An engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method and system

Abstract

The application discloses an engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method and system, and the early warning method comprises the following steps: arranging a microseismic sensor to collect rock fracturing earthquake-causing original waveform data, extracting waveform characteristic parameters and a focal mechanism parameter as a data set after wavelet transform filtering processing; inputting the data set into a TFT regression model to obtain space-time strong multi-dimensional parameter early warning index data during rock fracturing earthquake-causing; inputting characteristic variables in the data set and a target variable about whether earthquake-causing occurs into a random forest classification model to obtain a predicted rock fracturing earthquake-causing probability; finally, integrating the space-time strong multi-dimensional data obtained by the TFT regression prediction and the earthquake-causing probability obtained by the random forest classification into a new data set, training and evaluating the new data set by using a multi-layer perception MLP and a Transformer fusion model, obtaining rock fracturing earthquake-causing multi-parameter fusion early warning results by weighted fusion, and realizing space-time strong multi-dimensional advanced early warning of a reservoir liquid injection fracturing induced earthquake.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fracturing induced earthquake early warning, and in particular to a multi-information fusion early warning method and system for engineering rock mass fracturing induced earthquake. BACKGROUND

[0002] Hot dry rock, as a renewable energy source, has great potential for development. By using hydraulic fracturing technology, the heat in hot dry rock can be extracted to provide thermal energy for power generation. However, the hydraulic fracturing process may induce seismic activity, especially during the hot dry rock fracturing process, as the injection of high-pressure water changes the stress state of the underground rock, leading to the occurrence of earthquakes. This induced earthquake phenomenon not only poses a threat to equipment and personnel safety, but also may affect the environmental stability of the surrounding area. Therefore, it is of great significance to research and develop a method and system that can effectively early warn hot dry rock fracturing induced earthquakes, in order to ensure the safety and stability of hot dry rock development.

[0003] Microseismic / acoustic emission monitoring technology plays an irreplaceable important role in monitoring hot dry rock induced earthquakes. These technologies can monitor the generation and propagation of microcracks in rock bodies in real time, providing early warning signals of seismic activity. Microseismic / acoustic emission monitoring technology uses sensors placed underground or on the ground to capture tiny vibration signals, and through data analysis, identifies and locates seismic events. Compared with traditional seismic monitoring methods, microseismic / acoustic emission technology has higher sensitivity and resolution, and can detect extremely small-scale rock failure events, making it an ideal choice for hot dry rock induced earthquake early warning. For example, patent CN114779330A discloses a method for analyzing and predicting the orientation of main cracks in mining working face based on microseismic monitoring; patent CN115983465A discloses a method for constructing a time series prediction model for rock burst based on small sample learning; and patent CN116307683A provides an evaluation method for water fracturing activated fault induced earthquakes. The above achievements provide a reference for microseismic / acoustic emission monitoring and early warning of hot dry rock fracturing induced earthquakes.

[0004] However, the above methods for monitoring and early warning of rock fracturing induced seismic activity have the following shortcomings: First, the analysis method mainly relies on a single parameter for early warning, and does not fully consider the multi-parameter characteristics of seismic activity, resulting in low accuracy and reliability of early warning. Second, existing methods often ignore the complex dependence of time series data when processing seismic data, and do not fully utilize the potential information of time series data. In addition, existing early warning systems mostly fail to realize multi-model fusion, making it difficult to comprehensively utilize the advantages of different models and improve the overall performance of the early warning system.

[0005] Based on the above problems, an engineering rock mass fracturing earthquake-inducing multi-information fusion intelligent early warning method and system are urgently needed, which aims to realize accurate early warning of rock fracturing induced earthquakes through multi-parameter fusion, multi-model collaboration and deep mining of time series data, improve the reliability and stability of the early warning system, and provide strong support for ensuring the safety and sustainability of hot dry rock development. SUMMARY

[0006] The present scheme aims to address the problems and needs raised above and proposes an engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method and system. The above technical purposes are achieved and other technical effects are brought about due to the adoption of the following technical features.

[0007] One object of the present application is to propose an engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method, comprising the following steps:

[0008] S10: Collecting original waveform data induced by the rock fracturing earthquake-inducing process, filtering the obtained original waveform data to obtain effective waveform data;

[0009] S20: Extracting feature coefficients from the effective waveform data, and performing source mechanism inversion to obtain time-space strong multi-dimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and using the original waveform data induced by fracturing and the time-space strong multi-dimensional early warning index data as a data set;

[0010] S30: Inputting the data set into a TFT regression model for regression analysis, and pre-processing by normalization and missing value filling, training the model and evaluating the model performance to obtain predicted time-space strong multi-dimensional data;

[0011] S40: Extracting feature variables from the time-space strong multi-dimensional early warning index data set, and obtaining target variables about whether a seismic event occurs through analysis of historical microseismic data; pre-processing the extracted feature variables and target variables about whether a seismic event occurs, and then inputting them into a random forest classification model for classification analysis, using a multi-decision tree integration method for danger classification, training and performance evaluation of the pre-processed data through the random forest classification model to obtain a predicted seismic probability;

[0012] S50: Taking the predicted time-space strong multi-dimensional data and the seismic probability obtained by danger classification as input data, re-integrating to establish a new data set, and dividing it into a training set and a test set, using a multi-layer perception MLP and Transformer fusion model to train and evaluate the training set data, and using the test set data to weight and fuse the outputs of the multi-layer perception MLP and Transformer to obtain a rock fracturing earthquake-inducing multi-parameter fusion early warning result.

[0013] In addition, the engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method according to the present application can have the following technical features.

[0014] In one example of the present application, in step S10, the obtained original waveform data is filtered to obtain effective waveform data, specifically including the following steps:

[0015] S11: decompose the original waveform data to determine the wavelet base function and the decomposition layer number;

[0016] S12: perform threshold processing on the decomposed signal to remove noise components;

[0017] S13: reconstruct the signal after removing the noise to obtain the filtered effective waveform data.

[0018] In one example of the present application, in step S20, the effective waveform data is subjected to feature coefficient extraction and focal mechanism inversion to obtain time-space strong multi-dimensional early warning index data of waveform feature parameters, focal location and focal mechanism parameters, including:

[0019] S21: perform feature recognition on the induced waveform to extract waveform feature coefficients, including ring count, energy release rate, frequency band energy distribution, fractal feature, earthquake occurrence time and initial motion amplitude;

[0020] S22: perform inversion on the focal mechanism parameters, including rupture occurrence location index and rupture occurrence intensity index, wherein the rupture occurrence location index includes spatial concentration factor, spatial distribution intensity coefficient, error spatial form, tension-shear attribute, rupture surface occurrence and extension direction, and the rupture occurrence intensity index includes seismic moment magnitude, b value, stress drop and corner frequency;

[0021] S23: establish a data set according to the waveform feature coefficient data and the focal mechanism parameter data.

[0022] In one example of the present application, the step S30 specifically includes the following steps:

[0023] S31: construct a regression data set by combining the waveform feature coefficients and the focal mechanism parameters in the effective waveform data to form a regression data set;

[0024] S32: construct a TFT regression model and input the regression data set into the TFT regression model for regression prediction analysis;

[0025] S33: performance evaluation and result prediction, quantitatively evaluate the prediction results of the model by using the mean absolute scaled error MASE; output multi-parameter prediction data and prediction interval through the trained model to obtain the future seismic activity trend.

[0026] In an example of the present application, the TFT regression model comprises:

[0027] a data processing and feature selection module configured to select the most significant features from the input using a variable selection network (VSN), encode the context vectors of static covariates through a static covariate encoder (SCE);

[0028] an information flow modeling module configured to ensure effective information flow using a gated residual network (GRN), learn long-term dependencies of time series data using a time self-attention layer (TSL), and perform additional nonlinear processing on the output of the self-attention layer through a position-oriented feedforward layer (PFL);

[0029] a decoding and prediction module configured to enhance temporal features using a static enrichment layer (SEL) in a temporal fusion decoder (TFD);

[0030] a training optimization module configured to use mean squared error (MSE) as a loss function during training, use an Adam optimizer for parameter optimization, and update model parameters iteratively through a mini-batch stochastic gradient descent algorithm.

[0031] In an example of the present application, in S40, the extracted feature variables and target variables related to whether a seismic event occurs in the data set are preprocessed and then input into a random forest classification model for classification analysis, specifically including the following steps:

[0032] S41: Data set division and data preprocessing: combine the extracted feature variables and target variables related to whether a seismic event occurs in the spatial and temporal strong multi-dimensional early warning index data set to form a classification data set;

[0033] S42: Constructing a random forest model: generate multiple sub-samples from the classification data set through bootstrap sampling, train a decision tree for each sub-sample, reduce the risk of overfitting by integrating multiple trees, and finally determine the final class through majority voting;

[0034] S43: Model training and evaluation: use a cross-entropy loss function to optimize the model, calculate the classification effect of each tree during training; evaluate the performance of the model through evaluation indicators and analyze the classification performance of the model through a confusion matrix; wherein the evaluation indicators include: accuracy, recall rate, and comprehensive correct rate;

[0035] S44: Use the trained random forest model to classify and predict future seismic data, and output the probability of a seismic event occurring at each time point.

[0036] In an example of the present application, the accuracy A, recall rate R, and comprehensive correct rate F1 are calculated as follows:

[0037]

[0038] In the formula, TP is a true positive: true is a seismic event and the model recognition result is a seismic event, that is, the seismic phenomenon recognition is correct; FP is a false positive: true is not a seismic event but the model recognition result is a seismic event, that is, the non-seismic phenomenon recognition is incorrect; TN is a true negative: true is not a seismic event and the model recognition result is not a seismic event, that is, the non-seismic phenomenon recognition is correct; and FN is a false negative: true is a seismic event but the model recognition result is not a seismic event, that is, the seismic phenomenon recognition is incorrect.

[0039] In an example of the present application, in the step S50, the training set data is trained and evaluated using a multi-layer perception MLP and a Transformer fusion model, and the specific classification steps are as follows:

[0040] S51: Constructing a data set: integrating the output data of the TFT regression model and the random forest classification model, that is, the predicted spatiotemporal strong multidimensional data and the seismic probability, into a new data set;

[0041] S52: Constructing an MLP and a Transformer fusion model: training the MLP model and the Transformer model respectively, automatically selecting the MLP or the Transformer model according to the task characteristics of regression or classification, determining the best hyperparameter combination, further optimizing the hyperparameters, calculating the error of the training set and the validation set, judging whether the model is overfitting or underfitting, and evaluating the model performance;

[0042] S53: Outputting a rock fracturing seismic multi-parameter fusion early warning result: weighting and fusing the predicted spatiotemporal strong multidimensional data output by the MLP and the Transformer model and the classified predicted seismic probability, and finally generating a rock fracturing seismic multi-dimensional parameter fusion early warning result related to waveform difference features, source occurrence time, location and intensity.

[0043] In an example of the present application, constructing an MLP and a Transformer fusion model includes the following steps:

[0044] S521: Model training: when training the MLP, using mean square error MSE as the loss function, using the Adam optimizer to optimize the parameters, and updating the model parameters through the small batch stochastic gradient descent algorithm; when training the Transformer, using the cross-entropy loss function, using the Adam optimizer to optimize, and capturing the complex relationship between features through the multi-head attention mechanism;

[0045] S522: Model selection and hyperparameter optimization: automatically selecting the MLP or the Transformer model according to the task characteristics of regression or classification, determining the best hyperparameter combination through cross-validation, and further optimizing the hyperparameters to optimize the performance of the model;

[0046] S523: Model evaluation: calculate the error of the training set and the validation set in the training by MSE and cross entropy, judge whether the model is overfitting or underfitting, use root mean square error RMSE to evaluate the model performance for regression task, and use accuracy, precision and recall to evaluate for classification task.

[0047] Another object of the present application is to provide an engineering rock mass fracturing earthquake-inducing multi-information fusion early warning system, comprising:

[0048] The acquisition module is configured to acquire original waveform data induced by the rock fracturing induced earthquake process, and filter the obtained original waveform data to obtain effective waveform data.

[0049] The processing module is configured to extract feature coefficients from the effective waveform data and perform source mechanism inversion to obtain spatial and temporal strong multi-dimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and use the original waveform data induced by fracturing and the spatial and temporal strong multi-dimensional early warning index data as a data set.

[0050] The regression model module is configured to input the data set into the TFT regression model for regression analysis, and use normalization and missing value filling for preprocessing, train the model and evaluate the model performance to obtain predicted spatial and temporal strong multi-dimensional data.

[0051] The classification model module is configured to extract feature variables from the spatial and temporal strong multi-dimensional early warning index data set, obtain target variables about whether an earthquake occurs through analysis of historical microseismic data, preprocess the extracted feature variables and target variables about whether an earthquake occurs, and then input them into the random forest classification model for classification analysis, use the multi-decision tree integration method for danger classification, train and evaluate the preprocessed data through the random forest classification model, and obtain the predicted earthquake probability.

[0052] The weighted fusion module is configured to use the predicted spatial and temporal strong multi-dimensional data and the earthquake probability obtained by danger classification as input data, reorganize to establish a new data set, and divide it into a training set and a test set, use the multi-layer perception MLP and the Transformer fusion model to train and evaluate the training set data, use the test set data to weight the outputs of the multi-layer perception MLP and the Transformer, and obtain the rock fracturing induced earthquake multi-parameter fusion early warning result.

[0053] Compared with the prior art, the technical scheme has the following beneficial effects:

[0054] The technical scheme obtains and processes induced waveform data of dry hot rock fracturing induced earthquakes, uses wavelet transform filtering technology to obtain effective waveform data, and further extracts characteristic coefficients and focal mechanism parameters. These multi-parameter data are input into the TFT regression model for regression analysis, combined with the random forest classification model for classification analysis, and through the fusion of the multi-parameter prediction and the earthquake probability analysis of the model, the possibility of future earthquakes can be more accurately evaluated. The technical scheme effectively solves the data noise and non-linear problems in the traditional method, significantly improves the overall performance of the early warning system, and provides a scientific reference for the early warning of dry hot rock fracturing induced earthquakes.

[0055] The technical scheme comprehensively utilizes wavelet transform filtering, characteristic coefficient extraction, focal mechanism parameter inversion and other technical means to perform multi-dimensional analysis on the seismic waveform data induced by dry hot rock fracturing. Through the joint use of regression model and classification model, not only the future magnitude and energy release rate and other multi-parameter data can be predicted, but also the probability of future earthquake can be evaluated. The fusion model further integrates the regression prediction and classification analysis results, realizing comprehensive early warning of dry hot rock fracturing induced earthquakes. The technical scheme has significant advantages in data processing, model training and prediction evaluation, and can effectively improve the accuracy and stability of earthquake early warning, providing important technical support for earthquake safety monitoring during dry hot rock mining.

[0056] The fusion model of the technical scheme uses the advantages of MLP and Transformer to capture the non-linear features and time-dependent relationships in the data, improving the accuracy and stability of the prediction. Through the post-fusion model, the prediction results of regression and classification models are comprehensively utilized to improve the overall performance of the early warning system. MLP and Transformer models are suitable for different data features and task requirements, providing more flexible choices. The post-fusion model can reduce the bias and error of a single model, improve the stability and accuracy of the prediction, and finally realize multi-parameter fusion early warning of dry hot rock fracturing induced earthquakes through the fusion of regression prediction and hazard classification results.

[0057] The optimal embodiments of the present application will be described in more detail below with reference to the accompanying drawings, so that the features and advantages of the present application can be easily understood. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. The drawings are only used to show some embodiments of the present application, and not to limit all embodiments of the present application to this.

[0059] Figure 1 The flowchart of an engineering rock mass fracturing induced earthquake multi-information fusion early warning method according to an embodiment of the present application;

[0060] Figure 2 Wavelet transform filtering flow chart according to an embodiment of the present application;

[0061] Figure 3 FTF regression model structure schematic diagram according to an embodiment of the present application;

[0062] Figure 4 Random forest model structure schematic diagram according to an embodiment of the present application;

[0063] Figure 5 Space-time strong multi-dimensional parameter early warning index relationship diagram according to an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical scheme and advantages of the technical scheme of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in conjunction with the drawings of the specific embodiments of the present application. The same reference signs in the drawings represent the same parts. It should be noted that the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0065] Unless otherwise defined, technical terms or scientific terms used herein should be understood as their ordinary meanings to those of ordinary skill in the art to which the present application belongs. The terms "first", "second" and similar terms used in the description and claims of the present patent application do not necessarily mean any order, number or importance, but are only used to distinguish different components. Similarly, the terms "one" or "a" and similar terms do not necessarily mean a quantity limit. The terms "including" or "containing" and similar terms mean that the elements or objects before the term cover the elements or objects listed after the term and their equivalents, without excluding other elements or objects. The terms "connected" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like only represent relative positional relationships, which may change accordingly when the absolute position of the described object changes.

[0066] According to the first aspect of the present application, an engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method is provided, as shown in the figure, comprising the following steps: Figure 1

[0067] ​S10: According to the actual situation, arrange microseismic sensors, collect the original waveform data induced by rock fracturing induced seismic process, filter the obtained original waveform data to obtain effective waveform data; for example, wavelet transform can be used for filtering processing;

[0068] S20: Extracting feature coefficients and focal mechanism inversion are performed on the effective waveform data to obtain the time-space strong multi-dimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and the original waveform data induced by fracturing and the time-space strong multi-dimensional early warning index data are used as a data set;

[0069] S30: The data set is input into the TFT regression model for regression analysis, and the normalization and missing value filling method is used for preprocessing, the model is trained and the model performance is evaluated, and the predicted time-space strong multi-dimensional data is obtained to provide data for subsequent early warning of future seismic activity trend;

[0070] S40: Extracting feature variables from the time-space strong multi-dimensional early warning index data set, obtaining target variables about whether to occur induced seismicity through analysis of historical microseismic data; the extracted feature variables and target variables about whether to occur induced seismicity in the data set are preprocessed, and then input into a random forest classification model for classification analysis, and a multi-decision tree integration method is used for danger classification to solve data noise and nonlinear problems. Through random forest classification model training and performance evaluation of preprocessed data, the predicted induced seismic probability is obtained to provide data for subsequent early warning of future earthquake possibility;

[0071] S50: The predicted time-space strong multi-dimensional data and the induced seismic probability obtained by danger classification are used as input data to re-integrate and establish a new data set, and are divided into training set and test set. The multi-layer perception MLP and Transformer fusion model are used to train and evaluate the training set data, and the outputs of the multi-layer perception MLP and Transformer are weighted and fused using the test set data to obtain the rock fracturing induced seismic multi-parameter fusion early warning result.

[0072] The method obtains and processes the induced waveform data of hot dry rock fracturing induced seismicity, uses wavelet transform filtering technology to obtain effective waveform data, and further extracts feature coefficients and source mechanism parameters. These multi-parameter data are input into the TFT regression model for regression analysis, combined with the random forest classification model for classification analysis, through the multi-parameter prediction and induced seismic probability analysis of the fusion model, the possibility of future earthquake occurrence can be more accurately evaluated. The method effectively solves the data noise and nonlinear problems in the traditional method, significantly improves the overall performance of the early warning system, and provides a scientific reference for the early warning of hot dry rock fracturing induced earthquake.

[0073] The method comprehensively utilizes wavelet transform filtering, feature coefficient extraction, and focal mechanism parameter inversion, and performs multi-dimensional analysis on the seismic waveform data induced by dry hot rock fracturing. Through the joint use of regression model and classification model, not only the future magnitude and energy release rate and other multi-parameter data can be predicted, but also the probability of future seismicity can be evaluated. The fusion model further integrates the results of regression prediction and classification analysis, realizing comprehensive early warning of dry hot rock fracturing induced seismicity. The method has significant advantages in data processing, model training and prediction evaluation, and can effectively improve the accuracy and stability of earthquake early warning, providing important technical support for seismic safety monitoring in dry hot rock mining process.

[0074] The fusion model of the method utilizes the advantages of MLP and Transformer to capture the nonlinear features and time-dependent relationships in the data, improving the accuracy and stability of the prediction. Through the post-fusion model, the prediction results of regression and classification models are comprehensively utilized to improve the overall performance of the early warning system. MLP and Transformer models are suitable for different data features and task requirements, providing more flexible choices. The post-fusion model can reduce the bias and error of a single model, improve the stability and accuracy of the prediction, and ultimately realize multi-parameter fusion early warning of dry hot rock fracturing induced seismicity through the fusion of regression prediction and hazard classification results.

[0075] In one example of the present application, in step S10, the obtained original waveform data is subjected to wavelet transform filtering to obtain effective waveform data, as shown in Figure 2 The specific steps include the following:

[0076] S11: The original waveform data is decomposed to determine the wavelet basis function and the number of decomposition layers; for example, Daubechies db4 and db6 wavelets are selected as the basis function, and the formula is as follows:

[0077]

[0078] where ψ(t) is the Daubechies wavelet function, including db4 and db6 wavelet functions, φ(t) is the approximation function, b k is the filter coefficient of ψ(t), db4 wavelet k=4, M=4, db6 wavelet k=6, M=4. Select 3 to 6 layers of wavelet decomposition to decompose the original signal into different frequency band components.

[0079] S12: The decomposed signal is subjected to threshold processing to remove noise components; specifically, the high-frequency components are subjected to threshold processing, the threshold value is set according to the standard deviation of the noise, and finally all components are combined to obtain the noise-removed signal.

[0080] S13: reconstruct the signal after removing noise to obtain filtered effective waveform data. Specifically, the low frequency and high frequency components of the signal after removing noise are combined, and the filtered effective waveform data is recovered by inverse wavelet transform.

[0081] In one example of the present application, in step S20, the effective waveform data is subjected to feature coefficient extraction and focal mechanism inversion to obtain time-space strong multi-dimensional early warning index data of waveform feature parameters, focal location and focal mechanism parameters, including:

[0082] S21: feature recognition of induced waveform, extraction of waveform feature coefficients, including ringing count, energy release rate, frequency band energy distribution, fractal feature, earthquake occurrence time and initial motion amplitude; wherein the rupture occurrence time index includes earthquake occurrence time and initial motion amplitude;

[0083] Specifically, the ringing count is the number of times exceeding the preset threshold; the energy release rate is the driving force for judging crack propagation, and the formula is as follows: the frequency band energy distribution formula is as follows: P(f) = |N(f)| 2 , wherein P(f) is the energy density at frequency f, and X(f) is the Fourier transform value of the signal at frequency f; the fractal feature is described by the box dimension in the fractal dimension, and the formula is: , wherein N(x) is the number of boxes with radius x required to cover the waveform signal; the earthquake occurrence time is the time when the first wave peak or trough exceeding the set threshold appears; the initial motion amplitude is the amplitude of the first wave peak or trough in the initial stage of the waveform signal.

[0084] S22: inversion of focal mechanism parameters, including rupture occurrence location index and rupture occurrence intensity index, wherein the rupture occurrence location index includes spatial concentration factor, spatial distribution intensity coefficient, error spatial form, tension-shear attribute, rupture surface occurrence, and extension direction, and the rupture occurrence intensity index includes seismic moment magnitude, b value, stress drop, corner frequency; the time-space strong multi-dimensional parameter early warning index relationship diagram is as shown in Figure 5 ,

[0085] Specifically, (1) the rupture occurrence location index is as follows:

[0086] Spatial concentration factor: the spatial concentration degree of stress distribution at the source, and the formula is as follows:

[0087]

[0088] , wherein N is the number of rupture sources in unit volume V;

[0089] Spatial distribution intensity coefficient: the spatial intensity of stress distribution at the source, and the formula is as follows:

[0090]

[0091] where X i is the distance from the source to the center of volume, g i is the energy of the source i;

[0092] Error space pattern: the spatial difference between the theoretical calculation value and the actual observation value in the inversion process,

[0093]

[0094] where, the observed value of the jth component of the ith station, is the corresponding theoretical value, n is the number of stations, and m is the number of components;

[0095] Tension-shear property: the relative action of tension and shear in the rupture process, the formula is as follows:

[0096]

[0097] where L is the tension-shear ratio: L>1, the tensile fracture property is dominant; L<1, corresponding to the shear fracture property is dominant;

[0098] Rupture scale: the scale of rupture, the formula is as follows:

[0099]

[0100] where ΔA is the rupture area, M1 and M3 are the eigenvalues of the moment tensor, μ is the Lame constant, and ξ is the proportional coefficient of rupture area and displacement;

[0101] Rupture surface occurrence: the spatial orientation of the rupture surface, the formula is as follows:

[0102]

[0103] n=(cosβ,0,±sinβ)

[0104] where β is the angle between the motion direction and the normal direction of the source rupture surface, M1, M2, and M3 are the eigenvalues of the moment tensor, μ and λ are the Lame constants, and n is the spatial orientation;

[0105] Extension direction: the motion direction of the rupture surface, the formula is as follows:

[0106]

[0107] where β is the angle between the motion direction and the normal direction of the source rupture surface, and b is the motion direction;

[0108] (2) The rupture occurrence intensity index is as follows:

[0109] Moment magnitude: describes the size of an earthquake, formula as follows:

[0110]

[0111] Where M0 is the seismic moment,

[0112] b value: describes the frequency-size distribution of seismic activity, formula as follows:

[0113] b = log 10 (N) / M

[0114] Where N is the number of earthquakes greater than or equal to a certain magnitude M;

[0115] Stress drop: the change in the final state of the shear stress of a rupture surface before and after the rupture, formula as follows:

[0116]

[0117] Where M0 is the seismic moment, and R is the source radius;

[0118] Corner frequency: obtained by fitting the high-frequency part of the seismic wave spectrum, and its relationship with the seismic moment M0 and stress drop Δσ is as follows:

[0119] fc = 1 / 2π(Δσ / ρV s 3 ) 1 / 2 ,

[0120] Where ρ is the rock density, and Vs is the shear wave velocity.

[0121] S23: establishing a data set according to the waveform feature coefficient data and the focal mechanism parameter data.

[0122] In an example of the present application, the step S30 specifically comprises the following steps:

[0123] S31: constructing a regression data set, combining the waveform feature coefficients in the effective waveform data and the focal mechanism parameters to form a regression data set; the data set uses a table form, each row represents a data record at a time point, and each record contains all features at a time point. The Pandas library is used to load the data set, the feature variables are normalized, and missing data points are processed by linear interpolation, mean filling and other methods to ensure data integrity and consistency.

[0124] S32: constructing a TFT regression model, inputting the regression data set into the TFT regression model for regression prediction analysis;

[0125] The TFT regression model is selected because the TFT model enhances the time feature capturing ability through a time fusion decoder (TFD), and is particularly suitable for long time series prediction of hot dry rock seismic activity.

[0126] The TFT regression model is configured by constructing a TFT regression model, including:

[0127] The input multi-parameter data is 17-dimensional multi-parameter data including ring count, energy release rate, duration, frequency band energy distribution, fractal feature, earthquake occurrence time, initial motion amplitude, spatial concentration factor, spatial distribution intensity coefficient, error spatial shape, shear attribute, rupture scale, rupture surface occurrence, seismic moment magnitude, corner frequency, b value and extension direction; the number of neurons in the hidden layer is determined to be 256 through a grid search method or Bayesian optimization; the number of heads of the multi-head attention mechanism is 8; the number of layers of the Transformer is 4; and the output is 17-dimensional multi-parameter data at a future time point.

[0128] S33: Performance evaluation and result prediction, the prediction results of the model are quantitatively evaluated by using the mean absolute scaled error (MASE); the multi-parameter prediction data and the prediction interval are output by the trained model, and the future seismic activity trend is obtained, thereby providing data support for subsequent earthquake early warning. The mean absolute scaled error (MASE) is defined as follows:

[0129] Suppose y represents the true value, the predicted value, and n is the sample number, then the calculation formula of MASE is:

[0130]

[0131] The smaller the MASE value is, the better the prediction performance of the model is.

[0132] In one example of the present application, as shown in Figure 3 The TFT regression model includes:

[0133] The data processing and feature selection module is configured to select the most significant features from the input using a variable selection network (VSN) and encode the context vector of the static covariate through a static covariate encoder (SCE);

[0134] The information flow modeling module is configured to use a gated residual network (GRN) to ensure effective information flow, use a time self-attention layer (TSL) to learn the long-term dependence relationship of the time series data, and perform additional nonlinear processing on the output of the self-attention layer through a position-oriented feedforward layer (PFL);

[0135] The decoding and prediction module is configured to enhance the time features in the static enrichment layer (SEL) of the time fusion decoder (TFD);

[0136] The training optimization module is configured to use mean square error (MSE) as a loss function during training, use an Adam optimizer to optimize parameters, and iteratively update model parameters through a small batch stochastic gradient descent algorithm.

[0137] In one example of the present application, in the S40, the extracted feature variables and the target variable related to whether a shock occurs in the data set are preprocessed and then input into a random forest classification model for classification analysis, specifically including the following steps:

[0138] S41: Data set division and data preprocessing: the feature variables and the target variable related to whether a shock occurs in the time-space strong multi-dimensional early warning index data set are combined to form a classification data set; the data set is in table form, each row representing a data record at a time point, including feature variables and target variables. The Pandas library is used to load the classification data set, and the feature variables are normalized to ensure that each feature quantity is on the same scale, improve the model training effect, and use mean filling or linear difference method to ensure data integrity.

[0139] S42: Constructing a random forest model: as shown in Figure 4 , a plurality of sub-samples are generated from the classification data set by bootstrap sampling, a decision tree is trained for each sub-sample, the risk of overfitting is reduced by integrating multiple trees, and the final category is determined by majority voting method;

[0140] For the risk classification problem of whether a shock occurs, the category is determined by voting, the random forest integrates multiple trees to reduce the risk of overfitting, improve the classification accuracy and stability, and evaluate the feature importance. The random forest realizes a strong classifier by constructing multiple weak classifiers and combining the results, improves the classification performance, generates multiple sub-samples by bootstrap sampling method, and trains multiple decision trees.

[0141] The random forest model is constructed, including:

[0142] The input is multi-parameter data including ring count, energy release rate, duration, frequency band energy distribution, fractal feature, shock time, initial motion amplitude, spatial concentration factor, spatial distribution intensity coefficient, error spatial shape, shear and tensile properties, rupture scale, rupture surface occurrence, seismic moment magnitude, corner frequency, b value, and extension direction. The number of trees is determined to be 64 by grid search method or Bayesian optimization; the maximum depth is 7 layers; and the output is the shock probability of the current time node.

[0143] S43: Model training and evaluation: use cross-entropy loss function to optimize the model, calculate the classification effect of each tree during training; evaluate the performance of the model through evaluation indicators, and analyze the classification performance of the model through the confusion matrix; wherein the evaluation indicators include: accuracy, recall rate and comprehensive accuracy;

[0144] The definitions of accuracy, recall rate and F1-score are as follows:

[0145] Suppose TP (True Positive): true as a shock and the model recognition result is a shock, that is, the shock phenomenon is correctly identified; FP (False Positive): true as not a shock but the model recognition result is a shock, that is, the non-shock phenomenon is incorrectly identified; TN (True Negative): true as not a shock and the model recognition result is not a shock, that is, the non-shock phenomenon is correctly identified; FN (False Negative): true as a shock but the model recognition result is not a shock, that is, the shock phenomenon is incorrectly identified.

[0146] The three evaluation criteria are accuracy A, recall rate R and comprehensive accuracy F1, and the specific calculation formula is as follows:

[0147]

[0148] S44: Use the trained random forest model to classify and predict future earthquake data, output the probability of occurrence of each time point, and predict the possibility of future earthquake occurrence to provide reference for earthquake warning.

[0149] In an example of the present application, the accuracy A, recall rate R and comprehensive accuracy F1, and the specific calculation formula is as follows:

[0150]

[0151] In the formula, TP is the assumed true positive: true as a shock and the model recognition result is a shock, that is, the shock phenomenon is correctly identified; FP is the false positive: true as not a shock but the model recognition result is a shock, that is, the non-shock phenomenon is incorrectly identified; TN is the true negative: true as not a shock and the model recognition result is not a shock, that is, the non-shock phenomenon is correctly identified; FN is the false negative: true as a shock but the model recognition result is not a shock, that is, the shock phenomenon is incorrectly identified.

[0152] In an example of the present application, in the step S50, a multi-layer perception MLP and a Transformer fusion model are used to train and evaluate the training set data, and the specific classification steps are as follows:

[0153] S51: Constructing a dataset: integrating the output data of the TFT regression model and the random forest classification model, i.e., the predicted spatiotemporal strong multidimensional data and the earthquake probability, into a new dataset; the integrated dataset is in table form, each row containing the multi-parameter data predicted by regression and the earthquake probability predicted by classification. The integrated dataset is divided into a training set and a test set; the division ratio is 70% for training and 30% for testing to ensure the fairness of model training and evaluation.

[0154] S52: Building an MLP and Transformer fusion model: training the MLP model and the Transformer model respectively, and automatically selecting the MLP or Transformer model according to the task characteristics of regression or classification, determining the best combination of hyperparameters, and further optimizing the hyperparameters, calculating the error of the training set and the validation set, judging whether the model is overfitting or underfitting, and evaluating the performance of the model;

[0155] Among them, the MLP and Transformer model preparation includes:

[0156] The MLP configuration is as follows: the input is the multi-parameter data predicted by regression and the earthquake probability predicted by classification, the number of hidden layer neurons is two layers, each layer has 128 neurons, the activation function is ReLU, and the output is the final earthquake probability.

[0157] The Transformer configuration is as follows: the input is the multi-parameter data predicted by regression and the earthquake probability predicted by classification, the number of attention heads is 8, the number of hidden layer neurons is 256 per layer, the number of layers is 4, and the output is the final earthquake probability.

[0158] S53: Outputting rock fracturing earthquake multi-parameter fusion early warning results: weighting the predicted spatiotemporal strong multidimensional data and the classification predicted earthquake probability output by the MLP and Transformer model, and finally generating rock fracturing earthquake related waveform difference features, source occurrence time, location, and intensity of microseismic monitoring multi-dimensional parameter fusion early warning results.

[0159] In one example of the present application, building an MLP and Transformer fusion model includes the following steps:

[0160] S521: Model training: when training the MLP, use the mean square error MSE as the loss function, use the Adam optimizer for parameter optimization, and update the model parameters through the small batch stochastic gradient descent algorithm; when training the Transformer, use the cross-entropy loss function, use the Adam optimizer for optimization, and capture the complex relationships between features through the multi-head attention mechanism;

[0161] S522: Model Selection and Hyperparameter Tuning: Automatically select MLP or Transformer models based on the characteristics of the regression or classification task, determine the optimal hyperparameter combination through cross-validation, and further tune the hyperparameters to optimize the model performance.

[0162] S523: Model Evaluation: During training, the errors on the training and validation sets are calculated using MSE and cross-entropy to determine whether the model is overfitting or underfitting. For regression tasks, the root mean square error (RMSE) is used to evaluate model performance; for classification tasks, accuracy, precision, and recall are used. The root mean square error is defined as follows:

[0163]

[0164] Where n is the number of data points, y i It is the i-th actual value. It is the i-th predicted value.

[0165] According to a second aspect of the present invention, an engineering rock mass fracturing induced seismic multi-information fusion early warning system includes:

[0166] The acquisition module is configured to deploy microseismic sensors according to actual conditions, acquire raw waveform data induced by rock fracturing and seismic processes, and perform wavelet transform filtering on the acquired raw waveform data to obtain effective waveform data.

[0167] The processing module is configured to extract feature coefficients and invert the source mechanism of the effective waveform data to obtain spatiotemporal intensity multidimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and to use the original waveform data and spatiotemporal intensity multidimensional early warning index data induced by fracturing as the dataset.

[0168] The regression model module is configured to input the dataset into the TFT regression model for regression analysis, and to perform preprocessing by normalization and filling missing values, train the model and evaluate its performance, and obtain the predicted spatiotemporal strong multidimensional data to provide data for subsequent early warning of future seismic activity trends.

[0169] The classification model module is configured to extract feature variables from a spatiotemporal strong multidimensional early warning index dataset and obtain target variables regarding whether an earthquake will occur through the analysis of historical microseismic data. The feature variables and target variables regarding whether an earthquake will occur extracted from the dataset are preprocessed and then input into a random forest classification model for classification analysis. A multi-decision tree ensemble method is used for hazard classification to solve data noise and nonlinearity problems. The preprocessed data is trained and its performance is evaluated through the random forest classification model to obtain the predicted earthquake probability, providing data for subsequent early warning of the possibility of future earthquakes.

[0170] The weighted fusion module is configured to input the predicted spatiotemporal strong multidimensional data and the hazard classification-derived earthquake probability as input data, reorganize to establish a new data set, and divide into a training set and a test set. The training set data is trained and evaluated using a multi-layer perception MLP and a Transformer fusion model. The outputs of the MLP and the Transformer are weighted and fused using the test set data to obtain a rock fracturing earthquake multi-parameter fusion early warning result.

[0171] The system obtains and processes the induced waveform data of hot dry rock fracturing-induced earthquakes, uses wavelet transform filtering technology to obtain effective waveform data, and further extracts characteristic coefficients and source mechanism parameters. These multi-parameter data are input into the TFT regression model for regression analysis, combined with the random forest classification model for classification analysis, and through the multi-parameter prediction and earthquake probability analysis of the fusion model, the possibility of future earthquakes can be more accurately evaluated. The system effectively solves the data noise and non-linear problems in traditional methods, significantly improves the overall performance of the early warning system, and provides a scientific reference for the early warning of hot dry rock fracturing-induced earthquakes.

[0172] The system comprehensively utilizes wavelet transform filtering, characteristic coefficient extraction, and source mechanism parameter inversion, and performs multi-dimensional analysis on the hot dry rock fracturing-induced earthquake waveform data. Through the joint use of regression models and classification models, not only can the future magnitude and energy release rate and other multi-parameter data be predicted, but also the probability of future earthquakes can be evaluated. The fusion model further integrates the regression prediction and classification analysis results, realizing comprehensive early warning of hot dry rock fracturing-induced earthquakes. The system has significant advantages in data processing, model training, and prediction evaluation, and can effectively improve the accuracy and stability of earthquake prediction, providing important technical support for earthquake safety monitoring during hot dry rock mining.

[0173] The fusion model of the system utilizes the advantages of MLP and Transformer to capture the non-linear characteristics and time-dependent relationships in the data, improving the accuracy and stability of the prediction. Through the post-fusion model, the prediction results of the regression and classification models are comprehensively utilized to improve the overall performance of the early warning system. MLP and Transformer models are suitable for different data characteristics and task requirements, providing more flexible choices. The post-fusion model can reduce the bias and error of a single model, improving the stability and accuracy of the prediction. Ultimately, through the fusion of regression prediction and hazard classification results, multi-parameter fusion early warning of hot dry rock fracturing-induced earthquakes is realized.

[0174] The exemplary embodiments of the engineering rock mass fracturing earthquake-inducing multi-information fusion early warning method and system are described in detail above with reference to the preferred embodiments, however, those skilled in the art can understand that various modifications and changes can be made to the above specific embodiments without departing from the concept of the present application, and various technical features and structures of the present application can be combined without departing from the protection scope of the present application, and the protection scope of the present application is determined by the appended claims.

Claims

1. A method for early warning of rock mass fracturing and seismic induction through multi-information fusion, characterized in that, Includes the following steps: S10: Collect raw waveform data induced by rock fracturing and seismic process, and filter the acquired raw waveform data to obtain effective waveform data; S20: Extract feature coefficients and invert source mechanism of the effective waveform data to obtain spatiotemporal strong multidimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and use the original waveform data and spatiotemporal strong multidimensional early warning index data induced by fracturing as the dataset. S30: Input the dataset into the TFT regression model for regression analysis, and perform preprocessing by normalization and filling missing values. Train the model and evaluate its performance to obtain the predicted spatiotemporal strong multidimensional data. S40: Extract feature variables from the spatiotemporal strong multidimensional early warning index dataset, and obtain target variables regarding whether an earthquake will occur through the analysis of historical microseismic data; preprocess the feature variables and target variables regarding whether an earthquake will occur extracted from the dataset, and then input them into a random forest classification model for classification analysis; use the multi-decision tree ensemble method for hazard classification; and obtain the predicted earthquake probability by training and evaluating the preprocessed data through the random forest classification model. S50: The predicted spatiotemporal strong multidimensional data and the seismic probability obtained from hazard classification are used as input data to re-integrate and establish a new dataset, which is divided into training set and test set. The training set data is trained and evaluated using the multilayer perception MLP and Transformer fusion model. The outputs of the multilayer perception MLP and Transformer are weighted and fused using the test set data to obtain the multi-parameter fusion early warning result of rock fracturing seismicity.

2. The method for early warning of rock mass fracturing and seismic induction through multi-information fusion according to claim 1, characterized in that, In step S10, the acquired raw waveform data is filtered to obtain valid waveform data, specifically including the following steps: S11: Decompose the original waveform data to determine the wavelet basis function and the number of decomposition levels; S12: Perform thresholding on the decomposed signal to remove noise components; S13: Reconstruct the noise-removed signal to obtain the filtered effective waveform data.

3. The method for early warning of rock mass fracturing and seismic induction through multi-information fusion according to claim 1, characterized in that, In step S20, feature coefficients are extracted and focal mechanism inversion is performed on the effective waveform data to obtain spatiotemporal strong multi-dimensional early warning index data of waveform feature parameters, focal location, and focal mechanism parameters, including: S21: Induced waveform feature recognition, extract waveform feature coefficients, feature coefficients include ring count, energy release rate, frequency band energy distribution, fractal characteristics, vibration time and initial amplitude; S22: Inversion of focal mechanism parameters, which include rupture location index and rupture intensity index. The rupture location index includes spatial concentration factor, spatial distribution intensity coefficient, error spatial morphology, tension-shear properties, rupture surface attitude, and extension azimuth. The rupture intensity index includes earthquake moment magnitude, b-value, stress drop, and corner frequency. S23: Establish a dataset based on the waveform characteristic coefficient data and the source mechanism parameter data.

4. The method for early warning of rock mass fracturing and seismic induction through multi-information fusion according to claim 1, characterized in that, Step S30 specifically includes the following steps: S31: Construct a regression dataset by combining the waveform characteristic coefficients and source mechanism parameters in the effective waveform data to form a regression dataset; S32: Construct a TFT regression model and input the regression dataset into the TFT regression model for regression prediction analysis; S33: Performance evaluation and result prediction. The mean absolute scaling error (MASE) is used to quantitatively evaluate the model's prediction results. The trained model outputs multi-parameter prediction data and prediction intervals to obtain future seismic activity trends.

5. The multi-information fusion early warning method for rock mass fracturing and seismic induction as described in claim 4, characterized in that, The TFT regression model includes: The data processing and feature selection module is configured to use a variable selection network (VSN) to select the most salient features from the input and to encode the context vector of the static covariates through a static covariate encoder (SCE). The information flow modeling module is configured to use a gated residual network (GRN) to ensure effective information flow, a time self-attention layer (TSL) to learn the long-term dependencies of time-series data, and a position-guided feedforward layer (PFL) to perform additional nonlinear processing on the output of the self-attention layer. The decoding and prediction module is configured to enhance temporal features by utilizing the static enrichment layer SEL in the temporal fusion decoder TFD. The training optimization module is configured to use the mean squared error (MSE) as the loss function during training, and the Adam optimizer to optimize the parameters, iteratively updating the model parameters through the mini-batch stochastic gradient descent algorithm.

6. The method for early warning of rock mass fracturing and seismic induction through multi-information fusion according to claim 1, characterized in that, In step S40, the feature variables extracted from the dataset and the target variable regarding whether an earthquake occurred are preprocessed and then input into a random forest classification model for classification analysis. Specifically, this includes the following steps: S41: Dataset partitioning and data preprocessing: Extract feature variables from the spatiotemporal strong multidimensional early warning indicator dataset and combine them with the target variable regarding whether an earthquake occurred to form a classification dataset; S42: Construct a random forest model: Generate multiple subsamples from the classification dataset using bootstrap sampling, train a decision tree for each subsample, reduce the risk of overfitting by ensembling multiple trees, and finally determine the final category by majority voting. S43: Model Training and Evaluation: Use the cross-entropy loss function to optimize the model, and calculate the classification performance of each tree during training; The model's performance is evaluated using assessment metrics, and its classification performance is analyzed using a confusion matrix. The assessment metrics include: accuracy, recall, and overall correctness. S44: Use a trained random forest model to classify and predict future earthquake data, and output the probability of an earthquake occurring at each time point.

7. The method for early warning of rock mass fracturing and seismic induction through multi-information fusion according to claim 6, characterized in that, The specific formulas for calculating the precision A, recall R, and overall precision F1 are as follows: In the formula, TP is a true positive hypothesis: the actual earthquake is caused by the earthquake and the model identification result is also earthquake-causing, that is, the earthquake-causing phenomenon is correctly identified; FP is a false positive hypothesis: the actual earthquake is not caused by the earthquake, but the model identification result is earthquake-causing, that is, the non-earthquake phenomenon is incorrectly identified; TN is a true negative hypothesis: the actual earthquake is not caused by the earthquake and the model identification result is also earthquake-causing, that is, the non-earthquake phenomenon is correctly identified; FN is a false negative hypothesis: the actual earthquake is caused by the earthquake, but the model identification result is not earthquake-causing, that is, the earthquake-causing phenomenon is incorrectly identified.

8. The method for multi-information fusion and early warning of rock mass fracturing and seismic induction in engineering as described in claim 1, characterized in that, In step S50, the training set data is trained and evaluated using a fusion model of multilayer perceptual MLP and Transformer. The specific classification steps are as follows: S51: Constructing a dataset: Integrate the output data of the TFT regression model and the random forest classification model, i.e. the predicted spatiotemporal strong multidimensional data, with the seismic probability into a new dataset; S52: Construct a fusion model of MLP and Transformer: Train the MLP model and the Transformer model respectively, and automatically select the MLP or Transformer model according to the characteristics of the regression or classification task, determine the best hyperparameter combination, further tune the hyperparameters, calculate the error between the training set and the validation set, determine whether the model is overfitting or underfitting, and evaluate the model performance. S53: Output rock fracturing-induced seismic multi-parameter fusion early warning results: The predicted spatiotemporal strong multi-dimensional data and the seismic probability predicted by classification are weighted and fused to finally generate microseismic monitoring multi-parameter fusion early warning results related to waveform differences, source occurrence time, location, and intensity of rock fracturing-induced seismicity.

9. The multi-information fusion early warning method for rock mass fracturing and seismic induction as described in claim 8, characterized in that, Building a fusion model of MLP and Transformer involves the following steps: S521: Model Training: When training the MLP, the mean squared error (MSE) is used as the loss function, the Adam optimizer is used to optimize the parameters, and the model parameters are iteratively updated through the mini-batch stochastic gradient descent algorithm. When training the Transformer, the cross-entropy loss function is used, the Adam optimizer is used for optimization, and the complex relationships between features are captured through the multi-head attention mechanism; S522: Model Selection and Hyperparameter Tuning: Automatically select MLP or Transformer models based on the characteristics of the regression or classification task, determine the optimal hyperparameter combination through cross-validation, and further tune the hyperparameters to optimize the model performance. S523: Model Evaluation: During training, the error between the training set and the validation set is calculated using MSE and cross-entropy to determine whether the model is overfitting or underfitting. For regression tasks, the root mean square error (RMSE) is used to evaluate model performance, while for classification tasks, accuracy, precision, and recall are used for evaluation.

10. A multi-information fusion early warning system for rock mass fracturing and seismic induction, characterized in that, The acquisition module is configured to acquire raw waveform data induced by rock fracturing and seismic processes, and to filter the acquired raw waveform data to obtain effective waveform data. The processing module is configured to extract feature coefficients and invert the source mechanism of the effective waveform data to obtain spatiotemporal intensity multidimensional early warning index data of waveform feature parameters, source location and source mechanism parameters, and to use the original waveform data and spatiotemporal intensity multidimensional early warning index data induced by fracturing as the dataset. The regression model module is configured to input the dataset into the TFT regression model for regression analysis, and to perform preprocessing by normalization and filling missing values, train the model and evaluate its performance to obtain predicted spatiotemporal strong multidimensional data. The classification model module is configured to extract feature variables from the spatiotemporal strong multidimensional early warning index dataset and obtain target variables regarding whether an earthquake will occur through the analysis of historical microseismic data. The feature variables and target variables regarding whether an earthquake will occur extracted from the dataset are preprocessed and then input into a random forest classification model for classification analysis. The multi-decision tree ensemble method is used for hazard classification. The predicted earthquake probability is obtained by training and evaluating the preprocessed data through the random forest classification model. The weighted fusion module is configured to take the predicted spatiotemporal strong multidimensional data and the seismic probability obtained from hazard classification as input data, reintegrate them to establish a new dataset, and divide it into training and test sets. The training set data is trained and evaluated using a multilayer perceptron MLP and Transformer fusion model. The outputs of the multilayer perceptron MLP and Transformer are then weighted and fused using the test set data to obtain the multi-parameter fusion early warning result for rock fracturing seismicity.

Images