A method, device, equipment and medium for predicting the maximum magnitude of an artificial fracturing-induced earthquake

Abstract

This application discloses a method, apparatus, equipment, and medium for predicting the maximum magnitude of artificially fractured induced earthquakes, relating to the field of computer technology. It includes: acquiring historical earthquake data, historical geological data, and historical engineering data from a target location area to obtain target data; preprocessing the target data to obtain processed target data; training a deep learning model based on a multilayer perceptron and long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model; and inputting real-time geological data and real-time engineering data from the current target location area into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the model. This enables accurate prediction of the maximum magnitude of artificially fractured induced earthquakes.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, equipment and medium for predicting the maximum magnitude of an artificially fractured earthquake. Background Technology

[0002] Human activities (such as geothermal energy development, oil and gas extraction, reservoir impoundment, and wastewater injection) often lead to artificial fracturing-induced earthquakes. When the magnitude is large, it can cause certain losses and damage to human life. Therefore, studying the prediction of the maximum magnitude of artificially fracturing-induced earthquakes is of great significance for reducing the impact of disasters on human life.

[0003] However, predicting artificially induced earthquakes is particularly complex, involving various geological and engineering parameters. Traditional earthquake prediction methods mainly rely on seismological theory, statistical analysis, and empirical models. For example, the Shapiro model uses statistical analysis to predict the probability of earthquake occurrence; the McGarr model also uses statistical methods to predict the maximum magnitude of induced earthquakes; and the Turuntaev model uses physical and empirical models to estimate the maximum magnitude of induced earthquakes for the Basel project. These methods are effective for certain regions or situations, but they reveal their limitations when dealing with large-scale data and the nonlinear characteristics of seismic processes. Because the occurrence mechanism of artificially induced earthquakes is different from that of natural earthquakes, traditional methods face greater challenges in processing and interpreting earthquakes induced by human activities.

[0004] As can be seen from the above, how to accurately predict the maximum magnitude of earthquakes induced by artificial fracturing is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, equipment, and medium for predicting the maximum magnitude of artificially fractured induced earthquakes, capable of accurately predicting the maximum magnitude of such earthquakes. The specific solution is as follows:

[0006] Firstly, this application provides a method for predicting the maximum magnitude of artificially fractured earthquakes, including:

[0007] Historical seismic data, historical geological data, and historical engineering data of the target location area are collected to obtain target data. The target location area is an area where artificial fracturing-induced earthquakes have occurred. The historical seismic data includes the historical maximum magnitude. The historical geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, geostress changes, and fracture concentration. The historical engineering data includes fluid injection rate, fluid injection pressure, fluid injection depth, and total fluid injection volume.

[0008] The target data is preprocessed to obtain processed target data;

[0009] The processed target data is used to train a deep learning model based on multilayer perceptron and long short-term memory network to obtain a trained maximum magnitude prediction model.

[0010] Once real-time geological data and real-time engineering data for the target location area are obtained, the real-time geological data and real-time engineering data are input into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model.

[0011] Optionally, the preprocessing of the target data to obtain processed target data includes:

[0012] The target data is cleaned and normalized to obtain standardized target data;

[0013] Feature extraction is performed on the standardized target data, and data augmentation is performed on the feature-extracted data to obtain the processed target data.

[0014] Optionally, the deep learning model based on multilayer perceptron and long short-term memory network includes an input layer, a target data processing branch, a feature fusion layer, a fully connected layer and an output layer connected in sequence.

[0015] The target data processing branch includes a first data processing branch constructed based on a multilayer perceptron and a second data processing branch constructed based on a long short-term memory network. The input ends of the first data processing branch and the second data processing branch are both connected to the output end of the input layer, and the output ends of the first data processing branch and the second data processing branch are both connected to the input end of the feature fusion layer.

[0016] Accordingly, when real-time geological data and real-time engineering data of the current target location area are obtained, the real-time geological data and the real-time engineering data are input into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model, including:

[0017] The real-time geological data and the real-time engineering data are input into the target data processing branch through the input layer to obtain the feature data of the real-time geological data and the real-time engineering data;

[0018] The feature data is fused through the feature fusion layer to obtain comprehensive feature data;

[0019] The fully connected layer is used to process the comprehensive feature data to obtain the feature processing result, and the feature processing result is used as the maximum magnitude prediction result. Then, the maximum magnitude prediction result is output through the output layer.

[0020] Optionally, the first data processing branch based on a multilayer perceptron includes three fully connected layers connected in sequence, and the second data processing branch based on a long short-term memory network includes an embedding layer, a first long short-term memory network, and a second long short-term memory network connected in sequence.

[0021] The fully connected layer includes a number of neurons, and all fully connected layers use the ReLU activation function; the first long short-term memory network and the second long short-term memory network contain the same number of data units;

[0022] Accordingly, the step of training a deep learning model based on a multilayer perceptron and a long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model includes:

[0023] The input layer inputs the historical geological data in the processed target data into the first data processing branch in the target data processing branch to obtain the feature data of the historical geological data.

[0024] The input layer takes historical earthquake data and historical engineering data from the processed target data and inputs them into the second data processing branch. The first long short-term memory network in the second data processing branch is used to obtain the target short-term dependency relationship between the historical earthquake data and the historical engineering data. The second long short-term memory network in the second data processing branch is used to obtain the target long-term dependency relationship between the historical earthquake data and the historical engineering data. Based on the sequence information corresponding to the target short-term dependency relationship and the target long-term dependency relationship, the corresponding time series features are generated to complete the training of the deep learning model and obtain the trained maximum magnitude prediction model.

[0025] Optionally, the step of training a deep learning model based on a multilayer perceptron and a long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model includes:

[0026] The training parameters of the deep learning model based on multilayer perceptron and long short-term memory network are set, and the processed target data is input into the deep learning model based on multilayer perceptron and long short-term memory network for iterative training; the historical earthquake data also includes the frequency and time of historical earthquakes; the training parameters include the number of data samples in each iteration, the number of iterations, and the update step size of the training parameters after each iteration.

[0027] The training parameters are updated during iterative training, and the performance of the current deep learning model is evaluated on the validation set after each iteration of training.

[0028] The current deep learning model whose performance meets the preset verification and evaluation conditions is used as the maximum magnitude prediction model after training.

[0029] Optionally, after training the deep learning model based on multilayer perceptron and long short-term memory network using the processed target data to obtain the trained maximum magnitude prediction model, the method further includes:

[0030] The test set is input into the trained maximum magnitude prediction model to obtain prediction results. The trained maximum magnitude prediction model is evaluated by comparing the prediction results with the historical maximum magnitudes in the test set and by calculating evaluation indicators. The test set consists of historical data samples that have not participated in model training.

[0031] The target optimization method is determined based on the evaluation results of the trained maximum magnitude prediction model, and the trained maximum magnitude prediction model is optimized using the target optimization method; the target optimization method is any one or a combination of several of the hyperparameter optimization method, model structure optimization method, and training strategy optimization method.

[0032] Optionally, after obtaining real-time geological data and real-time engineering data for the current target location area, and inputting the real-time geological data and real-time engineering data into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model, the method further includes:

[0033] The maximum magnitude prediction model is evaluated based on the predicted maximum magnitude and the actual maximum magnitude, and then optimized based on the evaluation results of the maximum magnitude prediction model.

[0034] Secondly, this application provides a device for predicting the maximum magnitude of artificially fractured earthquakes, comprising:

[0035] The target data acquisition module is used to collect historical seismic data, historical geological data, and historical engineering data from the target location area to obtain target data. The target location area is an area where artificial fracturing-induced earthquakes have occurred. The historical seismic data includes the historical maximum magnitude. The historical geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, geostress changes, and fracture aggregation degree. The historical engineering data includes fluid injection rate, fluid injection pressure, fluid injection depth, and total fluid injection volume.

[0036] The target data preprocessing module is used to preprocess the target data to obtain processed target data;

[0037] The maximum magnitude prediction model training module is used to train a deep learning model based on multilayer perceptron and long short-term memory network using the processed target data to obtain the trained maximum magnitude prediction model.

[0038] The maximum magnitude prediction result output module is used to input the real-time geological data and real-time engineering data of the target location area into the maximum magnitude prediction model when the real-time geological data and real-time engineering data are obtained, so as to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model.

[0039] Thirdly, this application provides an electronic device, comprising:

[0040] Memory, used to store computer programs;

[0041] A processor is used to execute the computer program to implement the aforementioned method for predicting the maximum magnitude of artificially fractured earthquakes.

[0042] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned method for predicting the maximum magnitude of artificially fractured earthquakes.

[0043] This application provides a method for predicting the maximum magnitude of artificially fractured earthquakes. The method involves collecting seismic data, geological data, and engineering data to obtain target data. The seismic data includes the maximum magnitude and earthquake frequency. The geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, stress variations, and fracture aggregation. The engineering data includes fluid injection rate, fluid injection pressure, fluid injection depth, and total fluid injection volume. The target data is preprocessed to obtain processed target data. A pre-trained deep learning model based on a combination of multilayer perceptrons and long short-term memory networks is then used to further process the processed target data to obtain the maximum magnitude prediction result.

[0044] As can be seen from the above, this application utilizes a pre-trained deep learning model based on a combination of multilayer perceptrons and long short-term memory networks to process seismic data, geological data, and engineering data to obtain the maximum magnitude prediction result. That is, by leveraging the advantages of deep learning methods in processing large-scale data and in efficient prediction, the geological and engineering parameters of the study area are comprehensively analyzed, thereby accurately predicting the maximum magnitude of artificially fractured earthquakes. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of a method for predicting the maximum magnitude of an artificially fractured earthquake disclosed in this application;

[0047] Figure 2 This application discloses a flowchart of a specific method for predicting the maximum magnitude of an artificially fractured earthquake.

[0048] Figure 3 This is a schematic diagram of a deep learning model structure based on a multilayer perceptron and a long short-term memory network disclosed in this application;

[0049] Figure 4 This is a schematic diagram of a device for predicting the maximum magnitude of an artificially fractured earthquake disclosed in this application;

[0050] Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Predicting artificially fractured earthquakes is particularly complex, involving various geological and engineering parameters. Traditional earthquake prediction methods mainly rely on seismological theory, statistical analysis, and empirical models. For example, the Shapiro model uses statistical analysis to predict the probability of earthquake occurrence; the McGarr model also uses statistical methods to predict the maximum magnitude of induced earthquakes; and the Turuntaev model uses physical and empirical models to estimate the maximum magnitude of induced earthquakes for the Basel project. These methods are effective for certain regions or situations, but they reveal their limitations when dealing with large-scale data and the nonlinear characteristics of seismic processes. Therefore, this application provides a scheme for predicting the maximum magnitude of artificially fractured earthquakes. Leveraging the advantages of deep learning methods in handling large-scale data and efficient prediction, this scheme comprehensively analyzes the geological and engineering parameters of the study area, thereby accurately predicting the maximum magnitude of artificially fractured earthquakes.

[0053] See Figure 1 As shown in the embodiment of this application, a method for predicting the maximum magnitude of an artificially fractured earthquake is disclosed, including:

[0054] Step S11: Collect historical earthquake data, historical geological data, and historical engineering data in the target location area to obtain the target data.

[0055] In this embodiment, a target location area is determined, and historical seismic data, historical geological data, and historical engineering data are collected from this area to obtain target data. The target location area is a region where artificial fracturing-induced earthquakes have occurred. The historical seismic data includes, but is not limited to, the historical maximum magnitude, historical earthquake frequency, and historical earthquake occurrence time. The historical geological data includes, but is not limited to, lithological characteristics, fault length and properties, fault distribution characteristics, geostress changes, and fracture aggregation. The historical engineering data includes, but is not limited to, fluid injection rate, fluid injection pressure, fluid injection depth, and total fluid injection volume. It is understood that the specific data types of the aforementioned historical seismic data, historical geological data, and historical engineering data can be determined based on actual application conditions and are not specifically limited here.

[0056] Step S12: Preprocess the target data to obtain processed target data.

[0057] In this embodiment, the target data is preprocessed to obtain processed target data, and the target data is divided into training, validation, and testing groups according to a preset ratio for subsequent training of a deep learning model based on multilayer perceptrons and long short-term memory networks. Specifically, the preprocessing of the target data to obtain processed target data may include: cleaning and normalizing the target data to obtain standardized target data; extracting features from the standardized target data; and performing data augmentation on the feature-extracted data to obtain processed target data.

[0058] In one specific implementation, the data cleaning of the target data may include: deleting incomplete data or processing it using methods such as mean, median, interpolation, forward or backward padding; and processing outliers through visual detection and statistical detection. Further, the normalization of the target data may include: calculating the mean and standard deviation of the target data, and converting the original data into standardized data based on the mean and standard deviation. Furthermore, feature extraction is performed on the standardized target data, and data augmentation is applied to the feature-extracted data to obtain processed target data. This may include: extracting the frequency, maximum magnitude, and time of earthquakes within a certain time period from seismic data; quantifying different lithological characteristics in geological data, directly using fault length, and representing fault properties, individual fault characteristics, and fracture aggregation using numerical features to calculate the characteristics of in-situ stress changes; extracting the time series features of injection rate, injection pressure, and total injection from engineering data; enhancing seismic data using methods such as time migration, noise addition, and time scaling; enhancing geological data using methods such as random perturbation, data synthesis, and feature combination; and enhancing engineering data using methods such as interpolation generation, extended time windows, and data smoothing. It is understood that the specific processing methods described above can be determined based on actual application conditions and are not specifically limited here.

[0059] Step S13: Use the processed target data to train the deep learning model based on multilayer perceptron and long short-term memory network to obtain the trained maximum magnitude prediction model.

[0060] See Figure 2As shown, in this embodiment, a deep learning model combining a multilayer perceptron (MLP) and a long short-term memory (LSTM) network is used to analyze and train seismic data, geological data, and engineering data. The deep learning model based on the MLP and LSTM network includes an input layer, a target data processing branch, a feature fusion layer, a fully connected layer, and an output layer connected in sequence. The target data processing branch includes a first data processing branch built on the MLP and a second data processing branch built on the LSTM network. The inputs of both the first and second data processing branches are connected to the output of the input layer, and the outputs of both the first and second data processing branches are connected to the input of the feature fusion layer.

[0061] Furthermore, the first data processing branch based on a multilayer perceptron includes three fully connected layers connected in sequence, and the second data processing branch based on a long short-term memory network includes an embedding layer, a first long short-term memory network, and a second long short-term memory network connected in sequence. Each fully connected layer includes a number of neurons, and all fully connected layers use the ReLU activation function. The first and second long short-term memory networks contain the same number of data units. It should be noted that the number of neurons in each fully connected layer decreases sequentially according to the connection order of the fully connected layers; the specific number can be determined based on the actual application scenario and is not specifically limited here.

[0062] Accordingly, training the deep learning model based on a multilayer perceptron and a long short-term memory network using the processed target data to obtain the trained maximum magnitude prediction model may include: inputting historical geological data from the processed target data into a first data processing branch of the target data processing branch through the input layer to obtain feature data of the historical geological data; inputting historical earthquake data and historical engineering data from the processed target data into a second data processing branch through the input layer to obtain the target short-term dependency relationship of the historical earthquake data and the historical engineering data through the first long short-term memory network in the second data processing branch, and obtaining the target long-term dependency relationship of the historical earthquake data and the historical engineering data through the second long short-term memory network in the second data processing branch; and generating corresponding time series features based on the sequence information corresponding to the target short-term dependency relationship and the target long-term dependency relationship to complete the training of the deep learning model and obtain the trained maximum magnitude prediction model. Specifically, for historical geological data, the MLP method is used, divided into three fully connected layers, all using the ReLU activation function. Fully connected layer 1 contains *i* neurons, aiming to initially extract features from the geological data; fully connected layer 2 contains *j* neurons, further extracting higher-level features; and fully connected layer 3 contains *k* neurons, refining the features from the previous layer to generate the final geological data features. For historical earthquake and engineering data, the LSTM method is used, mainly consisting of two LSTM layers, each containing *p* units, with Dropout 0.2 used for regularization to prevent overfitting. LSTM layer 1 captures short-term dependencies in the time series and preserves sequence information, while LSTM layer 2 captures long-term dependencies in the time series, generating the final time series features.

[0063] In this embodiment, the training parameters need to be set first. The learning rate controls the parameter update step size; the batch size controls the number of data samples used to train the model in each iteration; and the number of training epochs determines the number of times the training set is traversed. Next, the final prediction model is obtained through forward propagation, loss calculation, backpropagation, and parameter update training. Simultaneously, at the end of each training iteration, the model performance is evaluated on the validation set, and the model that performs best on the validation set is saved. Specifically, training a deep learning model based on a multilayer perceptron and long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model may include: setting training parameters for the deep learning model based on the multilayer perceptron and long short-term memory network, and inputting the processed target data into the deep learning model for iterative training; the training parameters include the number of data samples in each iteration, the number of iterations, and the update step size of the training parameters after each iteration; updating the training parameters during iterative training, and evaluating the performance of the current deep learning model on a validation set after each iteration; and using the current deep learning model whose performance meets the preset validation evaluation conditions as the trained maximum magnitude prediction model. If the performance on the validation set no longer improves within several iterations, training is stopped early, and the current deep learning model is used as the trained maximum magnitude prediction model. That is, the model is trained using historical earthquake data and related parameters to learn complex patterns and correlations in the earthquake data. The deep learning model is trained using a pre-defined training set. Set appropriate learning rates, batch sizes, and training iterations to improve the model's convergence and generalization capabilities.

[0064] Furthermore, a test set is prepared, ensuring that the test set data has not been used for model training, preserving its purity to evaluate the model's generalization ability; the test set is input into the model to obtain prediction results, which are then compared with the true values, and the model performance is quantified using various evaluation metrics. Specifically, after training the deep learning model based on multilayer perceptrons and long short-term memory networks using the processed target data to obtain the trained maximum magnitude prediction model, the process may further include: inputting the test set into the trained maximum magnitude prediction model to obtain prediction results, and evaluating the trained maximum magnitude prediction model by comparing the prediction results with the historical maximum magnitudes in the test set and calculating evaluation metrics; the test set consists of historical data samples that have not participated in model training. The quantification of model performance using various evaluation metrics mainly includes mean-square error (MSE), mean absolute error (MAD), and R² coefficient of determination, where the formula for the mean-square error is as follows:

[0065] ;

[0066] The formula for the mean absolute error is as follows:

[0067] ;

[0068] The formula for the R² coefficient of determination is as follows:

[0069] ;

[0070] in, It is a predicted value. It is the actual value. It is the average of the true values. That is the number of samples.

[0071] Furthermore, after training the deep learning model based on multilayer perceptron and long short-term memory network using the processed target data to obtain the trained maximum magnitude prediction model, the process may further include: determining a target optimization method based on the evaluation results of the trained maximum magnitude prediction model, and optimizing the trained maximum magnitude prediction model using the target optimization method; the target optimization method is any one or a combination of several of the following: hyperparameter optimization method, model structure optimization method, and training strategy optimization method. Hyperparameter optimization is further divided into random search, grid search, and Bayesian optimization. Random search can be used for initial optimization, grid search can be used for fine-tuning, and Bayesian optimization is suitable for advanced optimization. These three methods can be combined according to specific circumstances. Random search involves defining the hyperparameter space and its range of values, then randomly selecting several hyperparameter combinations, training the model, evaluating its performance, and finally selecting the best-performing hyperparameter combination on the validation set. Grid search involves first defining the hyperparameters to be tuned and their possible ranges of values, traversing all combinations, training the model, evaluating its performance, and finally selecting the best-performing hyperparameter combination on the validation set. Bayesian optimization involves initially setting several hyperparameter combinations, training the model, evaluating its performance, then using the evaluation results to update the probabilistic model. Next, it selects the next set of hyperparameter combinations based on the probabilistic model, and finally repeats the first two steps until a stopping condition is met. Model structure tuning mainly includes adding regularization, adjusting the number of layers, and adjusting the number of neurons. These three methods can be combined according to the actual situation. Adding regularization is suitable for initial adjustments, while adjusting the number of layers and adding neurons is used when initial performance is unsatisfactory. Adding regularization mainly prevents overfitting by incorporating techniques such as Dropout and L2; adjusting the number of layers mainly adjusts the model depth; adjusting the number of neurons adjusts the model's learning ability. The training strategy optimization mainly includes learning rate scheduling and batch size adjustment. First, the batch size is determined, and then learning rate scheduling is introduced. Learning rate scheduling dynamically adjusts the learning rate during training, primarily using two methods: learning rate decay and cosine annealing. Batch size adjustment involves trying different batch sizes and finding the one that provides stable training and fast convergence.

[0072] Step S14: When real-time geological data and real-time engineering data of the current target location area are obtained, the real-time geological data and real-time engineering data are input into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model.

[0073] In this embodiment, real-time data is fed into a trained deep learning model, and the model's forward propagation calculations are used to perform real-time analysis and predict the maximum magnitude of induced earthquake events. Specifically, when real-time geological data and real-time engineering data for the current target location area are obtained, the real-time geological data and real-time engineering data are input into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model. This can include: inputting the real-time geological data and real-time engineering data into the target data processing branch through the input layer to obtain feature data of the real-time geological data and real-time engineering data; fusing the feature data through the feature fusion layer to obtain comprehensive feature data; processing the comprehensive feature data using the fully connected layer to obtain feature processing results, and using the feature processing results as the maximum magnitude prediction result; and then outputting the maximum magnitude prediction result through the output layer.

[0074] It is important to note that after making the actual prediction, the prediction results need to be compared and evaluated with the actual observation data to continuously optimize the maximum magnitude prediction model. Specifically, after obtaining real-time geological data and real-time engineering data for the current target location area, and inputting these data into the maximum magnitude prediction model to obtain the maximum magnitude prediction result output by the model, the process may further include: evaluating the maximum magnitude prediction model based on the maximum magnitude prediction result and the actual maximum magnitude, and optimizing the model based on the evaluation results. That is, comparing and evaluating the prediction results with the actual observation data; calculating indicators such as prediction error and accuracy to evaluate the model's predictive performance; further optimizing the parameters and structure of the deep learning model based on the feedback from the evaluation results; continuously monitoring model performance; and constantly adjusting and optimizing the model based on new earthquake event data. Prediction accuracy can be improved by increasing training data and adjusting model hyperparameters.

[0075] As can be seen from the above, this embodiment utilizes a pre-trained deep learning model combining multilayer perceptrons and long short-term memory networks to process seismic, geological, and engineering data to obtain the maximum magnitude prediction result. The deep learning model can learn from a large amount of artificial fracturing-induced earthquake event data, improving the accuracy of predicting the maximum magnitude of artificial fracturing-induced earthquakes. Secondly, the deep learning model has strong adaptability and generalization ability, capable of handling complex seismic data patterns and variations, and is effective for artificial fracturing operations under different geological conditions. Finally, the method based on big data and deep learning models provides scientific basis and credibility, helping to improve the understanding and management of earthquake-induced risks. In other words, by leveraging the advantages of deep learning methods in processing large-scale data and efficient prediction, the geological and engineering parameters of the study area are comprehensively analyzed, thereby accurately predicting the maximum magnitude of artificial fracturing-induced earthquakes.

[0076] See Figure 3 As shown, this application provides a specific method for predicting the maximum magnitude of artificially fractured induced earthquakes, including:

[0077] In this embodiment, the maximum magnitude prediction of artificially fractured earthquakes is completed through a process of data collection, data preprocessing, deep learning model design and training, model evaluation and optimization, and implementation.

[0078] Specifically, the data collection mainly focuses on three aspects: earthquake, geology, and engineering data. Earthquake data mainly includes maximum magnitude and earthquake frequency; geological data mainly includes lithological characteristics, fault length and properties, fault distribution characteristics, stress variations, and fracture aggregation; engineering data mainly includes fluid injection rate, pressure, depth, and total fluid injection volume.

[0079] Furthermore, the data preprocessing first involves data cleaning to remove duplicate data. Incomplete data can be deleted or processed using methods such as mean, median, interpolation, forward or backward imputation, etc. Outliers are detected visually, statistically, and then processed. Next, normalization is performed. This normalization primarily employs Z-Score standardization, which can handle data features with different means and standard deviations, regardless of whether it's seismic, geological, or engineering data. This reduces the impact of outliers on building the deep learning model, improving the stability of the deep learning model training process while preserving the data distribution. Then, feature extraction is performed. In seismic data, this mainly involves extracting the frequency of earthquakes, maximum magnitude, and time of occurrence within a specific time period. In geological data, it mainly quantifies different lithological characteristics, directly using fault length, and using numerical features to represent fault properties, individual fault characteristics, fracture aggregation, and calculating the characteristics of in-situ stress changes. In engineering data, it extracts the time series features of injection rate, injection pressure, and total injection, such as selecting the average or maximum value. Finally, when data is limited, data augmentation can be performed to facilitate input into the deep learning model. For seismic data, enhancement methods mainly include time migration, noise addition, and time scaling; for geological data, enhancement methods mainly include random perturbation, data synthesis, and feature combination; and for engineering data, enhancement methods mainly include interpolation generation, extended time windows, and data smoothing.

[0080] Furthermore, the design and training of the deep learning model begins with the selection of the model architecture. A deep learning model combining a Multilayer Perceptron (MLP) and a Long Short-Term Memory (LSTM) network is considered for analysis and training on seismic, geological, and engineering data. Next, the model is trained, starting with setting training parameters. This involves setting the learning rate to control the parameter update step size; setting the batch size to control the number of data samples used to train the model in each iteration; and setting the number of training epochs to traverse the training set completely. The training process mainly includes forward propagation, loss calculation, backpropagation, and parameter updates, through which the final prediction value is obtained. Simultaneously, validation is required. At the end of each training iteration, the model performance is evaluated on the validation set, and the best-performing model on the validation set is saved. It is important to note that during training, if the validation set performance no longer improves within a certain number of epochs, training is stopped early to prevent overfitting.

[0081] Furthermore, the model evaluation mainly includes preparing a test set, comparing predictions with true values, and calculating evaluation metrics. Preparing the test set ensures that the test set data has not been used for model training, preserving its purity to evaluate the model's generalization ability; comparing predictions with true values ​​involves inputting the test set data into the model, obtaining the prediction results, and comparing them with the true values; calculating evaluation metrics mainly involves quantifying model performance using various evaluation indicators. The model tuning mainly includes hyperparameter tuning, model structure tuning, training strategy tuning, and early stopping.

[0082] Furthermore, the practical application involves feeding real-time data into a pre-trained deep learning model and using forward propagation calculations to perform real-time analysis and predict the maximum magnitude of induced seismic events. Continuous optimization involves comparing and evaluating the prediction results with actual observation data. Indicators such as prediction error and accuracy are calculated to assess the model's predictive performance. Based on feedback from the evaluation results, the parameters and structure of the deep learning model are further optimized. Model performance is continuously monitored, and the model is constantly adjusted and optimized based on new seismic event data. Prediction accuracy can be improved by increasing training data and adjusting model hyperparameters.

[0083] As can be seen from the above, this embodiment utilizes a pre-trained deep learning model combining multilayer perceptrons and long short-term memory networks to process seismic, geological, and engineering data to obtain the maximum magnitude prediction result. The deep learning model can learn from a large amount of artificial fracturing-induced earthquake event data, improving the accuracy of predicting the maximum magnitude of artificial fracturing-induced earthquakes. Secondly, the deep learning model has strong adaptability and generalization ability, capable of handling complex seismic data patterns and variations, and is effective for artificial fracturing operations under different geological conditions. Finally, the method based on big data and deep learning models provides scientific basis and credibility, helping to improve the understanding and management of earthquake-induced risks. In other words, by leveraging the advantages of deep learning methods in processing large-scale data and efficient prediction, the geological and engineering parameters of the study area are comprehensively analyzed, thereby accurately predicting the maximum magnitude of artificial fracturing-induced earthquakes.

[0084] Accordingly, see Figure 4 As shown in the figure, this application discloses a device for predicting the maximum magnitude of an artificially fractured earthquake, comprising:

[0085] The target data acquisition module 11 is used to acquire historical seismic data, historical geological data, and historical engineering data in the target location area to obtain target data. The target location area is an area where artificial fracturing-induced earthquakes have occurred. The historical seismic data includes the historical maximum magnitude. The historical geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, geostress changes, and fracture aggregation degree. The historical engineering data includes fluid injection rate, fluid injection pressure, fluid injection depth, and total fluid injection volume.

[0086] Target data preprocessing module 12 is used to preprocess the target data to obtain processed target data;

[0087] The maximum magnitude prediction model training module 13 is used to train a deep learning model based on multilayer perceptron and long short-term memory network using the processed target data to obtain the trained maximum magnitude prediction model.

[0088] The maximum magnitude prediction result output module 14 is used to input the real-time geological data and real-time engineering data of the target location area into the maximum magnitude prediction model when the real-time geological data and real-time engineering data are obtained, so as to obtain the maximum magnitude prediction result output by the maximum magnitude prediction model.

[0089] As can be seen from the above, the embodiments of this application utilize a pre-trained deep learning model based on a combination of multilayer perceptrons and long short-term memory networks to process earthquake data, geological data, and engineering data to obtain the maximum magnitude prediction result. That is, by leveraging the advantages of deep learning methods in processing large-scale data and in efficient prediction, the geological and engineering parameters of the study area are comprehensively analyzed, thereby accurately predicting the maximum magnitude of artificially fractured earthquakes.

[0090] In some specific embodiments, the target data preprocessing module 12 may specifically include:

[0091] The target data standardization unit is used to perform data cleaning and normalization on the target data to obtain standardized target data.

[0092] The feature extraction unit is used to extract features from the standardized target data and perform data augmentation on the feature-extracted data to obtain processed target data.

[0093] In some specific embodiments, the maximum magnitude prediction model training module 13 may specifically include:

[0094] The training parameter setting unit is used to set the training parameters of the deep learning model based on multilayer perceptron and long short-term memory network, and input the processed target data into the deep learning model based on multilayer perceptron and long short-term memory network for iterative training; the historical earthquake data also includes the frequency and time of historical earthquakes; the training parameters include the number of data samples in each iteration, the number of iterations, and the update step size of the training parameters after each iteration.

[0095] The deep learning model performance evaluation unit is used to update the training parameters during iterative training and evaluate the performance of the current deep learning model on the validation set after each iteration of training.

[0096] The maximum magnitude prediction model determination unit is used to select the current deep learning model whose performance meets the preset verification and evaluation conditions as the trained maximum magnitude prediction model.

[0097] In some specific implementations, the deep learning model based on multilayer perceptron and long short-term memory network includes an input layer, a target data processing branch, a feature fusion layer, a fully connected layer, and an output layer connected in sequence.

[0098] The target data processing branch includes a first data processing branch constructed based on a multilayer perceptron and a second data processing branch constructed based on a long short-term memory network. The input ends of the first data processing branch and the second data processing branch are both connected to the output end of the input layer, and the output ends of the first data processing branch and the second data processing branch are both connected to the input end of the feature fusion layer.

[0099] Correspondingly, the maximum magnitude prediction result output module 14 may specifically include:

[0100] The feature data generation unit is used to input the real-time geological data and the real-time engineering data into the target data processing branch through the input layer to obtain feature data of the real-time geological data and the engineering data;

[0101] A feature data fusion unit is used to fuse the feature data through the feature fusion layer to obtain comprehensive feature data;

[0102] The maximum magnitude prediction result output unit is used to process the comprehensive feature data using the fully connected layer to obtain the feature processing result, and use the feature processing result as the maximum magnitude prediction result, and then output the maximum magnitude prediction result through the output layer.

[0103] In some specific implementations, the first data processing branch based on a multilayer perceptron includes three fully connected layers connected in sequence, and the second data processing branch based on a long short-term memory network includes an embedding layer, a first long short-term memory network, and a second long short-term memory network connected in sequence.

[0104] The fully connected layer includes a number of neurons, and all fully connected layers use the ReLU activation function; the first long short-term memory network and the second long short-term memory network contain the same number of data units;

[0105] Accordingly, the maximum magnitude prediction model training module 13 may specifically include:

[0106] The historical geological data feature extraction unit is used to input the historical geological data in the processed target data into the first data processing branch in the target data processing branch through the input layer, so as to obtain the feature data of the historical geological data.

[0107] The historical earthquake data and historical engineering data feature extraction unit is used to input the historical earthquake data and historical engineering data from the processed target data into the second data processing branch through the input layer, so as to obtain the target short-term dependency relationship of the historical earthquake data and the historical engineering data through the first long short-term memory network in the second data processing branch, and obtain the target long-term dependency relationship of the historical earthquake data and the historical engineering data through the second long short-term memory network in the second data processing branch, and generate corresponding time series features based on the sequence information corresponding to the target short-term dependency relationship and the target long-term dependency relationship, so as to complete the training of the deep learning model and obtain the trained maximum magnitude prediction model.

[0108] In some specific embodiments, the artificial fracturing-induced earthquake maximum magnitude prediction device may further include:

[0109] The maximum magnitude prediction model testing module is used to input the test set into the trained maximum magnitude prediction model to obtain the prediction result, and to evaluate the trained maximum magnitude prediction model by comparing the prediction result with the historical maximum magnitude in the test set and calculating the evaluation index; the test set is a historical data sample that has not participated in the model training.

[0110] The first maximum magnitude prediction model optimization module is used to determine the target optimization method based on the evaluation results of the trained maximum magnitude prediction model, and to optimize the trained maximum magnitude prediction model using the target optimization method; the target optimization method is any one or a combination of several of the hyperparameter optimization method, model structure optimization method, and training strategy optimization method.

[0111] The second maximum magnitude prediction model optimization module is used to evaluate the maximum magnitude prediction model based on the maximum magnitude prediction results and the actual maximum magnitude, and to optimize the maximum magnitude prediction model based on the evaluation results of the maximum magnitude prediction model.

[0112] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the artificial fracturing-induced earthquake maximum magnitude prediction method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0113] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0114] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0115] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the artificial fracturing-induced earthquake maximum magnitude prediction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0116] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for predicting the maximum magnitude of artificially fractured earthquakes. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0118] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0119] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0120] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0121] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the maximum magnitude of an induced seismic event resulting from artificial fracturing, characterized in that, The method comprises the following steps: data acquisition is performed on historical earthquake data, historical geological data and historical engineering data in a target location area to obtain target data; the target location area is an area where man-made fracturing induced earthquakes occur; the historical earthquake data includes a historical maximum magnitude; the historical geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, ground stress changes and fracture accumulation degree; the historical engineering data includes fluid injection rate, fluid injection pressure, fluid injection depth and total fluid injection volume; preprocessing is performed on the target data to obtain processed target data; a deep learning model based on a multi-layer perceptron and a long short-term memory network is trained using the processed target data to obtain a trained maximum magnitude prediction model; when real-time geological data and real-time engineering data in the target location area are obtained, the real-time geological data and the real-time engineering data are input into the maximum magnitude prediction model to obtain a maximum magnitude prediction result output by the maximum magnitude prediction model.

2. The method of claim 1, wherein, The preprocessing of the target data to obtain processed target data comprises: data cleaning and normalization processing are performed on the target data to obtain standardized target data; feature extraction is performed on the standardized target data, and data enhancement is performed on the data after feature extraction to obtain processed target data.

3. The method of claim 1, wherein, The deep learning model based on a multi-layer perceptron and a long short-term memory network comprises an input layer, a target data processing branch, a feature fusion layer, a full connection layer and an output layer connected in sequence; The target data processing branch comprises a first data processing branch based on a multi-layer perceptron and a second data processing branch based on a long short-term memory network, and the input ends of the first data processing branch and the second data processing branch are connected with the output end of the input layer, and the output ends of the first data processing branch and the second data processing branch are connected with the input end of the feature fusion layer. Correspondingly, when the real-time geological data and the real-time engineering data in the target location area are obtained, the real-time geological data and the real-time engineering data are input into the maximum magnitude prediction model to obtain a maximum magnitude prediction result output by the maximum magnitude prediction model, which comprises: the real-time geological data and the real-time engineering data are input into the target data processing branch through the input layer to obtain feature data of the real-time geological data and the real-time engineering data; the feature data is fused through the feature fusion layer to obtain comprehensive feature data; the comprehensive feature data is processed using the full connection layer to obtain a feature processing result, and the feature processing result is taken as the maximum magnitude prediction result, and then the maximum magnitude prediction result is output through the output layer.

4. The method of claim 3, wherein, The first data processing branch based on a multi-layer perceptron comprises three full connection layers connected in sequence, and the second data processing branch based on a long short-term memory network comprises an embedding layer, a first long short-term memory network and a second long short-term memory network connected in sequence. The full connection layer includes a plurality of neurons, and the full connection layer uses a ReLU activation function; the first long short-term memory network and the second long short-term memory network include the same number of data units; Correspondingly, the training of the deep learning model based on the multi-layer perceptron and the long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model comprises: The historical geological data in the processed target data is input into the first data processing branch in the target data processing branch through the input layer to obtain feature data of the historical geological data; The historical earthquake data and the historical engineering data in the processed target data are input into the second data processing branch through the input layer, the target short-term dependency relationship of the historical earthquake data and the historical engineering data is obtained through the first long short-term memory network in the second data processing branch, the target long-term dependency relationship of the historical earthquake data and the historical engineering data is obtained through the second long short-term memory network in the second data processing branch, and corresponding time sequence features are generated based on the sequence information corresponding to the target short-term dependency relationship and the target long-term dependency relationship to complete the training of the deep learning model to obtain a trained maximum magnitude prediction model.

5. The method of claim 1, wherein, The training of the deep learning model based on the multi-layer perceptron and the long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model comprises: The training parameters of the deep learning model based on the multi-layer perceptron and the long short-term memory network are set, and the processed target data is input into the deep learning model based on the multi-layer perceptron and the long short-term memory network for iterative training; the historical earthquake data further includes historical earthquake occurrence frequency and historical earthquake occurrence time; the training parameters include the number of data samples in each iteration, the number of iterations, and the update step of the training parameters after each iteration; The training parameters are updated during the iterative training process, and the performance of the current deep learning model is evaluated on a validation set after each iteration; The current deep learning model that meets the preset validation evaluation condition is used as the trained maximum magnitude prediction model.

6. The method of claim 1, wherein, After the training of the deep learning model based on the multi-layer perceptron and the long short-term memory network using the processed target data to obtain a trained maximum magnitude prediction model, the following steps are further included: A test set is input into the trained maximum magnitude prediction model to obtain a prediction result, and the trained maximum magnitude prediction model is evaluated by comparing the prediction result with historical maximum magnitudes in the test set and calculating evaluation indexes; the test set is a historical data sample that has not participated in model training; Determine a target tuning method according to the evaluation result of the trained maximum magnitude prediction model, and optimize the trained maximum magnitude prediction model through the target tuning method; the target tuning method is any one or a combination of several of the hyperparameter tuning method, the model structure tuning method, and the training strategy tuning method.

7. The method of predicting the maximum magnitude of induced seismicity from hydraulic fracturing according to any one of claims 1 to 6, wherein, After the real-time geological data and the real-time engineering data on the target location area are acquired, the real-time geological data and the real-time engineering data are input into the maximum magnitude prediction model to obtain a maximum magnitude prediction result output by the maximum magnitude prediction model, and the method further includes: Evaluate the maximum magnitude prediction model based on the maximum magnitude prediction result and an actual maximum magnitude, and optimize the maximum magnitude prediction model based on the evaluation result of the maximum magnitude prediction model.

8. A device for predicting the maximum magnitude of an artificially fractured earthquake, characterized in that, The method includes: A target data acquisition module is configured to acquire historical earthquake data, historical geological data, and historical engineering data on a target location area to obtain target data; the target location area is an area where man-made fracturing-induced earthquakes exist, the historical earthquake data includes historical maximum magnitudes, the historical geological data includes lithological characteristics, fault length and properties, fault distribution characteristics, ground stress changes, and fracture accumulation degrees, and the historical engineering data includes fluid injection rates, fluid injection pressures, fluid injection depths, and fluid injection total volumes. A target data preprocessing module is configured to preprocess the target data to obtain processed target data. A maximum magnitude prediction model training module is configured to train a deep learning model based on a multilayer perceptron and a long short-term memory network by using the processed target data to obtain a trained maximum magnitude prediction model. A maximum magnitude prediction result output module is configured to, when real-time geological data and real-time engineering data on the target location area are acquired, input the real-time geological data and the real-time engineering data into the maximum magnitude prediction model to obtain a maximum magnitude prediction result output by the maximum magnitude prediction model.

9. An electronic device, comprising: The method includes: A memory is configured to save a computer program. A processor is configured to execute the computer program to implement the man-made fracturing-induced earthquake maximum magnitude prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is saved, and the computer program is executed by a processor to implement the man-made fracturing-induced earthquake maximum magnitude prediction method according to any one of claims 1 to 7.

Images