Method and system for intelligently removing clusters from seismic directory based on optimization algorithm

Abstract

The invention discloses a seismic directory intelligent cluster removal method and system based on an optimization algorithm. The main research content comprises the following steps: 1) preprocessing original seismic directory data subjected to manual cluster removal; 2) constructing an earthquake characteristic system fusing multi-dimensional characteristics such as space-time distribution and earthquake magnitude relationship, wherein the earthquake characteristic system comprises key characteristics such as time, longitude and latitude, earthquake magnitude and the like and a plurality of dynamic characteristics; 3) establishing a self-adaptive feature selection mechanism based on a particle swarm optimization algorithm, and automatically screening a feature combination which has the most discriminating capability on seismic cluster identification through an intelligent search strategy; and 4) constructing a random forest classification model optimized by PSO, realizing intelligent identification of background earthquakes and cluster earthquakes, and overcoming the limitation that a traditional method depends on empirical parameters. Through innovation of the technical principle, the limitation of a traditional method is broken through, the clump removing precision, the adaptability and the stability are remarkably improved, and the method has definite practical value and popularization prospects.

Description

Technical Field

[0001] This invention belongs to the field of intelligent de-clustering technology for earthquake catalogs, and particularly relates to an intelligent de-clustering method and system for earthquake catalogs based on optimization algorithms. Background Technology

[0002] Earthquake catalog declustering is a crucial task in seismic activity characteristic analysis. It aims to classify monitored natural earthquake events into background earthquakes and clustered earthquakes (primarily aftershock sequences and some earthquake swarms). This process removes aftershocks and earthquake swarm sequences triggered by moderate to strong earthquakes from the background earthquake list, facilitating a more accurate estimation of the regional seismic background. This work is of great significance for seismic hazard assessment, earthquake prediction model development, and earthquake statistical analysis.

[0003] Current mainstream techniques include spatiotemporal window methods and statistical modeling methods. Among spatiotemporal window methods, the Gardner-Knopoff method, by setting time and spatial windows, treats earthquakes within a specific spatiotemporal window as clustered earthquakes and removes them. Generally, the larger the magnitude, the larger the corresponding spatiotemporal window. Its advantages are its simple principle and ease of operation, making it one of the most widely used methods for removing clusters from earthquake catalogs. The Reasenberg method, also known as the pseudo-random spatiotemporal window method, mainly considers stress redistribution and Ohmori's law to determine the spatiotemporal interactions between earthquakes. This method connects earthquake events based on their spatiotemporal distance, considering the largest event in the cluster as the mainshock and other events as aftershocks. If event A is the mainshock of event B, and event B is the mainshock of event C, then events A, B, and C are all considered to belong to the same cluster. The relationship between individual events and clustered earthquake events is determined through spatial and temporal interaction formulas. In statistical modeling methods, the ETAS (Epidemic-Type Aftershock Sequence) model, initially proposed by Ogata Yosihiko (1988), belongs to the category of point process models. Its core idea is that each earthquake (including the mainshock and aftershocks) can act as a trigger source, generating a series of aftershocks. This model divides the earthquake sequence into background events and triggering events. Background events are caused by long-term seismic activity rates and are related to regional tectonic activity; triggering events are triggered by previous earthquakes (mainshocks or aftershocks). The trigger rate of the ETAS model follows a modified Omori law, and the magnitude distribution obeys the Gutenberg-Richter law, while also considering the temporal and spatial distribution of earthquakes. The stochastic debundling method based on this model does not simply determine whether each earthquake is a background event or a triggering event, but rather gives the probability of each event being a background event, thus effectively avoiding the subjectivity of manually selecting initial parameters.

[0004] However, existing earthquake catalog de-clusters generally suffer from drawbacks such as strong parameter dependence, insufficient feature utilization, and poor regional adaptability. The Gardner-Knopoff and Reasenberg methods heavily rely on fixed parameters based on historical data from specific regions, failing to adapt to the differences in tectonic features across different earthquake zones. While the ETAS statistical method is theoretically sound, its computational complexity is extremely high, it is prone to getting trapped in local optima, and its parameter estimation process is unstable. These traditional methods utilize only basic features such as time, space, and magnitude, ignoring the rich statistical information and spatial correlation features of earthquake sequences. Their classification accuracy does not exceed 90%, and their performance varies significantly when applied to different seismic tectonic regions, lacking intelligent parameter optimization and feature selection capabilities. Summary of the Invention

[0005] This invention aims to explore a novel method for de-clustering seismic catalogs based on intelligent optimization algorithms, addressing the shortcomings of traditional methods in feature selection and classification accuracy, and providing an intelligent de-clustering method and system for seismic catalogs based on optimization algorithms.

[0006] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent de-clustering method for seismic catalogs based on optimization algorithms, which includes the following steps:

[0008] S1. Obtain the original seismic catalog data of the target area after manual removal, and preprocess each original seismic catalog data to obtain the preprocessed dataset.

[0009] S2. Extract multidimensional features, including spatiotemporal distribution features, energy release features, and dynamic expansion features, from the preprocessed dataset, and normalize or standardize all extracted multidimensional features to obtain feature subsets.

[0010] S3. An improved particle swarm optimization algorithm is used to adaptively select features from the feature subset. The improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism. During the feature selection process, the classification accuracy of the random forest model is used as the fitness index to evaluate the quality of the feature subset. The key seismic features contained in the original seismic catalog data are always kept in the selected state. The particle swarm optimization algorithm iteratively optimizes and filters all feature subsets except for the key seismic features to obtain the optimal dynamic features. Finally, the optimal dynamic features are combined with the key seismic features to obtain the optimal feature subset.

[0011] S4. Use the optimal feature subset as input to train the random forest model. During the training process, use the particle swarm optimization algorithm to optimize the parameters of the random forest model to obtain the trained random forest model.

[0012] S5. The trained random forest model is used to classify and identify background earthquakes and aftershock sequences of the input earthquake events to be classified, thereby removing clusters from the earthquake catalog.

[0013] Based on the above scheme, each step can be implemented in the following preferred manner.

[0014] As a preferred embodiment of the first aspect mentioned above, in S1, the original earthquake catalog data includes the earthquake event's occurrence time, longitude, latitude, focal depth, and magnitude.

[0015] As a preferred embodiment of the first aspect above, in S1, the preprocessing includes three parts: the first part is to unify the time unit of the original earthquake catalog data; the second part is to calculate the fractal dimension, b-value and minimum complete magnitude; and the third part is to label the category of each earthquake event, wherein the category label is either a background earthquake or an aftershock sequence.

[0016] As a preferred embodiment of the first aspect above, in S2, the nearest neighbor distance algorithm is used to extract multidimensional features. The spatiotemporal distribution features include the time interval between two adjacent seismic events, the spatial distance calculated based on the latitude and longitude of the seismic events, and the spatial density representing the frequency of earthquakes within a unit area. The energy release features include magnitude and focal depth. The dynamic expansion features include fault correlation parameters and historical seismic activity background values.

[0017] As a preferred embodiment of the first aspect above, in S3, the key seismic features include the longitude, latitude, time of occurrence, and magnitude of the seismic event.

[0018] As a preferred option for the first aspect mentioned above, in S4, the parameters to be optimized in the random forest model include the number of decision trees, the maximum depth, and the node splitting criterion.

[0019] As a preferred option of the first aspect mentioned above, in S4, when optimizing the parameters of the random forest model, the parameters to be optimized in the random forest model are used as particle dimensions, and the particle position corresponds to a set of parameter combinations. The cross-validation accuracy on the optimal feature subset is used as the fitness index to evaluate the quality of the parameter combinations. Finally, the optimal parameter combination is found by iterative search to achieve the highest classification accuracy of the random forest model.

[0020] As a preferred embodiment of the first aspect mentioned above, the training of the random forest model is further evaluated using a k-fold cross-validation strategy: the dataset consisting of the optimal feature subset is randomly divided into k disjoint subsets, training and testing are repeated k times, and the mean of the k test results is calculated as the final performance metric; the final performance metric includes accuracy, recall and F1-Score.

[0021] Secondly, the present invention provides an intelligent de-clustering system for earthquake catalogs based on an optimization algorithm, comprising:

[0022] The data preprocessing module is used to obtain the original seismic catalog data of the target area after manual removal, and to preprocess each original seismic catalog data to obtain the preprocessed dataset.

[0023] The feature processing module is used to extract multidimensional features, including spatiotemporal distribution features, energy release features and dynamic expansion features, from the preprocessed dataset, and to normalize or standardize all the extracted multidimensional features to obtain feature subsets.

[0024] The adaptive feature selection module is used to adaptively select feature subsets using an improved particle swarm optimization algorithm. The improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism. During the feature selection process, the classification accuracy of the random forest model is used as a fitness index to evaluate the quality of the feature subsets. The key seismic features contained in the original seismic catalog data are always kept in the selected state. The particle swarm optimization algorithm iteratively optimizes and filters all feature subsets except for the key seismic features to obtain the optimal dynamic features. Finally, the optimal dynamic features are combined with the key seismic features to obtain the optimal feature subset.

[0025] The model training module is used to train the random forest model by taking the optimal feature subset as input and using the particle swarm optimization algorithm to optimize the parameters of the random forest model during the training process, so as to obtain the trained random forest model.

[0026] The classification and de-clustering module is used to classify and identify background earthquakes and aftershock sequences of the input earthquake events to be classified using a trained random forest model, thereby achieving de-clustering of the earthquake catalog.

[0027] Fifthly, the present invention provides a computer electronic device, which includes a memory and a processor;

[0028] The memory is used to store computer programs;

[0029] The processor is configured to, when executing the computer program, implement the intelligent de-clustering method for earthquake catalogs based on optimization algorithms as described in any of the solutions of the first aspect above.

[0030] Compared with traditional seismic catalog de-clustering methods (such as the Gardner and Knopoff spatiotemporal window method, the Reasenberg pseudo-random window method, the ETAS model, etc.) and single machine learning methods, the de-clustering scheme based on intelligent optimization algorithms in this invention has significant advantages in both technical principles and practical effects, specifically reflected in the following aspects:

[0031] I. This invention addresses the core pain points of traditional methods and enhances the scientific rigor of de-clumping techniques.

[0032] 1) This invention overcomes reliance on experience and achieves adaptive optimization of features and parameters. Traditional methods (such as the spatiotemporal window method and the ETAS model) require manual setting of key parameters (such as the size of the time window and the spatial decay coefficient). Parameter selection depends on domain experience, which can easily lead to bias in the de-furling results. This invention optimizes both feature selection and random forest parameters by improving the PSO algorithm: In the feature selection stage, PSO automatically searches for the optimal feature combination, avoiding the subjectivity of manual feature selection; in the model training stage, PSO iteratively optimizes the number and depth of decision trees in the random forest, without relying on preset empirical values.

[0033] 2) This invention overcomes the limitations of model assumptions and adapts to complex seismic activity. Traditional statistical methods such as the ETAS model are based on strict mathematical assumptions (such as aftershock attenuation following Omori's law and spatial isotropic distribution), making it difficult to explain atypical clustering phenomena. This invention, with data-driven approach at its core, uses a multi-dimensional feature system encompassing theoretically derived features and scalable dynamic features, capable of capturing subtle differences in complex seismic activity across different regions; random forests fit nonlinear relationships through ensemble decision trees, without requiring pre-set physical model assumptions.

[0034] II. This invention improves de-clustering accuracy and generalization ability, enhancing its practical application value.

[0035] 1) Multi-dimensional feature selection and fusion enhances classification accuracy. Traditional methods rely on single-dimensional features (such as the spatiotemporal window method which depends on time and spatial distance, and the ETAS model which focuses on magnitude-time correlation), easily overlooking the multi-factor coupling characteristics of seismic activity. The multi-dimensional feature system constructed in this invention covers inherent static features while also supporting the addition of dynamic features. It can be flexibly supplemented according to the seismological characteristics of the study area, exhibiting good scalability. When the model of this invention performs declination, it prioritizes fixing the most critical features for seismology and dynamically selects all remaining features to provide the feature combination most beneficial to declination accuracy. Due to the richness of feature dimensions and the rationality of the selection and filtering, the distinction between background earthquakes and clustered earthquakes is greatly improved, with an overall classification accuracy of over 95%, significantly higher than traditional methods (75%-85%).

[0036] 2) Cross-validation ensures model stability. This invention employs k-fold cross-validation to reduce the impact of data partitioning randomness on the results, resulting in smaller fluctuations in performance metrics.

[0037] III. Synergistic Improvement of Technological Innovation and Practicality

[0038] 1) This invention deeply integrates intelligent optimization with machine learning. Current approaches either indiscriminately pile up potentially useful features, leading to redundancy, or manually try to filter them one by one, but this is inefficient due to the large number of combinations. This invention combines the global optimization capability of PSO with the strong classification capability of random forests, forming a closed-loop mechanism of "feature selection - parameter optimization - cross-validation," representing an innovative application of intelligent algorithms in the field of seismology.

[0039] 2) This invention is adaptable to diverse application scenarios. This method can be directly applied to scenarios such as earthquake catalog preprocessing, seismic activity analysis, and medium- to long-term earthquake hazard assessment. After removing clusters from the historical earthquake catalog, the background seismic activity rate can be extracted more accurately, providing reliable input for earthquake probability models. Attached Figure Description

[0040] Figure 1 This is the overall flowchart of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the adaptive feature selection in the improved particle swarm optimization algorithm of this invention;

[0042] Figure 3 This is a schematic diagram of a confusion matrix provided in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the ROC curve of a trained random forest model on the validation set, as given in an embodiment of the present invention.

[0044] Figure 5 This is a schematic diagram showing the comparison of model classification accuracy given in an embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of the model performance evaluation results given in an embodiment of the present invention;

[0046] Figure 7 This is a system block diagram of the present invention. Detailed Implementation

[0047] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0048] In the description of this invention, it should be understood that the terms "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of those features.

[0049] Before describing the specific implementation process of this invention, the concepts involved in this invention will be explained below.

[0050] Earthquake Catalog: Based on the arrival times of seismic waves observed by regional seismic stations, corresponding earthquake event information is generated and compiled into an earthquake catalog. Basic information includes the time of occurrence (year, month, day, hour, minute, second), epicenter location (longitude, latitude), focal depth, magnitude, etc. The earthquake catalog for a certain region includes the mainshock, aftershocks caused by the mainshock, and earthquake swarms, exhibiting highly heterogeneous and variable spatiotemporal clustering characteristics.

[0051] Background earthquakes are relatively independent seismic events that are not closely related to other earthquakes in time and space. The spatial distribution of background earthquakes is generally considered to reflect the current stress state of the region.

[0052] Cluster earthquakes: a sequence of earthquakes that are closely related in time and space, including aftershock sequences triggered by the mainshock and sequences of earthquake swarms.

[0053] like Figure 1 As shown, in a preferred embodiment of the present invention, the above-mentioned intelligent de-clustering method for earthquake catalogs based on optimization algorithms includes the following steps S1 to S5. The specific implementation process of each step will be described in detail below.

[0054] S1. Obtain the original seismic catalog data of the target area after manual removal, and preprocess each original seismic catalog data to obtain the preprocessed dataset.

[0055] It should be noted that in S1 of this invention, the original earthquake catalog data includes the earthquake event's occurrence time, longitude, latitude, focal depth, and magnitude.

[0056] It should be noted that in S1 of this invention, the above preprocessing includes three parts: the first part is to unify the time unit of the original earthquake catalog data; the second part is to calculate the fractal dimension, b-value and minimum complete magnitude; and the third part is to label the category of each earthquake event, wherein the category label is the background earthquake or aftershock sequence.

[0057] In this embodiment S1, the specific process of preprocessing the original seismic catalog data is as follows:

[0058] 1) Select a portion of the earthquake catalog data (after manual declustering) for the target area between 2002 and 2013 and form a dataset;

[0059] 2) Based on the results of manual declustering, the background earthquake and aftershock sequences are marked with 1 and 0 respectively in the dataset;

[0060] 3) Unify the time (year, month, day, hour, minute, second) of earthquake catalog data to "day" as the unit;

[0061] 4) Calculate parameters such as fractal dimension (box counting method), b-value, and minimum complete magnitude for subsequent feature calculations and data filtering.

[0062] S2. Extract multidimensional features, including spatiotemporal distribution features, energy release features, and dynamic expansion features, from the preprocessed dataset, and normalize or standardize all extracted multidimensional features to obtain feature subsets.

[0063] It should be noted that in S2 of this invention, the nearest neighbor distance algorithm is used to extract multidimensional features. The aforementioned spatiotemporal distribution features include the time interval between two adjacent seismic events, the spatial distance calculated based on the latitude and longitude of the seismic events, and the spatial density representing the frequency of earthquakes within a unit area. The aforementioned energy release features include magnitude and focal depth. The aforementioned dynamic expansion features include fault correlation parameters and historical seismic activity background values.

[0064] In this embodiment S2, the essence of earthquake clustering is the aggregation characteristics of earthquake events in terms of spatiotemporal distribution and energy release. By extracting quantitative parameters that reflect these characteristics, a basic feature space for distinguishing between background earthquakes and clustered earthquakes can be constructed. To this end, in this invention S2, a multi-dimensional earthquake feature system is constructed: the original earthquake catalog data is transformed into feature vectors with discriminative power, providing a data foundation for subsequent classification and supporting dynamic expansion of the number of features.

[0065] The specific process of constructing the above-mentioned multidimensional earthquake characteristic system is as follows:

[0066] S21. Data Input and Feature Extraction

[0067] The preprocessed seismic catalog data for the target area is read in. Based on seismological theory (such as the nearest neighbor distance (NND) algorithm) and statistical analysis, multidimensional features are derived from the above seismic catalog data, covering the following dimensions:

[0068] 1) Spatiotemporal distribution characteristics: such as the time interval between earthquake events, spatial distance (calculated based on latitude and longitude), spatial density (frequency of earthquakes per unit area), etc.

[0069] 2) Energy release characteristics: such as magnitude, focal depth, etc.;

[0070] 3) Dynamically extended features: Based on the actual application scenario (such as the difference in seismic activity in different tectonic regions), features such as fault correlation parameters and historical seismic activity background values ​​can be added. The number of features is not fixed.

[0071] S22. Feature Processing

[0072] The extracted multidimensional features are normalized or standardized (or an appropriate processing method can be selected according to actual needs) to eliminate the impact of dimensional differences on subsequent algorithms.

[0073] S3. An improved particle swarm optimization algorithm is used to adaptively select features from the feature subset. The improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism into the particle swarm optimization algorithm. In the feature selection process, the classification accuracy of the random forest model is used as the fitness index to evaluate the quality of the feature subset. The key seismic features contained in the original seismic catalog data are always kept in the selected state. The particle swarm optimization algorithm iteratively optimizes and filters all feature subsets except for the key seismic features to obtain the optimal dynamic features. Finally, the optimal dynamic features are combined with the key seismic features to obtain the optimal feature subset.

[0074] It should be noted that in S3 of this invention, the key seismic features include the longitude, latitude, time of occurrence, and magnitude of the seismic event.

[0075] In this embodiment S3, the principle of Particle Swarm Optimization (PSO) will first be explained to facilitate understanding by those skilled in the art. PSO is a metaheuristic optimization algorithm based on swarm intelligence, proposed by James Kennedy and Russell Eberhart in 1995. It simulates the social behavior of biological groups such as flocks of birds and schools of fish, finding optimal solutions to complex optimization problems through cooperation and information sharing among particles.

[0076] In the process of adaptive feature selection using PSO, this invention considers that not all extracted features contribute to cluster identification, and some features may be redundant or interfering. Therefore, this invention introduces the PSO algorithm for feature selection. This algorithm, by simulating the swarm intelligence behavior of particle swarms, searches for the optimal feature subset (i.e., the most critical feature combination for classifying background earthquakes and cluster earthquakes) in the feature space, reducing feature dimensionality, improving model training efficiency, performance, and generalization ability, and maximizing the discriminative power of features for the classification task.

[0077] The specific process of adaptive feature selection using the PSO algorithm includes the following steps:

[0078] S31. Feature Coding

[0079] This invention treats each feature as a dimension and uses binary encoding to represent the particle position ("1" indicates that the feature is selected, and "0" indicates that it is not selected). The particle dimension is consistent with the total number of features.

[0080] S32. Fitness Function Design

[0081] This invention uses the classification accuracy of a random forest model as a fitness index to evaluate the quality of feature subsets. A higher fitness value indicates stronger discriminative power of the feature combination.

[0082] S33. PSO Optimization Process

[0083] Initialization: Randomly generate a certain number of particles (feature subset), and set the particle velocity, position range, and maximum number of iterations.

[0084] Iterative optimization: Particles update their position and velocity by tracking their individual optimal solutions (the best subset of features in their own history) and the global optimal solutions (the best subset of features in the history of the group), and continuously iterate to select the feature combination with the highest fitness.

[0085] In each iteration, the updates to particle position and particle velocity follow the formulas below:

[0086] ,

[0087] ;

[0088] in, and Particles In the The second iteration and the first The speed corresponding to the next iteration; parameters Inertial weight; These are individual learning factors and group learning factors, respectively. For particles Its own historical best position; and They represent particles respectively In the The second iteration and the first The feature selection state of the next iteration; The globally optimal position; The value is a random number in the interval [0,1]. Through this update rule, the particle swarm can balance exploration and development in the search space, thus gradually converging to the optimal feature subset.

[0089] Termination condition: Output the optimal feature subset when the maximum number of iterations is reached or the fitness value tends to stabilize.

[0090] In this embodiment S3, unlike traditional methods that use PSO, this invention introduces a fixed constraint feature mechanism in the above update process and designs an improved particle swarm optimization algorithm for adaptive feature selection of feature subsets, such as... Figure 2 As shown. Specifically, for key seismic features already confirmed in seismic knowledge (such as longitude, latitude, time, magnitude, etc.), they are always kept at "1" in the binary feature selection vector to ensure they are not discarded during optimization. The traditional PSO algorithm is then used to search for all other feature subsets besides the key seismic features. This approach fully utilizes PSO's ability to search for features of other variables while avoiding the loss of important information due to randomness, thus improving the model's robustness and interpretability.

[0091] S4. Use the optimal feature subset as input to train the random forest model. During the training process, use the particle swarm optimization algorithm to optimize the parameters of the random forest model to obtain the trained random forest model.

[0092] It should be noted that in S4 of this invention, the parameters to be optimized in the random forest model include the number of decision trees, the maximum depth, and the node splitting criterion.

[0093] It should be noted that in S4 of this invention, when optimizing the parameters of the random forest model, the parameters to be optimized in the random forest model are used as particle dimensions, and the particle position corresponds to a set of parameter combinations. The cross-validation accuracy on the optimal feature subset is used as the fitness index to evaluate the quality of the parameter combinations. Finally, the parameter combination that makes the random forest model have the highest classification accuracy is found through iterative search as the optimal parameter combination.

[0094] In this embodiment S4, considering that the PSO algorithm is essentially an optimization algorithm for finding the optimal solution, this invention applies the PSO algorithm to the optimal parameter combination search of the Random Forest (RF) model. The Random Forest model is briefly described below. Machine learning is a technique that allows computers to automatically learn patterns from data and make predictions or decisions. Common methods include linear models, neural networks, and decision trees. To improve the stability and accuracy of a single model, ensemble learning has emerged: combining multiple "weak models" into a "strong model." Random forest is a typical ensemble learning method. It trains multiple decision trees by randomly sampling data and features, and then votes (classification) or averages (regression) their predictions. This approach not only reduces overfitting but also has strong generalization ability, and therefore is widely used in many fields such as finance, healthcare, and recommendation systems.

[0095] Based on the aforementioned background knowledge, the random forest designed in this invention reduces the risk of overfitting by integrating the classification results of multiple decision trees and exhibits strong nonlinear classification capabilities. Optimizing key parameters of the random forest (number of decision trees, maximum depth, node splitting criteria, etc.) using the PSO algorithm further enhances model performance. Based on the selected optimal feature combination, high-precision classification of background earthquakes and clustered earthquakes can be achieved, overcoming the limitations of traditional methods that rely on empirical parameters.

[0096] The specific process of optimizing random forest parameters using the PSO algorithm includes the following steps:

[0097] The optimal feature combination is divided into a training set (70%) and a validation set (30%), and a random forest model is trained on the training set. During parameter optimization, the key parameters of the random forest model are used as particle dimensions, with each particle position corresponding to a set of parameter combinations. The cross-validation accuracy is used as the fitness metric to evaluate the quality of the parameter combinations. The PSO algorithm is used iteratively to find the parameter combination that achieves the highest classification accuracy for the random forest.

[0098] S5. The trained random forest model is used to classify and identify background earthquakes and aftershock sequences of the input earthquake events to be classified, thereby removing clusters from the earthquake catalog.

[0099] It should be noted that, further, the method of the present invention also includes evaluating the trained random forest model using a k-fold cross-validation strategy: the dataset consisting of the optimal feature subset is randomly divided into k disjoint subsets, training and testing are repeated k times, and the mean of the k test results is calculated as the final performance index; the final performance index includes accuracy, recall and F1-Score.

[0100] In this embodiment, a pre-trained random forest model is used to classify the input seismic events, outputting the distinction between "background earthquake" and "aftershock sequence". The overall de-clustering classification accuracy of the model is above 95%. The classification accuracy for aftershock sequences is relatively high, reaching 96.04%; the classification accuracy for background earthquakes is relatively lower, but still reaches 88.43%. The classification effect is described using a confusion matrix and ROC curves, as shown below. Figure 3 and Figure 4 As shown. Figure 3 In the confusion matrix, TN and TP represent true positives (correctly classified positives) and true negatives, respectively, while FN and FP represent false positives (misclassified positives) and false negatives, respectively.

[0101] like Figure 5As shown, this embodiment compares the currently mainstream ETAS model with the method of this invention. Both models were tested on the same dataset for debundling. The ETAS model achieved a classification accuracy of only 75.54%, while the method of this invention... Figure 5 The optimized PSO-RF classification accuracy was 95.2%.

[0102] In addition to the experiments described above, this invention also uses a k-fold cross-validation strategy to evaluate the random forest model: the dataset is randomly divided into k disjoint subsets, training and testing are repeated k times, and the mean of the k test results is calculated as the final performance metric (accuracy, recall, F1-score). The model performance evaluation results are as follows: Figure 6 As shown (here k=5).

[0103] In summary, the method of this invention can achieve adaptive adjustment of model parameters through particle swarm optimization, integrate multi-dimensional feature information and dynamically select the optimal feature combination, and achieve high-precision classification of over 95% and good regional adaptability while ensuring efficient computation, thus making up for the shortcomings of existing mainstream technologies.

[0104] It should also be noted that the intelligent de-clustering method for seismic catalogs based on optimization algorithms in the above embodiments can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides an intelligent de-clustering system for seismic catalogs based on optimization algorithms, corresponding to the intelligent de-clustering method for seismic catalogs based on optimization algorithms provided in the above embodiments, such as... Figure 7 As shown, it includes:

[0105] The data preprocessing module is used to obtain the original seismic catalog data of the target area after manual removal, and to preprocess each original seismic catalog data to obtain the preprocessed dataset.

[0106] The feature processing module is used to extract multidimensional features, including spatiotemporal distribution features, energy release features and dynamic expansion features, from the preprocessed dataset, and to normalize or standardize all the extracted multidimensional features to obtain feature subsets.

[0107] The adaptive feature selection module is used to adaptively select feature subsets using an improved particle swarm optimization algorithm. The improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism. During the feature selection process, the classification accuracy of the random forest model is used as a fitness index to evaluate the quality of the feature subsets. The key seismic features contained in the original seismic catalog data are always kept in the selected state. The particle swarm optimization algorithm iteratively optimizes and filters all feature subsets except for the key seismic features to obtain the optimal dynamic features. Finally, the optimal dynamic features are combined with the key seismic features to obtain the optimal feature subset.

[0108] The model training module is used to train the random forest model by taking the optimal feature subset as input and using the particle swarm optimization algorithm to optimize the parameters of the random forest model during the training process, so as to obtain the trained random forest model.

[0109] The classification and de-clustering module is used to classify and identify background earthquakes and aftershock sequences of the input earthquake events to be classified using a trained random forest model, thereby achieving de-clustering of the earthquake catalog.

[0110] It is understood that the intelligent de-clustering method for seismic catalogs based on optimization algorithms described in S1-S5 above can essentially be implemented by a computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the intelligent de-clustering method for seismic catalogs based on optimization algorithms provided in the above embodiments, which includes a memory and a processor;

[0111] The memory is used to store computer programs;

[0112] The processor is configured to implement the intelligent de-clustering method for earthquake catalogs based on optimization algorithms in the above embodiments when executing the computer program.

[0113] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0114] It is understood that 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.

[0115] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.

[0116] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for intelligent de-clustering of earthquake catalog based on optimization algorithm, characterized in that, The method comprises the following steps: S1, obtaining original earthquake catalog data of a target area after manual removal of aftershocks, and preprocessing each original earthquake catalog data to obtain a preprocessed data set; S2, extracting multi-dimensional features including time and space distribution features, energy release features and dynamic expansion features from the preprocessed data set, and performing normalization or standardization processing on all extracted multi-dimensional features to obtain a feature subset; S3, performing adaptive feature selection on the feature subset by using an improved particle swarm optimization algorithm, wherein the improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism into the particle swarm optimization algorithm; In the feature selection process, the classification accuracy of the random forest model is used as an adaptability index to evaluate the advantages and disadvantages of the feature subset, and the key earthquake features contained in the original earthquake catalog data are always kept in the selected state. All feature subsets except the key earthquake features are iteratively optimized and screened according to the particle swarm optimization algorithm to obtain the optimal dynamic features. Finally, the optimal dynamic features and the key earthquake features are combined to obtain the optimal feature subset; S4, using the optimal feature subset as the input of the random forest model, training the random forest model, and optimizing the parameters of the random forest model during the training process by using the particle swarm optimization algorithm to obtain the trained random forest model; S5, using the trained random forest model to classify and identify background earthquakes and aftershock sequences of inputted seismic events, and realizing earthquake catalog de-clustering.

2. The optimization algorithm-based intelligent de-clustering method of earthquake catalogues according to claim 1, characterized in that, In S1, the original earthquake catalog data includes the occurrence time, longitude, latitude, focal depth and magnitude of the seismic event.

3. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 1, characterized in that, In S1, the preprocessing includes three parts: the first part is to unify the time unit of the original earthquake catalog data, the second part is to calculate the fractal dimension, b value and minimum complete magnitude, and the third part is to label the class label of each seismic event, which is a background earthquake or an aftershock sequence.

4. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 2, characterized in that, In S2, the nearest neighbor distance algorithm is used to extract multi-dimensional features, the time and space distribution features include the time interval between two adjacent seismic events, the spatial distance calculated based on the longitude and latitude of the seismic event, and the spatial density representing the frequency of earthquakes in a unit area, the energy release features include the magnitude and focal depth, and the dynamic expansion features include the fault correlation parameter and the historical seismic activity background value.

5. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 2, characterized in that, In S3, the key earthquake features include the longitude, latitude, occurrence time and magnitude of the seismic event.

6. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 1, characterized in that, In S4, the parameters to be optimized in the random forest model include the number of decision trees, the maximum depth and the node splitting standard.

7. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 6, characterized in that, In S4, when optimizing the parameters of the random forest model, the parameters to be optimized in the random forest model are used as the particle dimension, the particle position corresponds to a group of parameter combinations, the cross-validation accuracy on the optimal feature subset is used as the adaptability index to evaluate the advantages and disadvantages of the parameter combinations, and finally the parameter combination that makes the classification accuracy of the random forest model highest is found as the optimal parameter combination through iterative search.

8. The optimization algorithm based intelligent de-clustering method of earthquake catalogues according to claim 1, characterized in that, Also included is the use of a k-fold cross-validation strategy to evaluate the trained random forest model: the data set consisting of the optimal feature subset is randomly divided into k non-intersecting subsets, and the training and testing are repeated k times, and the mean of the k test results is calculated as the final performance indicator; the final performance indicator includes accuracy, recall rate and F1-Score.

9. A seismic catalog intelligent de-clustering system based on optimization algorithms, characterized in that, It comprises: A data preprocessing module is configured to obtain original earthquake catalog data of a target area after manual removal, and to preprocess each original earthquake catalog data to obtain a preprocessed data set. A feature processing module is configured to extract multi-dimensional features including spatio-temporal distribution features, energy release features and dynamic expansion features from the preprocessed data set, and to normalize or standardize all extracted multi-dimensional features to obtain a feature subset. An adaptive feature selection module is configured to use an improved particle swarm optimization algorithm to perform adaptive feature selection on the feature subset, and the improved particle swarm optimization algorithm is obtained by introducing a fixed constraint feature mechanism to the particle swarm optimization algorithm. During the feature selection process, the classification accuracy of the random forest model is used as the fitness indicator to evaluate the pros and cons of the feature subset, and the key earthquake features contained in the original earthquake catalog data are always kept in the selected state, and all feature subsets except the key earthquake features are iteratively optimized and screened according to the particle swarm optimization algorithm to obtain the optimal dynamic feature, and finally the optimal dynamic feature and the key earthquake feature are combined to obtain the optimal feature subset. A model training module is configured to use the optimal feature subset as the input of the random forest model, train the random forest model, and use the particle swarm optimization algorithm to optimize the parameters of the random forest model during the training process to obtain a trained random forest model. A classification and identification de-clustering module is configured to use the trained random forest model to classify and identify background earthquakes and aftershock sequences of inputted seismic events to be classified, and to realize earthquake catalog de-clustering.

10. A computer electronic device, comprising: It comprises a memory and a processor; The memory is configured to store a computer program; The processor is configured to execute the computer program to implement the intelligent earthquake catalog de-clustering method based on the optimization algorithm according to any one of claims 1-8.

Images