area. If the absolute value of the caliber item is not greater than the second error tolerance, the secondary verification is deemed to have passed, and the set of partition areas composed of the fine-tuned partition areas is confirmed as the set of partition areas after balancing verification. If it is greater than the second error tolerance, it is determined that the area of ​​the abnormal area is inconsistent with the registration caliber of the partition area system, and the suspected conflicting abnormal area number and its cross-partition list are output. The suspected conflicting abnormal area number and its cross-partition list are fed back to the abnormal sample set spatial aggregation stage. The abnormal sample set spatial aggregation stage re-splits the area of ​​the suspected conflicting abnormal area according to the partition affiliation and then returns to execute the secondary verification. The re-split verification of the same abnormal area is performed for a maximum of a preset number of rounds. If it still fails after exceeding the preset number of rounds, the process is terminated and a balancing failure prompt and a manual verification list are output. When the abnormal sample set is empty, the area of ​​the abnormal region is zero, the net area of ​​the regular partition is equal to the sum of the areas of the partitions after fine-tuning, and the secondary verification degenerates into a re-verification of the result of the primary verification. Summarize the first verification difference and its relative magnitude, fine-tuning details table, number of fine-tuning rounds, and the second verification difference to generate a balance verification file, which is then attached to the appendix of the standardized intensity description document according to the configuration. When the fine-tuning amount of any partition area in the fine-tuning details table causes its sorting key to cross the adjacent partitions at the same level, a sequential review prompt is triggered. When the relative magnitude of the single-verification difference of a consecutive preset number of earthquake cases exceeds the preset error limit, a data source management prompt indicating that there is a systematic accuracy problem in the upstream partition area data source is output.

[0017] Secondly, embodiments of the present invention provide a machine learning-based intensity monitoring system for implementing a machine learning-based intensity monitoring method, the system comprising: The data acquisition unit is used to acquire the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset. The data preprocessing unit is used to perform data preprocessing on the original dataset, output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; The machine learning intensity discrimination unit is used to input the feature vector set into the intensity discrimination model, output the discrimination confidence distribution of each survey point at each intensity level, and take the maximum confidence as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; The sample separation unit is used to separate all survey points corresponding to the feature vector set into a valid sample set and an abnormal sample set based on the confidence score and a preset confidence threshold. The partition area summarization and sorting unit is used to perform area grouping and summarization and five-level stable sorting on the partition record set based on the effective sample set, and output an ordered partition area set; The area balancing verification unit is used to sum the areas of each partition in the ordered partition area set whose intensity value is higher than the preset lower limit level, compare it with the total area of ​​the whole domain, identify the cumulative deviation and eliminate it according to the partition area ratio; after splitting the abnormal area obtained by performing spatial aggregation on the abnormal sample set into abnormal sub-areas falling into each cross-partition according to the partition affiliation, the regular partition net area after deducting the abnormal sub-areas on the partition side is compared with the whole domain net area benchmark after deducting the registered area of ​​the abnormal area on the whole domain side to perform a second verification, and output the partition area set and abnormal area area after balancing verification; The document generation unit is used to input the balanced and verified zonal area set, the abnormal area, the statistical results of the confidence score, and the basic parameter set into the multi-granularity document generation module. Based on the magnitude in the basic parameter set and the number of county-level administrative regions involved, it adaptively determines the upper limit of the number of township-level listings, generates a standardized intensity description document, exports it, and automatically opens and presents it.

[0018] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by at least one processor, which are executed by at least one processor to enable the at least one processor to perform the method proposed in the first aspect of the embodiments of the present invention.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0020] Figure 1This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart of the steps of an intensity monitoring method based on machine learning provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the functional units of an intensity monitoring system based on machine learning provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. 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 embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] The present invention will be further described below with reference to the accompanying drawings.

[0023] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0024] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0025] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0026] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for a machine learning-based intensity monitoring system.

[0027] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data interaction with the data server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the electronic program of the intensity monitoring system based on machine learning stored in the memory 1005 through the processor 1001 and executes the intensity monitoring method based on machine learning provided in the embodiment of the present invention.

[0028] Reference Figure 2 The present invention provides a machine learning-based intensity monitoring method, the method comprising: S201: Obtain the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset; S202: Perform data preprocessing on the original dataset to output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; S203: Input the feature vector set into the intensity discrimination model, output the discrimination confidence distribution of each survey point at each intensity level, and take the maximum confidence as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; S204: Based on the confidence score and the preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an abnormal sample set; S205: Based on the effective sample set, perform area grouping and summarization and five-level stable sorting on the partitioned record set, and output an ordered partitioned area set; wherein, the five-level stable sorting includes sorting in descending order by intensity level, and within the same intensity level, sorting in descending order by the corresponding area of ​​province, city, county, and township. S206: Sum the areas of each partition in the ordered partition area set whose intensity value is higher than the preset lower limit level, compare it with the total area of ​​the whole domain, identify the cumulative deviation and eliminate it according to the partition area ratio; after splitting the abnormal area obtained by performing spatial aggregation on the abnormal sample set into abnormal sub-areas falling into each cross-partition according to the partition affiliation, compare the regular partition net area after deducting the abnormal sub-areas on the partition side with the whole domain net area benchmark after deducting the registered area of ​​the abnormal area on the whole domain side to perform a second verification, and output the partition area set and abnormal area after balancing verification; S207: Input the balanced and verified set of zonal areas, the area of ​​the abnormal zone, the statistical results of the confidence score, and the set of basic parameters into the multi-granularity document generation module. Based on the magnitude in the set of basic parameters and the number of county-level administrative regions involved, adaptively determine the upper limit of the number of township-level listings, generate a standardized intensity description document, export it, and automatically open and present it.

[0029] In one optional implementation, the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the seismic event are obtained, and the raw dataset is output, including: S2011: Read the epicenter records from the epicenter data table and extract the fields of earthquake time, epicenter location, epicenter latitude and longitude, magnitude, major axis direction of isoseismal lines, major and minor axis lengths, and surface rupture zone length. Fill missing fields with standard placeholders and output the basic parameter set. Among them, the epicenter latitude and longitude and the magnitude are used to construct the epicentral distance feature and magnitude feature in subsequent feature encoding. The magnitude is used for the differential setting of confidence threshold and the adaptive determination of the upper limit of the number of township-level listings. The major axis direction of isoseismal lines, major and minor axis lengths, and surface rupture zone lengths are used to generate the description of the earthquake situation overview section in the document generation process. S2012: Read intensity zoning records from the intensity map statistics table. Each intensity zoning record should contain at least the fields of intensity level, zoning area, province, city, county, township name, and township area. Output the zoning record set. The intensity level and township area are used as sorting keys for subsequent area grouping and multi-level stable sorting. The area of ​​each zoning zone and the total area of ​​the whole region are used as the summation objects and comparison benchmarks for subsequent balancing verification. S2013: Read the survey point records from the survey point statistics table. Each survey point record contains at least the survey point number, latitude and longitude, fourth-level administrative region name, on-site intensity assessment, description of damage phenomenon, and site environment category. Output the survey point record set. The on-site intensity assessment, description of damage phenomenon, and site environment category are encoded to form feature vector components and used as model training labels. The fourth-level administrative region name is used for subsequent zoning matching of survey points and intensity zones. S2014: Read instrument intensity records from the instrument intensity statistics table and count them in groups according to preset standard intensity intervals (e.g., with an interval width of 0.1 or 0.5), and output the instrument intensity record set; wherein, the instrument intensity value of each survey point is used to construct the instrument intensity component of the feature vector, and the grouping and counting results of each interval are used to generate the instrument intensity distribution description paragraph in the standardized intensity description document, and the document generation process skips the instrument intensity distribution description paragraph when the instrument intensity record set is empty; S2015: Read the total area of ​​the earthquake zone from the global basic geographic data table. The total area is the sum of all land areas within the scope of this earthquake intensity assessment. Output the total area as the global comparison benchmark for the balancing and verification process. S2016: Perform cross-table consistency checks on the basic parameter set, the partition record set, the survey point record set, the instrument intensity record set, and the total area of ​​the entire region. The cross-table consistency checks include record number conservation checks, survey point number primary key uniqueness checks, and fourth-level administrative region name foreign key matching checks. Construct a regional affiliation index with the survey point number as the primary key and the fourth-level administrative region name as the foreign key, and construct a spatial association index with the survey point latitude and longitude and the epicenter latitude and longitude. Output the original dataset with the association index.

[0030] In one optional implementation, data preprocessing is performed on the original dataset to output a normalized dataset, and feature encoding is performed on the survey point data in the normalized dataset to construct a feature vector set, including: S2021: Perform format validation on each field of the original dataset, validate the numerical type and valid range of numeric fields, mark records that exceed the range as invalid records, and fill missing fields with standard placeholders. Output the format-validated intermediate dataset, the list of invalid records, and the records filled with placeholders; wherein, the longitude validation range is 73°E~135°E, and the latitude validation range is 3°N~54°N. S2022: Perform whitespace stripping, invisible character removal, and garbled character replacement on the intermediate dataset after format verification, and output the cleaned intermediate dataset; S2023: Perform administrative region name normalization processing on the cleaned intermediate dataset. Remove duplicate provincial and municipal district names from the municipalities directly under the central government. Assign the county-level divisions under the provincial government according to the hierarchical relationship between the province and the county-level city. Generate standardized labels for towns, townships, and streets at the township level. Output the normalized intermediate dataset. The normalized four-level administrative region names serve as the comparison key for deduplication of districts with the same name, the basis for generating administrative region codes, and the name key for subsequent multi-level stable sorting, ensuring the uniqueness of districts with the same name in the data preprocessing, feature encoding, and sorting stages. S2024: For the normalized intermediate dataset, perform duplicate detection on the survey point records using the combination of survey point number and latitude / longitude as the key, and perform duplicate removal on the partition records using the combination of intensity level, fourth-level administrative region name, and township name to eliminate duplicate registrations of the same township between adjacent intensity partitions. Output a normalized dataset, effective sample size, and a list of duplicate records; the effective sample size serves as the basis for subsequent sample partitioning and is checked against the sum of the sample sizes of the subsequently separated effective sample set and outlier sample set. S2025: Perform feature encoding on each survey point record in the standardized dataset to construct a feature vector containing at least the epicentral distance feature, magnitude feature, site shear wave velocity feature, instrumental intensity feature, damage phenomenon code, site environment category code, and administrative division code, forming a feature vector set and a corresponding intensity label vector; wherein, the epicentral distance feature is calculated from the survey point latitude and longitude and the epicenter latitude and longitude using an ellipsoidal model, the site shear wave velocity feature is obtained from the site environment category through a preset lookup table, and the instrumental intensity feature is obtained by matching the nearest instrumental intensity record, and when no matching record is found, it is filled with a standard missing placeholder and a missing marker is added, without using estimated values ​​derived from magnitude and epicentral distance to fill in, so as to avoid the instrumental intensity feature being completely collinear with the existing magnitude and epicentral distance features; when estimation filling is necessary, it uses... The estimated intensity calculated using the empirical attenuation formula is filled in and an estimated label is added. At this point, partial collinearity is accepted. This is mitigated by retaining the true values ​​of the magnitude and epicentral distance features of the sample and reducing the weight of the sample in the fitness assessment using the estimated label. The feature vector set is output to the model inference stage as the discriminant input, and the intensity label vector is output to the model training stage and the hyperparameter optimization stage as the supervision label. The true distribution of the zoning assignment is constructed by matching the intensity zoning to which the survey points belong through the zoning assignment association index. If the zoning is assigned, it is 1; otherwise, it is 0. This serves as the supervision label for the zoning assignment cross-entropy. The intensity discrimination model adopts a dual-task output structure. The first output head outputs the discrimination confidence distribution of each intensity level, and the second output head outputs the assignment probability of each intensity zoning. The two output heads share the main feature extraction and are trained with joint loss. Epicentral distance characteristics: Geodesic distances are calculated primarily using the ellipsoidal model (Vincenty inverse equation, WGS-84 ellipsoid, semi-major axis a = 6378137 m, flattening f = 1 / 298.257223563), whose simplified spherical form is the Haversine formula: In the formula, The epicentral distance (km) from the l-th survey point to the epicenter; l is the index of the survey point; R is the average radius of the Earth, which is taken as R=6371.0088 km in this embodiment; The latitude (in radians) of the l-th survey point; The longitude (in radians) of the l-th survey point; The latitude (in radians) of the epicenter; The longitude (in radians) of the epicenter; Instrument intensity estimation fill: This is only enabled when estimation fill is absolutely necessary and the collinearity constraint has been handled as described above. When there is no matching instrument intensity record for the survey point, the estimated intensity is calculated using an empirical attenuation formula. In the formula, The estimated instrument intensity for the l-th survey point; The magnitude of the earthquake; All regression coefficients are derived from measured magnitude, epicentral distance, and instrument intensity data of historical earthquakes, calibrated using least-squares regression. This is a standard calibration method in this field (example magnitude:). ≈ 1.5, ≈ -2~-3, (≈ 4~6, for example only); the regional seismic motion attenuation relationship is obtained by calibrating historical earthquake cases based on a regional intensity attenuation statistical model (such as a major axis elliptical intensity attenuation model); Here, the near-field saturation constant is taken as [value missing in original text]. =25 km; Site shear wave velocity mapping: Site shear wave velocity characteristics are obtained by mapping from site environment categories through a preset lookup table. An example lookup table is: Hard Site ≥500 m / s, medium hard surfaces 270~500 m / s, soft surfaces <270 m / s; where, The equivalent shear wave velocity of the site; Encoding of destructive phenomena: The encoding of destructive phenomena uses a preset dictionary of destructive phenomena categories to perform word segmentation and category mapping on the description of the destructive phenomena. Example dictionary entries are: building collapse: category 1, wall cracking: category 2, ground cracks: category 3, landslide collapse: category 4, object reaction: category 5, to obtain the count vector of each category of destructive phenomena, or to encode the description of the destructive phenomena into a fixed-length vector using a statistical word frequency weighting method. S2026: Summarize the invalid record list, the placeholder-filled records, the duplicate record list, and the estimated markers to generate a data quality report containing the original sample size, the removed sample size, the effective sample size, and the distribution of removal reasons at each stage; the effective sample size in the data quality report is used to perform quantity verification on the separation results of the effective sample set and the abnormal sample set, and serves as the verification basis for the statistical description of the survey points in the document generation stage.

[0031] In one optional implementation, the training method of the intensity discrimination model includes: S001: Divide the feature vector set and the corresponding intensity label vector into training set, validation set and test set according to a preset ratio (example: training set: validation set: test set = 6:2:2). When dividing, perform stratified sampling according to intensity level to make the proportion of samples of each intensity level in the three subsets consistent. S002: Calculate the sample size of each intensity level in the training set. For intensity levels with a sample size below a preset lower limit, using the attenuation relationship between epicentral distance and intensity as a priori, perform interpolation oversampling on the samples of these intensity levels to generate synthetic samples and attach synthetic labels, outputting an enhanced training set. The synthetic labels are used to apply a lower weight to the synthetic samples than to the real samples in the fitness evaluation of subsequent hyperparameter optimization (example: real sample weight 1.0, synthetic sample weight 0.5). Simultaneously, according to a preset retention ratio (example: 10%), the synthetic samples with attached synthetic labels are synchronously injected into the validation set to evaluate the robustness of the candidate model to the synthetic samples during hyperparameter optimization and to support the differentiated sample weights in the fitness function. S003: Using the hyperparameters of the machine learning model (e.g., maximum depth of the tree model, learning rate, minimum number of leaf samples, subsampling rate) as decision variables, and the fitness function of the weighted combination of the intensity prediction bias and partition assignment cross-entropy of the candidate model trained on the augmented training set on the validation set, an improved Golden Jackal optimization algorithm is executed to iteratively optimize and output the optimal hyperparameter vector; the machine learning model can be a gradient boosting tree, random forest, multilayer perceptron, or other model with multi-classification output capability (gradient boosting tree is used as an example in this example). The formula for the fitness function is: In the formula, The fitness function; For decision variables; Let j be the true intensity of the j-th validation sample; j is the validation sample index. For decision variables The candidate model predicted intensity is output by the model constructed from the corresponding hyperparameter vectors; The sample weights for the j-th validation sample are: (1.0 for real samples, 0.5 for synthetic samples, and 0.8 for estimated padding samples with additional estimation labels); when the validation set does not contain synthetic samples or estimated padding samples, all... Take 1.0). As the first weighted weight, As the second weighting weight, this embodiment takes =0.7、 =0.3 (The logarithm base in cross-entropy does not affect the meaning of the relative proportion of the two weights; the natural logarithm can also be used). Cross-entropy for partition affixation; The true distribution of the partition affiliation of the j-th verification sample (affiliation is 1 if it is affiliation, otherwise it is 0, constructed by the partition affiliation association index as described in S2025); c represents the attribution probability output by the candidate model (given by the second output head of the dual-task output structure described in S2025); c is the intensity partition category index. The improved Golden Jackal optimization algorithm described in this embodiment is an improved form formed by introducing a weighted collaboration of three leaders, refraction back learning, and a golden sine perturbation mechanism on the basis of the dual-leader framework of the known Golden Jackal optimization algorithm. The difference between it and the known Golden Jackal optimization algorithm is that: the number of leaders is expanded from two to three and the prey position is estimated by weighting the inverse of the normalized fitness; a three-layer position update in the gray wolf collaborative manner is introduced in the prey position estimation and position update; and refraction back learning and golden sine perturbation are superimposed. S004: Instantiate the model structure using the optimal hyperparameter vector, fit the intensity discrimination model on the enhanced training set, and evaluate the intensity discrimination accuracy (statistically calculated according to intensity level) and partition assignment accuracy on the test set, and output the model evaluation report; the accuracy of each intensity level in the model evaluation report is used for the subsequent differential setting of confidence threshold.

[0032] In one optional implementation, the iterative optimization of the improved golden jackal optimization algorithm includes: S0031: Encode the decision variables into position vectors of the improved Golden Jackal optimization algorithm, generate chaotic sequences using Logistic mapping and map them to the hyperparameter search space to generate candidate initial populations; The formula is: In the formula, This is the (b+1)th chaotic variable; Let b be the b-th chaotic variable; is the parameter for full chaos, typically 4; b is the index of the chaotic variable; The initial values ​​for chaos are set to avoid the set of fixed points {0.25, 0.5, 0.75}. In the formula, The i-th candidate individual in the initial candidate population; Let i be the i-th chaotic variable; This is the upper bound of the hyperparameter search space; This is the lower bound of the hyperparameter search space; For individual indexes; S0032: For each candidate individual in the candidate initial population, a refracted reverse individual is generated with a dynamic scale factor that decays with iteration. The refracted reverse individual and the one with the better fitness value among the original candidate individuals are retained to form the initial population. The formula is: In the formula, Let be the dynamic scaling factor for the t-th iteration; This is the upper bound of the dynamic scaling factor. As the lower bound of the dynamic scaling factor, this embodiment takes... =2.5、 =1; t is the iteration number index; This represents the maximum number of iterations. This refers to the i-th refracted reverse individual in the t-th iteration; It is the i-th candidate individual in the t-th iteration; S0033: Based on the fitness function, calculate and sort the fitness values ​​of all individuals in the initial population or the population updated in the last iteration, and select the top three individuals as the first leader, the second leader, and the third leader, respectively. S0034: In each iteration, the first leader, the second leader and the third leader estimate the prey position, and enter the exploration or development phase by combining the prey escape mechanism and the gray wolf cooperation mechanism. The three-layer cooperative position update is performed with the prey position as the convergence guide to obtain the temporary individual position. The formula for weighted prey location estimation by three leaders is: In the formula, The normalized fitness inverse weight of the first leader individual; The normalized fitness inverse weight of the second leader individual; The normalized fitness inverse weight of the third leader individual; The normalized fitness inverse weight of the m-th leader individual; m is the leader individual index, belonging to... ; The fitness function is to substitute the candidate decision variable vectors corresponding to the position vectors of the individuals in parentheses into the fitness function and output the corresponding fitness values. The first leader in the t-th iteration; The second leader in the t-th iteration; The third leader in the t-th iteration; Let be the prey position in the t-th iteration; For the m-th leader individual in the t-th iteration; Escape energy and phase transition, the formula is: In the formula, Let be the prey escape energy in the t-th iteration; The numbers are uniformly random numbers in the interval [0,1]. When the value is ≥1, the exploration phase begins (large-scale individual search). When the value is less than 1, the development stage begins (individuals converge towards the prey location to hunt). Three-tier collaboration location update: Development phase ( <1), the formula is: In the formula, Let be the distance vector between an individual and the m-th leader individual; Let be the swing coefficient between the individual and the m-th leader individual; For the m-th leader individual in the t-th iteration; For the i-th individual in the t-th iteration; The temporary component generated by the collaboration of the primary leader individual; The control coefficient generated by the m-th leader; The control coefficient generated by the primary leader; For the t-th iteration Distance vector between the first leader and the individual leader; The temporary component generated by the collaboration between the first and second leading individuals; The swing coefficient between the individual and the second leader; The control coefficient generated for the second leader; Temporary components generated for prey location coordination; Let be the convergence factor for the t-th iteration; The first uniform random number between [0,1] generated for the m-th collaboration; The second uniform random number between [0,1] generated in the m-th collaboration; The temporary individual position obtained from the fusion of three layers of collaboration; The swing coefficient between the individual and the third leader individual; The control coefficient generated by the third leader; Exploration phase ( 1) The formula is: In the formula, The temporary component generated by the collaboration of the m-th leader individual; For random individuals; S0035: Perform a position perturbation of the temporary individual position using a combination of sine and cosine using the golden ratio constant to obtain the perturbated position; The formula is: In the formula, The position after the disturbance; The first perturbation phase parameter is updated generation by generation, ∈[0,2π]; The second perturbation phase parameter is updated generation by generation, ∈[0,π]; The golden ratio constant, ; S0036: Using the current iteration's dynamic scaling factor A refracted inverse solution is generated for the perturbed position. The refracted inverse solution with the better fitness value among the perturbed positions is retained as the new generation position of the temporary individual, thus obtaining an updated population. S0037: Recalculate the fitness values ​​of all individuals in the updated population and update the first leader, the second leader, and the third leader, and return the results to the prey location estimation stage to form an iterative closed loop; S0038: When the maximum number of iterations is reached or the change in the optimal fitness value for a consecutive preset number of generations is less than the convergence threshold, the iteration is terminated, the first leader individual is output as the global optimal individual, and the global optimal individual is decoded to obtain the optimal hyperparameter vector.

[0033] In one optional implementation, the feature vector set is input into the intensity discrimination model, which outputs the discrimination confidence distribution of each survey point at each intensity level, and takes the maximum confidence score as the confidence score of that survey point, including: S2031: Verify that the intensity discrimination model is loaded and the version tag is consistent with the current configuration; verify that the sample size of the feature vector set is consistent with the effective sample size output by the data preprocessing stage, and that the dimension of each feature vector is consistent with the preset feature definition; if any verification fails, terminate the inference and output the corresponding error message. S2032: After the verification is passed, the feature vector set of all survey points is constructed as an inference input matrix and input into the intensity discrimination model in batches at one time. The original output of the model is normalized to obtain the discrimination confidence distribution of each survey point at each intensity level after normalization, which sums to 1. S2033: For each survey point, extract the maximum value in the discrimination confidence distribution as the confidence score of that survey point, record the intensity level corresponding to the maximum value as the discrimination intensity, and extract the second highest confidence level to calculate the difference between the confidence score and the second highest confidence level as the discrimination dissonance degree; the discrimination intensity is output to the confidence threshold comparison stage as the index key of the differentiation threshold, the confidence score is output to the sample separation stage as the separation criterion, and survey points whose discrimination dissonance degree exceeds the preset dissonance limit are marked with a review prompt in the abnormal separation file; the statistical results of the confidence scores are output to the document generation stage as the data source for the statistical description of the survey points; the confidence scores, discrimination intensity, and discrimination confidence distribution are cached and stored with the survey point number as the key, and the cached results are directly read when the same dataset is processed repeatedly to ensure output determinism.

[0034] In one optional implementation, based on the confidence score and a preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an outlier sample set, including: S2041: Based on the discrimination accuracy of each intensity level in the model evaluation report, configure a confidence threshold for each intensity level. Intensity levels with a discrimination accuracy not lower than the average discrimination accuracy of each intensity level correspond to a confidence threshold not lower than a preset benchmark threshold. Intensity levels with a discrimination accuracy lower than the average correspond to a confidence threshold lower than the preset benchmark threshold, thus forming a differentiated threshold configuration table. The formula is: In the formula, Here, k represents the confidence threshold for intensity level k; k is the intensity level variable. The accuracy rate of intensity level k in the model evaluation report; This represents the average accuracy across all levels. The baseline threshold is set to 0.75 in this embodiment; The sensitivity coefficient is set to 0.5 in this embodiment; The lower bound of the threshold is set to 0.6 in this embodiment; The upper bound of the threshold is set to 0.9 in this embodiment; The truncation function restricts the results to upper and lower bounds. Based on the discrimination accuracy of each intensity level in the model evaluation report, a confidence threshold is configured for each intensity level. Intensity levels with high discrimination accuracy correspond to higher confidence thresholds, while those with low discrimination accuracy correspond to lower confidence thresholds. Specifically, for intensity levels with low discrimination accuracy, the overall confidence of the model output is relatively low, so a relatively lower confidence threshold is used to ensure basic coverage of effective samples at that level and avoid excessive rejection of effective samples. Conversely, for intensity levels with high discrimination accuracy, the overall confidence of the model output is relatively high, so a relatively higher confidence threshold is used to filter out low-confidence samples, thereby maintaining a balanced number of effective samples across all intensity levels. S2042: For each survey point, using the discrimination intensity level of that survey point as the index key, retrieve the corresponding threshold from the differential threshold configuration table, and compare the confidence score of the survey point with the corresponding threshold: survey points with a confidence score not lower than the corresponding threshold are included in the valid sample set, and survey points with a confidence score lower than the corresponding threshold are included in the abnormal sample set. An abnormal separation marker containing the separation time, the threshold used, and the confidence score is added to the survey points included in the abnormal sample set; survey points with a confidence score equal to the corresponding threshold are included in the valid sample set. S2043: The sum of the sample sizes of the valid sample set and the abnormal sample set is compared with the valid sample size output by the data preprocessing stage. If the comparison is successful, the partition area summary is started. If the comparison fails, the process is terminated and a location prompt for the uncovered survey points is output. S2044: Summarize the differentiated threshold configuration table, each anomaly separation marker, and the quantity verification results to generate anomaly separation files; perform spatial aggregation on the anomaly sample set, and based on the administrative division association index and the spatial association index, aggregate anomaly survey points that are spatially less than the preset aggregation radius (example: 10 km) and belong to the same county-level administrative region into the same anomaly area, count the number of anomaly points within the area, calculate the registered area of ​​the anomaly area and record the background intensity level, and output the anomaly area list and the anomaly area area; when the anomaly sample set is empty, output empty anomaly area markers, and the anomaly area area is set to zero; the anomaly area list and the anomaly area area are output to the balancing verification stage as deduction objects for secondary verification, and the anomaly separation files are attached to the appendix of the standardized intensity description document according to the configuration.

[0035] In one optional implementation, based on the valid sample set, area grouping and summarizing and a five-level stable sort are performed on the partitioned record set to output an ordered set of partition areas, including: S2051: Verify that the sum of the sample sizes of the valid sample set and the abnormal sample set has been verified, and verify that each survey point in the valid sample set can be matched with at least one intensity partition record in the partition record set according to the zoning association index, and add an isolated point mark to the survey points that cannot be matched. S2052: For each intensity zone, count the number of valid survey points falling into that zone from the valid sample set, obtain the zone coverage count table, and identify the zero-coverage zone with a coverage count of zero. S2053: Using intensity level as the first-level grouping key and the names of the four-level administrative regions (province, city, county, and township) as the hierarchical grouping keys, the partition record set is grouped and summarized. Multiple partition records of the same intensity level and the same township are merged and accumulated. High-precision decimal arithmetic is used to perform area accumulation, with a decimal context precision of not less than 28 significant digits, to obtain the total area of ​​each intensity level and township combination, as well as the total area of ​​the county, city, and province at each level upwards. Among them, high-precision decimal arithmetic is used to eliminate floating-point rounding errors introduced by the calculation process of this system itself. The first verification is mainly aimed at the difference in the first verification value, which mainly comes from the difference between the rounding of the map patch calculation and the registration caliber of the upstream partition area data source. The two have different functions and are not interchangeable. S2054: A stable sorting algorithm is used to perform a five-level stable sorting on the grouped summary results, outputting an ordered set of partition areas and a sorting process archive: the first level is sorted in descending order by intensity level, and the second to fifth levels are sorted in descending order by the corresponding area of ​​province, city, county, and township, respectively. The relative order of records under the same sorting key value remains unchanged; before sorting, the administrative region name field participating in the comparison is normalized using the name after data preprocessing; the partition coverage count table is output to the balancing verification stage and the document generation stage along with the ordered set of partition areas, and the zero-coverage partition list is output to the balancing verification stage to perform the upper limit constraint on the area of ​​zero-coverage partitions, and is output to the document generation stage for zero-coverage prompts.

[0036] In one optional implementation, the areas of each partition in the ordered partition area set whose intensity value is higher than a preset lower limit are summed and compared with the total area of ​​the entire region. Accumulated deviations are identified and eliminated by apportioning them according to the partition area ratio. After spatial aggregation of the abnormal sample set is performed, the abnormal areas are split into abnormal sub-areas falling into each cross-boundary partition according to their partition affiliation. The regular partition net area (after deducting the abnormal sub-areas) is then compared with the baseline of the total net area (after deducting the registered area of ​​the abnormal areas) on the entire region side for secondary verification. The result is the output of the balanced and verified partition area set and the abnormal area area, including: S2061: Confirm the range of zones participating in the summation, and only include zones with intensity values ​​higher than the preset lower limit level (VI in this embodiment). The area of ​​zones with intensity values ​​not higher than the preset lower limit level is separately recorded as the area of ​​the unincluded zone. Establish the accounting identity: Total area of ​​the whole region = Sum of zone areas + Area of ​​unincluded zone + First verification difference. This identity is used as the accounting basis for the first verification. The first verification difference reflects the deviation of the upstream zone area data from the whole region benchmark, rather than the result of the definition of the identity itself. S2062: Sum the areas of each partition involved in the summation, calculate the first verification difference between the total area of ​​the entire region minus the area of ​​the unincluded region and the sum of the areas of the partitions. If the absolute value of the first verification difference is not greater than the first error tolerance, it is determined that no fine-tuning is required. If it is greater than the first error tolerance, it is determined that there is a cumulative deviation of the partition area system relative to the global benchmark. If no fine-tuning is required, the summed area of ​​each partition involved in the summation is directly used as the fine-tuned area of ​​that partition. The formula is: In the formula, This is the difference in a single verification test. The total area of ​​the entire region; Area not included; Let p be the summation area of ​​the p-th partition participating in the summation; p is the index of the partition participating in the summation. ≤ (First error tolerance in this embodiment) When the value is 0.01km2, it is determined that no fine-tuning is needed; when the value is greater than the first error tolerance, it is determined that there is a cumulative deviation of the partition area system relative to the global benchmark. S2063: In cases where a cumulative deviation is determined, the first verification difference is distributed to each partition according to the proportion of each partition area to the total area of ​​the partitions, resulting in a fine-tuned partition area; wherein, for the zero-coverage partition, the fine-tuning amount shall not exceed the preset upper limit of the partition area, and the excess shall be proportionally transferred to the partitions with sample coverage; after fine-tuning, the first verification difference is re-verified and falls within the first error tolerance; if it does not fall within the tolerance, iterative fine-tuning is performed to the upper limit of the preset number of rounds (10 rounds in this embodiment); if it still does not fall within the tolerance after exceeding the upper limit of the preset number of rounds, the process is terminated and a balancing failure prompt is output; The formula is: In the formula, For the fine-tuned first The summation area of ​​the p-th participating partition in the round; For the first The summation area of ​​the p-th participating partition in the round; For round index; For the first The difference in the first round of verification; in the first iteration Take the original total area of ​​the corresponding partition; when there is a zero-coverage partition and its proportionally allocated fine-tuning amount reaches the upper limit of the zero-coverage partition fine-tuning proportion. When the excess portion is transferred to the sample-covered partitions and the area distribution of each partition is changed, multiple rounds of iteration are required for convergence. This is the reason for setting an upper limit on the number of iteration rounds. In the case of no zero-covered partitions or the transfer amount not reaching the upper limit, convergence is achieved in one round of allocation. In the formula, To fine-tune the upper limit of the proportion of zero-coverage partitions, this embodiment takes... =0.5%; S2064: For each anomalous area, it is divided into anomalous sub-areas falling into each intensity zone it crosses, such that the sum of the anomalous sub-areas of each anomalous area is not greater than the registered area of ​​the anomalous area, and no anomalous sub-area is greater than the fine-tuned area of ​​its respective zone; the anomalous sub-areas are allocated according to the proportion of the geographical intersection area of ​​the anomalous area and each cross-intensity zone to the total area of ​​the anomalous area, and satisfy the above two constraints; when the registered area of ​​the anomalous area is inconsistent with the sum of the geographical intersection area, the registered area is used as the standard and scaled proportionally; the anomalous sub-areas falling into each zone are subtracted from the anomalous area in the fine-tuned area of ​​each zone, and then summed to obtain the net area of ​​the regular zone; the registered area of ​​the anomalous area is subtracted from the total area of ​​the entire region to obtain the net area benchmark of the entire region; the second-order verification difference between the net area benchmark of the entire region after subtracting the area of ​​the unincluded area and the net area of ​​the regular zone is calculated; The formula is: In the formula, The sum of the areas of the sub-regions of the abnormal region A; The set of partitions that are connected to the abnormal region A; To fall into the partition The area of ​​the anomalous sub-sub; For partitioned indexes; Registered area for abnormal zones; To fall into the partition The fine-tuned area of ​​each partition is such that the sum of the areas of all abnormal sub-areas of each abnormal region is not greater than the registered area of ​​the abnormal region, and the area of ​​any abnormal sub-area is not greater than the fine-tuned area of ​​the partition in which it belongs. In the formula, q represents the net area of ​​the regular partition after deducting the area of ​​the abnormal sub-partition; q is the partition index that is not connected to the abnormal area. To fall into the partition The area of ​​the partition after fine-tuning; In the formula, Based on the net area of ​​the entire region; The total area of ​​the entire region; To fall into the partition The area of ​​the partition after fine-tuning; This is the difference between the two checks; substituting the above definition and expanding, we get: In the formula, This is the residual check difference after fine-tuning (it is zero when fully converged); The secondary verification difference consists of the primary verification residual and the anomaly area consistency check item. The first item is the residual primary verification difference after fine-tuning, and the second item is the difference between the sum of the sub-areas of the anomaly area and the registered area of ​​the anomaly area, directly reflecting the consistency between the anomaly area and the registered area system. To isolate the influence of the primary verification residual, the determination is based on the caliber item. The absolute value is compared with the second error tolerance; when there are multiple abnormal areas, the caliber item of each abnormal area is calculated and compared with the second error tolerance one by one, so as to locate the specific abnormal area where the registration caliber is inconsistent. S2065: If the absolute value of the caliber item is not greater than the second error tolerance, the secondary verification is deemed to have passed, and the set of partition areas composed of the fine-tuned partition areas is confirmed as the set of partition areas after balancing verification; if it is greater than the second error tolerance, it is deemed that the area of ​​the abnormal area is inconsistent with the registered caliber of the partition area system, and the abnormal area number suspected of conflict and the list of partitions it crosses are output; the abnormal area number suspected of conflict and the list of partitions it crosses are fed back to the abnormal sample set spatial aggregation stage, and the abnormal sample set spatial aggregation stage re-splits the area of ​​the suspected abnormal area according to the partition affiliation and then returns to execute the secondary verification; the re-split verification of the same abnormal area is performed for a maximum of a preset number of rounds (example: 3 rounds), and if it still fails after exceeding the preset number of rounds, the process is terminated and a balancing failure prompt and a manual verification list are output; S2066: When the abnormal sample set is empty, the area of ​​the abnormal region is zero, the net area of ​​the regular partition is equal to the sum of the areas of the partitions after fine-tuning, and the secondary verification degenerates into a re-verification of the result of the primary verification. S2067: Summarize the first verification difference and its relative magnitude. The fine-tuning details table, the number of fine-tuning rounds, and the difference in the secondary verification are used to generate a balance verification file, which is then attached to the appendix of the standardized intensity description document according to the configuration. When the fine-tuning amount of any partition area in the fine-tuning details table causes its sorting key to cross the adjacent partitions at the same level, a sequence review prompt is triggered. S2068: When the relative magnitude of the single-verification difference of a preset number (e.g., 5) of consecutive earthquake cases exceeds the preset error limit (e.g., 0.1%), output a data source governance prompt indicating that there is a systematic accuracy problem in the upstream partition area data source.

[0037] In one optional implementation, the balanced and verified set of zonal areas, the area of ​​the abnormal zone, the statistical results of the confidence score, and the set of basic parameters are input into a multi-granularity document generation module. Based on the magnitude in the set of basic parameters and the number of county-level administrative regions involved, an upper limit for the number of township-level listings is adaptively determined. A standardized intensity description document is generated, exported, and automatically opened for presentation, including: S2071: When the magnitude is not less than a preset major earthquake threshold (e.g., preset major earthquake threshold = 7.0), or the number of county-level administrative regions involved is not less than a preset number threshold (e.g., preset number threshold = 30), the upper limit of the number of townships listed is set to a first preset value (e.g., first preset value = 10). Only townships with the highest area ranking and not exceeding the upper limit are listed in each intensity zone. The excess is represented by a summary count statement. When the magnitude is less than the preset major earthquake threshold and the number of county-level administrative regions involved is less than the preset number threshold, all township names are listed. When the user-specified listing upper limit is detected, the user-specified value is given priority and a manual overwrite log is recorded. S2072: Construct a multi-level chapter framework from top to bottom, including granularity of earthquake situation overview, granularity of intensity distribution across the entire region, granularity of hierarchical administrative divisions, and detailed granularity of townships, and attach data source tags to each chapter to indicate the data source; fill in the content according to the chapter framework, wherein the earthquake situation overview chapter references the fields of earthquake occurrence time, epicenter location, magnitude, and isoseismal major axis direction, major and minor axis length, and surface rupture zone length from the basic parameter set; fields in the basic parameter set that have standard placeholders are converted to default sentence structure and the placeholder literals are not written into the document; generate a special paragraph on the anomaly area when an anomaly area exists, and skip the paragraph when the anomaly area is empty; generate a paragraph on the instrument intensity distribution when the grouping and counting results of the instrument intensity record set are not empty, and skip the paragraph when they are empty. S2073: Compare all numerical expressions in the document with the upstream data corresponding to their data source tags item by item, and verify the cross-level closure of the same value between multiple granular chapters. If there is any inconsistency, locate and correct it. After the verification is passed, perform format standardization, export it to the preset format and automatically open and present it with the system default program. The manual overwrite log and verification and correction record are archived with the document.

[0038] This invention also provides a machine learning-based intensity monitoring system 300, with reference to... Figure 3 The system may include the following units: Data acquisition unit 301 is used to acquire the basic parameter set, zonal record set, survey point record set, instrument intensity record set and total area of ​​the earthquake event, and output the raw dataset. The data preprocessing unit 302 is used to perform data preprocessing on the original dataset, output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; The machine learning intensity discrimination unit 303 is used to input the feature vector set into the intensity discrimination model, output the discrimination confidence distribution of each survey point at each intensity level, and take the maximum confidence as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; The sample separation unit 304 is used to separate all survey points corresponding to the feature vector set into a valid sample set and an abnormal sample set based on the confidence score and the preset confidence threshold. The partition area summarization and sorting unit 305 is used to perform area grouping and summarization and five-level stable sorting on the partition record set based on the effective sample set, and output an ordered partition area set; wherein, the five-level stable sorting includes sorting in descending order by intensity level, and within the same intensity level, sorting in descending order by the corresponding area of ​​province, city, county, and township. The area balancing verification unit 306 is used to sum the areas of each partition in the ordered partition area set whose intensity value is higher than the preset lower limit level, compare it with the total area of ​​the whole domain, identify the cumulative deviation and eliminate it according to the partition area ratio; after splitting the abnormal area obtained by performing spatial aggregation on the abnormal sample set into abnormal sub-areas falling into each cross-partition according to the partition affiliation, the regular partition net area after deducting the abnormal sub-areas on the partition side is compared with the whole domain net area benchmark after deducting the registered area of ​​the abnormal area on the whole domain side to perform a second verification, and output the partition area set and abnormal area after balancing verification; The document generation unit 307 is used to input the balanced and verified zonal area set, the abnormal area, the statistical results of the confidence score, and the basic parameter set into the multi-granularity document generation module, adaptively determine the upper limit of the number of township-level listings based on the magnitude in the basic parameter set and the number of county-level administrative regions involved, generate a standardized intensity description document, export it, and automatically open and present it.

[0039] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus; The processor, communication interface, and memory communicate with each other through a communication bus. Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the machine learning-based intensity monitoring method of the present invention.

[0040] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or Non-Volatile Memory (NVM), such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0041] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0042] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the machine learning-based intensity monitoring method of the present invention.

[0043] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0048] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine learning-based intensity monitoring method, characterized in that, The method includes: Obtain the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset; Perform data preprocessing on the original dataset to output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; The feature vector set is input into the intensity discrimination model, which outputs the discrimination confidence distribution of each survey point at each intensity level, and the maximum confidence score is taken as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; Based on the confidence score and the preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an abnormal sample set. Based on the effective sample set, the area grouping and summarization and five-level stable sorting are performed on the partition record set to output an ordered set of partition areas; The summation of the areas of each partition with an intensity value higher than the preset lower limit in the ordered partition area set is compared with the total area of ​​the entire domain to identify the cumulative deviation and eliminate it according to the partition area ratio; the abnormal area obtained by performing spatial aggregation on the abnormal sample set is split into abnormal sub-areas falling into each cross-partition according to the partition affiliation, and the regular partition net area after deducting the abnormal sub-areas on the partition side is compared with the benchmark of the net area of ​​the entire domain after deducting the registered area of ​​the abnormal area on the entire domain side to perform a second verification, and output the partition area set and abnormal area after balancing verification; The balanced and verified set of zonal areas, the area of ​​the abnormal zone, the statistical results of the confidence score, and the set of basic parameters are input into the multi-granularity document generation module. Based on the magnitude of the earthquake in the set of basic parameters and the number of county-level administrative regions involved, the upper limit of the number of township-level listings is adaptively determined, a standardized intensity description document is generated, exported, and automatically opened and presented.

2. The intensity monitoring method based on machine learning according to claim 1, characterized in that, Obtain the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset, including: Read the epicenter records from the epicenter data table and extract the fields of earthquake time, epicenter location, epicenter latitude and longitude, magnitude, major axis direction of isoseismal lines, major and minor axis lengths, and surface rupture zone length. Fill missing fields with standard placeholders and output the basic parameter set. Read the intensity zoning records from the intensity map statistics table. Each intensity zoning record should contain at least the fields of intensity level, zoning area, province, city, county, township name and township area. Output the zoning record set. Read the survey point records from the survey point statistics table. Each survey point record should contain at least the fields of survey point number, latitude and longitude, name of the fourth-level administrative region, on-site intensity assessment, description of the damage phenomenon and site environment category. Output the survey point record set. Read the instrument intensity records from the instrument intensity statistics table, group and count them according to the preset standard intensity interval, and output the instrument intensity record set; Read the total area of ​​the earthquake zone from the global basic geographic data table. The total area is the sum of the land areas within the scope of this earthquake intensity assessment. Output the total area. Perform cross-table consistency checks on the basic parameter set, the partition record set, the survey point record set, the instrument intensity record set, and the total area of ​​the entire region. Construct a regional affiliation index with the survey point number as the primary key and the name of the fourth-level administrative region as the foreign key. Construct a spatial association index with the latitude and longitude of the survey points and the latitude and longitude of the epicenter. Output the original dataset with the association index.

3. The intensity monitoring method based on machine learning according to claim 2, characterized in that, Data preprocessing is performed on the original dataset to output a normalized dataset, and feature encoding is performed on the survey point data in the normalized dataset to construct a feature vector set, including: Perform format validation on each field of the original dataset, validate the numeric type and valid range for numeric fields, mark records that exceed the range as invalid records, and fill missing fields with standard placeholders. Output the format-validated intermediate dataset, the list of invalid records, and the records filled with placeholders. Perform whitespace stripping, invisible character removal, and garbled character replacement on the intermediate dataset after format verification, and output the cleaned intermediate dataset; The cleaned intermediate dataset is normalized by administrative region names. Duplicate provincial and municipal district names are removed from the municipalities directly under the central government. The county-level divisions under the provincial government are attached according to the hierarchical relationship between the province and the county-level city. Standardized labels are generated for towns, townships and streets at the township level. The normalized intermediate dataset is then output. For the normalized intermediate dataset, duplicate detection is performed on the survey point records using the combination of survey point number and latitude and longitude as the key, and duplicate deduplication is performed on the partition records using the combination of intensity level, fourth-level administrative region name and township name as the key, eliminating duplicate registration of the same township between adjacent intensity partitions, and outputting a normalized dataset, effective sample size and duplicate record list. For each survey point record in the standardized dataset, feature encoding is performed to construct a feature vector that includes at least epicentral distance features, magnitude features, site shear wave velocity features, instrument intensity features, damage phenomenon codes, site environment category codes, and administrative division codes, forming a feature vector set and a corresponding intensity label vector; wherein, for survey points without matching instrument intensity records but which need to be estimated and filled, the estimated intensity calculated by the empirical attenuation formula is used to fill the estimated intensity and an estimated label is added. The invalid record list, the placeholder-filled records, the duplicate record list, and the estimated markers are summarized to generate a data quality report that includes the original sample size, the number of removed samples, the number of valid samples, and the distribution of removal reasons at each stage.

4. The intensity monitoring method based on machine learning according to claim 3, characterized in that, The training method for the intensity discrimination model includes: The feature vector set and the corresponding intensity label vector are divided into a training set, a validation set and a test set according to a preset ratio; The sample size of each intensity level in the statistical training set is calculated. For intensity levels with a sample size lower than the preset lower limit, the relationship between epicentral distance and intensity attenuation is used as a priori. Interpolation and oversampling are performed on the samples of the intensity level with a sample size lower than the preset lower limit to generate synthetic samples and add synthetic labels. The enhanced training set is then output. Using the hyperparameters of the machine learning model as decision variables, and the weighted combination of the intensity prediction bias and partition assignment cross-entropy of the candidate model trained on the enhanced training set on the validation set as the fitness function, the improved Golden Jackal optimization algorithm is executed to iteratively find the optimal hyperparameter vector. The model structure is instantiated using the optimal hyperparameter vector, and an intensity discrimination model is fitted on the enhanced training set. The intensity discrimination accuracy and partition assignment accuracy are evaluated on the test set, and a model evaluation report is output.

5. The intensity monitoring method based on machine learning according to claim 4, characterized in that, The iterative optimization of the improved golden jackal optimization algorithm includes: The decision variables are encoded as position vectors of the improved Golden Jackal optimization algorithm, and a chaotic sequence is generated by the Logistic mapping and mapped to the hyperparameter search space to generate candidate initial populations. For each candidate individual in the candidate initial population, a refracted reverse individual is generated with a dynamic scale factor that decays with iteration. The refracted reverse individual with the better fitness value among the original candidate individuals is retained to form the initial population. Based on the fitness function, calculate and sort the fitness values ​​of all individuals in the initial population or the population updated in the last iteration, and select the top three individuals as the first leader, the second leader, and the third leader, respectively. In each iteration, the first leader, the second leader, and the third leader estimate the prey's location. Combining the prey escape mechanism and the gray wolf cooperation mechanism, they enter the exploration or development phase. The three-layer cooperative location update is performed with the prey's location as the convergence guide to obtain the temporary individual location. The position of the temporary individual is perturbed by a combination of sine and cosine using the golden ratio constant to obtain the perturbed position; The refracted inverse solution is generated for the perturbed position using the current iterative dynamic scale factor. The refracted inverse solution with the better fitness value among the perturbed positions is retained as the new generation position of the temporary individual, thus obtaining an updated population. The fitness values ​​of all individuals in the updated population are recalculated and the first leader, the second leader, and the third leader are updated. The results are then fed back to the prey location estimation stage to form an iterative closed loop. The iteration terminates when the maximum number of iterations is reached or the change in the optimal fitness value for a consecutive preset number of generations is less than the convergence threshold. The first leader individual is output as the global optimal individual, and the global optimal individual is decoded to obtain the optimal hyperparameter vector.

6. The intensity monitoring method based on machine learning according to claim 5, characterized in that, The feature vector set is input into the intensity discrimination model, which outputs the discrimination confidence distribution of each survey point at each intensity level, and takes the maximum confidence score as the confidence score of that survey point, including: Verify that the intensity discrimination model is loaded and the version tag is consistent with the current configuration; verify that the sample size of the feature vector set is consistent with the effective sample size output by the data preprocessing stage, and that the dimension of each feature vector is consistent with the preset feature definition; if any verification fails, terminate the inference and output the corresponding error message. After the verification is passed, the feature vector set of all survey points is constructed as an inference input matrix and input into the intensity discrimination model in batches at one time. The original output of the model is normalized to obtain the discrimination confidence distribution of each survey point at each intensity level after normalization, which sums to 1. For each survey point, the maximum value in the discrimination confidence distribution is extracted as the confidence score of that survey point, the intensity level corresponding to the maximum value is recorded as the discrimination intensity, and the second highest confidence score is extracted to calculate the difference between the confidence score and the second highest confidence score as the discrimination divergence degree.

7. The intensity monitoring method based on machine learning according to claim 6, characterized in that, Based on the confidence score and the preset confidence threshold, all survey points corresponding to the feature vector set are separated into a valid sample set and an outlier sample set, including: Based on the discrimination accuracy rate of each intensity level in the model evaluation report, a confidence threshold is configured for each intensity level. Intensity levels with a discrimination accuracy rate not lower than the average discrimination accuracy rate of each intensity level correspond to a confidence threshold not lower than a preset benchmark threshold. Intensity levels with a discrimination accuracy rate lower than the average value correspond to a confidence threshold lower than the preset benchmark threshold, thus forming a differentiated threshold configuration table. For each survey point, the corresponding threshold is retrieved from the differential threshold configuration table using the discrimination intensity level of the survey point as the index key. The confidence score of the survey point is compared with the corresponding threshold, and all survey points in the feature vector set are separated into a valid sample set and an abnormal sample set. The sum of the sample sizes of the valid sample set and the abnormal sample set is compared with the valid sample size output from the data preprocessing stage. If the comparison is successful, the partition area aggregation is initiated. If the comparison fails, the process is terminated and a location prompt for the uncovered survey points is output. Summarize the differentiated threshold configuration table, each anomaly separation marker, and the quantity verification results to generate anomaly separation files; perform spatial aggregation on the anomaly sample set, and based on the administrative division association index and the spatial association index, aggregate anomaly survey points that are spatially less than the preset aggregation radius and belong to the same county-level administrative region into the same anomaly area, count the number of anomaly points within the area, calculate the registered area of ​​the anomaly area and record the background intensity level, and output the anomaly area list and the anomaly area area; when the anomaly sample set is empty, output an empty anomaly area marker, and the anomaly area is set to zero.

8. The intensity monitoring method based on machine learning according to claim 7, characterized in that, Based on the effective sample set, area grouping and summarization and five-level stable sorting are performed on the partitioned record set to output an ordered set of partition areas, including: The verification of the sum of the sample sizes of the valid sample set and the abnormal sample set has been passed, and the verification of each survey point in the valid sample set can be matched with at least one intensity partition record in the partition record set according to the zoning association index. The survey points that cannot be matched are marked as isolated points. For each intensity zone, the number of valid survey points falling into that zone from the valid sample set is counted to obtain a zone coverage count table, and zero-coverage zones with a coverage count of zero are identified. Using the intensity level as the first-level grouping key and the names of the four-level administrative regions (province, city, county, and township) as the hierarchical grouping keys, the partition record set is grouped and summarized. Multiple partition records of the same intensity level and the same township are merged and accumulated, and high-precision decimal arithmetic is used to perform surface accumulation to obtain the total area of ​​each intensity level and township combination, as well as the total area of ​​the county, city, and province at each level upwards. A stable sorting algorithm is used to perform a five-level stable sort on the grouped summary results, outputting an ordered set of partition areas and a sorting process archive. The partition coverage count table is output along with the ordered set of partition areas to the balancing verification stage and the document generation stage. The list of zero-coverage partitions is output to the balancing verification stage to perform upper limit constraints on the area of ​​zero-coverage partitions and is output to the document generation stage for zero-coverage prompts.

9. The intensity monitoring method based on machine learning according to claim 8, characterized in that, The summation of the areas of each zone in the ordered zone area set whose intensity value is higher than the preset lower limit level is compared with the total area of ​​the entire region to identify cumulative deviations and eliminate them by allocating them according to the proportion of the zone area. The abnormal areas obtained by spatial aggregation of the abnormal sample set are then divided into abnormal sub-areas falling into each cross-zone according to their zone affiliation. The regular zone net area after deducting the abnormal sub-areas on the zone side is compared with the total net area benchmark after deducting the registered area of ​​the abnormal areas on the entire region side to perform a second verification. The result is the output of the balanced and verified zone area set and the abnormal area area, including: Confirm the range of zones to be included in the summation, and only include zones with intensity values ​​higher than the preset lower limit level. The area of ​​zones with intensity values ​​not higher than the preset lower limit level is separately recorded as the area of ​​the unincluded zone. Establish a calculation identity as the calculation basis for the first verification. The areas of each partition involved in the summation are summed, and the first verification difference between the total area of ​​the whole region minus the area of ​​the non-included area and the sum of the areas of the partitions is calculated. If the absolute value of the first verification difference is not greater than the first error tolerance, it is determined that no fine-tuning is required. If it is greater than the first error tolerance, it is determined that there is a cumulative deviation of the partition area system relative to the whole region benchmark. In cases where there is a cumulative deviation, the first verification difference is distributed to each partition according to the proportion of each partition area to the total area of ​​the partitions, thus obtaining the fine-tuned partition area. Each anomalous area is divided into anomalous sub-areas falling into each intensity zone it crosses, according to its zoning affiliation. The sum of the anomalous sub-areas of each anomalous area is not greater than the registered area of ​​the anomalous area, and no anomalous sub-area is greater than the fine-tuned area of ​​its respective zone. The anomalous sub-areas falling into each zone are subtracted from the anomalous area in the fine-tuned area, and then summed to obtain the net area of ​​the regular zones. The registered area of ​​the anomalous area is subtracted from the total area of ​​the entire region to obtain the net area benchmark of the entire region. The second-order verification difference between the net area benchmark of the entire region (after deducting the area of ​​the unincluded area) and the net area of ​​the regular zones is calculated. The second-order verification difference consists of a primary verification residual term and a caliber term, where the caliber term is the difference between the sum of the anomalous sub-areas of each anomalous area and the registered area of ​​the anomalous area. If the absolute value of the caliber item is not greater than the second error tolerance, the secondary verification is deemed to have passed, and the set of partition areas composed of the fine-tuned partition areas is confirmed as the set of partition areas after balancing verification. If it is greater than the second error tolerance, it is determined that the area of ​​the abnormal area is inconsistent with the registration caliber of the partition area system, and the suspected conflicting abnormal area number and its cross-partition list are output. The suspected conflicting abnormal area number and its cross-partition list are fed back to the abnormal sample set spatial aggregation stage. The abnormal sample set spatial aggregation stage re-splits the area of ​​the suspected conflicting abnormal area according to the partition affiliation and then returns to execute the secondary verification. The re-split verification of the same abnormal area is performed for a maximum of a preset number of rounds. If it still fails after exceeding the preset number of rounds, the process is terminated and a balancing failure prompt and a manual verification list are output. When the abnormal sample set is empty, the area of ​​the abnormal region is zero, the net area of ​​the regular partition is equal to the sum of the areas of the partitions after fine-tuning, and the secondary verification degenerates into a re-verification of the result of the primary verification. Summarize the first verification difference and its relative magnitude, fine-tuning details table, number of fine-tuning rounds, and the second verification difference to generate a balance verification file, which is then attached to the appendix of the standardized intensity description document according to the configuration. When the fine-tuning amount of any partition area in the fine-tuning details table causes its sorting key to cross the adjacent partitions at the same level, a sequential review prompt is triggered. When the relative magnitude of the single-verification difference of a consecutive preset number of earthquake cases exceeds the preset error limit, a data source management prompt indicating that there is a systematic accuracy problem in the upstream partition area data source is output.

10. A machine learning-based intensity monitoring system, used to implement the machine learning-based intensity monitoring method as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition unit is used to acquire the basic parameter set, zonal record set, survey point record set, instrument intensity record set, and total area of ​​the earthquake event, and output the raw dataset. The data preprocessing unit is used to perform data preprocessing on the original dataset, output a normalized dataset, and perform feature encoding on the survey point data in the normalized dataset to construct a feature vector set; The machine learning intensity discrimination unit is used to input the feature vector set into the intensity discrimination model, output the discrimination confidence distribution of each survey point at each intensity level, and take the maximum confidence as the confidence score of the survey point; wherein, the intensity discrimination model is a model built based on machine learning algorithm and trained by iterative optimization of hyperparameters using an improved Golden Jackal optimization algorithm; The sample separation unit is used to separate all survey points corresponding to the feature vector set into a valid sample set and an abnormal sample set based on the confidence score and the preset confidence threshold. The partition area summarization and sorting unit is used to perform area grouping and summarization and five-level stable sorting on the partition record set based on the effective sample set, and output an ordered partition area set; The area balancing verification unit is used to sum the areas of each partition in the ordered partition area set whose intensity value is higher than the preset lower limit level, compare it with the total area of ​​the whole domain, identify the cumulative deviation and eliminate it according to the partition area ratio; after splitting the abnormal area obtained by performing spatial aggregation on the abnormal sample set into abnormal sub-areas falling into each cross-partition according to the partition affiliation, the regular partition net area after deducting the abnormal sub-areas on the partition side is compared with the whole domain net area benchmark after deducting the registered area of ​​the abnormal area on the whole domain side to perform a second verification, and output the partition area set and abnormal area after balancing verification; The document generation unit is used to input the balanced and verified zonal area set, the abnormal area, the statistical results of the confidence score, and the basic parameter set into the multi-granularity document generation module. Based on the magnitude of the earthquake in the basic parameter set and the number of county-level administrative regions involved, it adaptively determines the upper limit of the number of township-level listings, generates a standardized intensity description document, exports it, and automatically opens and presents it.