Fracture activity prediction method and system based on numerical simulation
Through the fracture activity prediction method based on numerical simulation, combined with the finite element method and Bayesian network model, the problems of lack of data and insufficient information in the prior art are solved, and efficient and accurate fracture activity prediction and uncertainty analysis are achieved.
Patent Information
- Application Number
- CN202411554371.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-04
- Publication Date
- 2025-02-25
AI Technical Summary
The prior art has problems such as lack of data, insufficient information and poor data quality in monitoring and prediction of fault activity, and it is difficult to fully capture the dynamic characteristics of fault activity.
Using a fault activity prediction method based on numerical simulation, a variety of geological data were collected and pre-processed, the stress distribution and deformation of the fault zone was simulated by using the finite element method, a Bayesian network model was constructed and its structure and parameters were optimized through Q-Learning, and the prediction results of multiple models were combined using Bagging, and the real-time monitoring data was assimilated into the numerical simulation model.
An accurate fault activity prediction model has been established, which improves the accuracy and reliability of the prediction, can dynamically reflect the latest geological conditions, provide detailed prediction results and their uncertainty analysis, and supports efficient monitoring and prevention of fault activities.
Smart Images
Figure CN119511346A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of fault activity prediction, and in particular, relates to a method and system for predicting fault activity based on numerical simulation. Background Art
[0002] Conventional technologies have many deficiencies in the monitoring and prediction of fault activity. First, conventional technologies often rely on a single data source or a limited data set, which makes the model lack data and insufficient information when facing a complex and changing geological environment, making it difficult to fully capture the dynamic characteristics of fault activity. Second, data preprocessing and analysis methods are relatively simple, and fail to fully remove noise or fill in missing data, resulting in poor quality of the basic data of the model, affecting the accuracy and reliability of the prediction results.
[0003] In view of this, the present invention is proposed. Summary of the invention
[0004] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method and system for predicting fault activity based on numerical simulation, thereby solving the problems raised in the above-mentioned background technology.
[0005] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:
[0006] A method for predicting fault activity based on numerical simulation includes the following steps:
[0007] (1): Collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform preprocessing;
[0008] (2): Use the finite element method to simulate the stress distribution and deformation of the fault zone;
[0009] (3): Use the acquired data to build a prediction model, use Q-Learning to optimize the structure and parameters of the Bayesian network, and use Bagging to combine the prediction results of multiple models;
[0010] (4): Assimilate the real-time monitoring data into the numerical simulation model and iteratively update the simulation model;
[0011] (5): Use the optimized Bayesian network model to predict the probability of fault activity and visualize the prediction results. Finally, perform uncertainty analysis based on the prediction results.
[0012] Optionally, the steps for preprocessing the collected data are:
[0013] A1: Use the formula X clean =median(X i-k , ..., Xi , ..., X i+k ) removes noise and fills missing values from the acquired data, where X clean is the cleaned data, X is the original data, and k is the filter window size;
[0014] A2: Use the formula The cleaned data is standardized, where X norm is the standardized data, μ is the mean of the data, and σ is the standard deviation of the data
[0015] A3: Extract features from the standardized data and reduce the data dimension.
[0016] Optionally, the steps of simulating the stress distribution and deformation of the fault zone using the finite element method are as follows:
[0017] B1: Based on the collected geological data, a geometric model of the fault zone is established and discretized, a finite element mesh is generated, and boundary conditions and initial conditions are applied to the model.
[0018] B2: Use the formula K·u=F to solve the stress distribution and deformation of the fracture zone, where K is the global stiffness matrix, u is the node displacement vector, and F is the external load vector;
[0019] B3: According to the distribution of stress and deformation, the stress concentration area is adaptively refined to complete the simulation of the simulated fracture zone.
[0020] Optionally, the steps to build a prediction model using the acquired data are:
[0021] C1: Use convolutional neural networks to extract features from image data and use long short-term memory networks (LSTM) to process time series data;
[0022] C2: Construct a Bayesian network model to describe the conditional probability relationship of fault activity, and its expression is: Among them, P(X i |Parents(X i )) is node X i Conditional probability conditional on its parent node;
[0023] C3: Use variational autoencoders for variational inference to estimate the parameters of the Bayesian network;
[0024] C4: Construct a spatiotemporal Bayesian network, combine spatial and temporal data to capture the spatiotemporal dependencies of fault activities and use variational autoencoders for variational inference, estimate the parameters of the Bayesian network, extract features and make predictions through convolutional neural networks and long short-term memory networks, and generate a prediction model based on the prediction results.
[0025] Optionally, the steps for optimizing the structure and parameters of the Bayesian network using Q-Learning are:
[0026] D1: Define the initial structure and parameters of the Bayesian network, initialize the Q table to generate Q values, use the greedy strategy to select actions, execute the selected actions in the current structural state, and obtain a new structural state and corresponding rewards;
[0027] D2: Iterate the Q value until the Q value reaches the maximum number of iterations to obtain the optimized Bayesian network structure and parameters.
[0028] Optionally, use bagging to combine the predictions of multiple models:
[0029] E1: Generate B training subsets from the original data set by sampling with replacement, and train a Bayesian network model for each subset to form a model set f1, f2, ..., f B ;
[0030] E2: For new input data x, the prediction results of multiple Bayesian network models are combined through the Bagging method, and its expression is: in, The prediction results of the combined model, B is the number of Bayesian network models trained, f b (x) is the prediction result of the b-th Bayesian network model for the input data x;
[0031] E3: The prediction results of all models are averaged and integrated to obtain the final prediction result.
[0032] Optionally, the steps to assimilate real-time monitoring data into numerical simulation are:
[0033] F1: Set the initial state vector C0 and the numerical simulation model M to predict the state at the next moment. The expression is: in, is the predicted state vector, w k is the model noise;
[0034] F2: Obtain real-time monitoring data and use the Kalman filter to assimilate the real-time monitoring data and numerical simulation data;
[0035] F3: Iterate steps F1-F2 at each time step, continuously update the state vector and error covariance matrix, and assimilate new monitoring data in real time.
[0036] Optionally, the steps of using the optimized Bayesian network model to predict the probability of fault activity are:
[0037] J1: Use Q-Learning to optimize the structure and parameters of the Bayesian network, perform probabilistic reasoning through the Bayesian network, calculate the probability distribution of fault activity, and use the belief propagation algorithm to perform probabilistic reasoning in the Bayesian network to obtain the inference prediction results;
[0038] J2: Use visualization tools to display inference prediction results.
[0039] Optionally, the steps for uncertainty analysis are:
[0040] S1: Use the formula H(X) = -∑ x∈χ P(x)logP(x) calculates the entropy of the prediction result, measures the uncertainty, and uses the formula Calculate the variance of the prediction results to measure the degree of dispersion of the prediction value, where H(X) is the entropy of the random variable $X$, χ is all possible states of X, P(x) is the probability that X is in state x, Var(X) is the variance of X, is the expectation of X, μ is the expected value of X;
[0041] S2: Evaluate the impact of different evidence variables on the prediction results, determine which variables contribute the most to the prediction of fault activity, and finally evaluate the importance of the evidence variables.
[0042] A fault activity prediction system based on numerical simulation, comprising:
[0043] Data acquisition module, used to collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform pre-processing;
[0044] Numerical simulation module, used to simulate the stress distribution and deformation of the fault zone using the finite element method;
[0045] The model building module is used to build a prediction model using the acquired data, optimize the structure and parameters of the Bayesian network using Q-Learning, and combine the prediction results of multiple models using Bagging;
[0046] Data assimilation module, used to assimilate real-time monitoring data into the numerical simulation model and iteratively update the simulation model;
[0047] The data display and analysis module is used to use the optimized Bayesian network model to make probability predictions of fault activities and visualize the prediction results. Finally, uncertainty analysis is performed based on the prediction results.
[0048] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:
[0049] By integrating multiple data sources and performing complex preprocessing and analysis, combined with advanced numerical simulation and optimization techniques, an accurate prediction model is established. The prediction capability of the model is further improved through innovative Bayesian network optimization and integration methods. At the same time, real-time monitoring data is dynamically integrated and continuously updated to reflect the latest conditions. Ultimately, detailed prediction results and their uncertainty analysis are provided, thereby providing efficient and accurate decision-making support for the monitoring and prevention of fault activities.
[0050] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0052] In the figure:
[0053] Figure 1 Flowchart of the fault activity prediction method.
[0054] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0055] The present invention will now be described in further detail with reference to the accompanying drawings.
[0056] See also Figure 1 As shown, in this embodiment, a method for predicting fault activity based on numerical simulation is provided, comprising the following steps:
[0057] (1): Collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform preprocessing;
[0058] In this embodiment, the steps of preprocessing the collected data are:
[0059] A1: Use the formula X clean =median(X i-k , ..., X i , ..., X i+k ) removes noise and fills missing values from the acquired data, where X clean is the cleaned data, X is the original data, and k is the filter window size;
[0060] A2: Use the formula The cleaned data is standardized, where X normis the standardized data, μ is the mean of the data, and σ is the standard deviation of the data
[0061] A3: Extract features from the standardized data and reduce the data dimension. Through sophisticated data preprocessing steps, the high quality and consistency of the data are ensured, the impact of noise and missing values is eliminated, and the comparability of the data is guaranteed through standardization. Further feature extraction and dimensionality reduction steps effectively reduce the complexity of the data, making subsequent modeling and analysis more efficient and accurate, thereby improving the reliability and performance of the overall prediction model.
[0062] (2): Use the finite element method to simulate the stress distribution and deformation of the fault zone;
[0063] In this embodiment, the steps of simulating the stress distribution and deformation of the fracture zone using the finite element method are as follows:
[0064] B1: Based on the collected geological data, a geometric model of the fault zone is established and discretized, a finite element mesh is generated, and boundary conditions and initial conditions are applied to the model.
[0065] B2: Use the formula K·u=F to solve the stress distribution and deformation of the fracture zone, where K is the global stiffness matrix, u is the node displacement vector, and F is the external load vector;
[0066] B3: According to the distribution of stress and deformation, the stress concentration area is adaptively refined to complete the simulation of the simulated fracture zone.
[0067] It should be noted that by applying boundary conditions and initial conditions, the model can truly reflect the actual geological conditions and stress state, ensuring the reliability and accuracy of the simulation. By using the formula for solving the stress distribution and deformation of the fault zone, the stress and deformation response of each node can be accurately calculated, thereby showing in detail the stress distribution and deformation characteristics of the fault zone under different conditions. This precise calculation method can effectively capture the subtle changes in the fault zone under different stress conditions and provide a deeper geomechanical analysis. Furthermore, by adaptively refining the grid in the stress concentration area, the method can dynamically adjust the resolution of the grid to improve the simulation accuracy, especially in high stress or high deformation areas, making the simulation results more accurate and detailed. This adaptive grid refinement technology not only improves the calculation efficiency and saves computing resources, but also ensures high-precision simulation in key areas.
[0068] (3): Use the acquired data to build a prediction model, use Q-Learning to optimize the structure and parameters of the Bayesian network, and use Bagging to combine the prediction results of multiple models;
[0069] In this embodiment, the step of using the acquired data to build a prediction model is as follows:
[0070] C1: Use convolutional neural networks to extract features from image data and use long short-term memory networks (LSTM) to process time series data;
[0071] C2: Construct a Bayesian network model to describe the conditional probability relationship of fault activity, and its expression is: Among them, P(X i |Parents(X i )) is node X i Conditional probability conditional on its parent node;
[0072] C3: Use variational autoencoders for variational inference to estimate the parameters of the Bayesian network;
[0073] C4: Construct a spatiotemporal Bayesian network, combine spatial and temporal data to capture the spatiotemporal dependencies of fault activities and use variational autoencoders for variational inference, estimate the parameters of the Bayesian network, extract features and make predictions through convolutional neural networks and long short-term memory networks, and generate a prediction model based on the prediction results.
[0074] By combining multiple networks and variational inference techniques, a powerful prediction model was formed, which not only improved the accuracy of the prediction of fault activity, but also enabled detailed spatiotemporal analysis, thus providing a scientific basis for geological disaster warning and prevention. This method significantly improved the comprehensiveness and accuracy of the prediction through multi-level and multi-angle analysis and modeling.
[0075] In this embodiment, the steps of using Q-Learning to optimize the structure and parameters of the Bayesian network are as follows:
[0076] D1: Define the initial structure and parameters of the Bayesian network, initialize the Q table to generate Q values, use the greedy strategy to select actions, execute the selected actions under the current structural state, and obtain a new structural state and corresponding rewards; initialize the Q table and generate Q values, which provides a metric for the optimization process and evaluates the effects of different structures and parameter combinations through changes in Q values.
[0077] D2: Iterate the Q value until the Q value reaches the maximum number of iterations to obtain the optimized Bayesian network structure and parameters. The greedy strategy is used to select actions and the selected actions are executed in the current structural state. This can balance exploration and utilization during the optimization process, and can not only discover new and potentially better structures, but also make full use of existing experience and knowledge to gradually improve the performance of the model. After executing the selected action at each step, a new structural state and corresponding rewards are obtained. The reward mechanism can evaluate the contribution of each adjustment to the model performance, thereby guiding the subsequent optimization direction.
[0078] In addition, by iteratively updating the Q value, the model can gradually learn and optimize until the Q value reaches the maximum number of iterations, ensuring the stability and convergence of the optimization process. Finally, the optimized Bayesian network structure and parameters are obtained, making the model achieve the highest prediction accuracy and robustness.
[0079] In this embodiment, the steps of using Bagging to combine the prediction results of multiple models are as follows:
[0080] E1: Generate B training subsets from the original data set by sampling with replacement, and train a Bayesian network model for each subset to form a model set f1, f2, ..., f B ;
[0081] E2: For new input data x, the prediction results of multiple Bayesian network models are combined through the Bagging method, and its expression is: in, The prediction results of the combined model, B is the number of Bayesian network models trained, f b (x) is the prediction result of the b-th Bayesian network model for the input data x;
[0082] E3: The prediction results of all models are averaged and integrated to obtain the final prediction results. First, multiple training subsets are generated by sampling with replacement to ensure the diversity and representativeness of each subset. This method can effectively alleviate the overfitting problem so that each training subset can capture the different characteristics of the original data. Secondly, by training an independent Bayesian network model for each training subset, a model set is formed. This set can cover various patterns and changes of the data, thereby improving the stability and generalization ability of the overall model. For new input data, the Bagging method is used to combine the prediction results of multiple Bayesian network models. Through the integration of multiple models, the deviations and errors that may be caused by a single model can be effectively reduced, and the reliability of the prediction results can be enhanced.
[0083] (4): Assimilate the real-time monitoring data into the numerical simulation model and iteratively update the simulation model;
[0084] In this embodiment, the steps of assimilating the real-time monitoring data into the numerical simulation are:
[0085] F1: Set the initial state vector C0 and the numerical simulation model M to predict the state at the next moment. The expression is: in, is the predicted state vector, w k is the model noise;
[0086] F2: Obtain real-time monitoring data and use the Kalman filter to assimilate the real-time monitoring data and numerical simulation data;
[0087] F3: Iterate steps F1-F2 at each time step, continuously update the state vector and error covariance matrix, and assimilate new monitoring data in real time. In this embodiment, first, the initial state vector and numerical simulation model are set to provide a clear basis and structure for subsequent predictions, and the state change at the next moment can be accurately predicted through numerical simulation. Then, real-time monitoring data is obtained and data assimilation is performed. The real-time monitoring data is combined with the numerical simulation data through the Kalman filter, which can effectively correct and update the prediction model so that the model can reflect the latest field observations. The application of the Kalman filter not only improves the adaptability of the model, but also continuously optimizes the state vector and error covariance matrix through its iterative process to ensure the high accuracy and high reliability of the model. By continuously performing data assimilation and model updating at each time step, and assimilating new monitoring data in real time, the prediction model can be dynamically adjusted and optimized to cope with the constant changes in the environment and conditions.
[0088] (5): Use the optimized Bayesian network model to predict the probability of fault activity and visualize the prediction results. Finally, perform uncertainty analysis based on the prediction results.
[0089] In this embodiment, the steps of using the optimized Bayesian network model to predict the probability of fault activity are as follows:
[0090] J1: Use Q-Learning to optimize the structure and parameters of the Bayesian network, perform probabilistic reasoning through the Bayesian network, calculate the probability distribution of fault activity, and use the belief propagation algorithm to perform probabilistic reasoning in the Bayesian network to obtain the inference prediction results;
[0091] J2: Use visualization tools to display the inference prediction results. Through precise probabilistic reasoning combined with efficient visualization, the prediction of fault activity is not only more accurate, but also provides users with intuitive decision support tools, thus significantly improving the scientificity and effectiveness of geological disaster prevention and control. Through this method, we can better cope with complex and changing geological environments, improve the practicality and reliability of prediction models, and provide strong technical support and theoretical basis for research and practice in related fields.
[0092] In this embodiment, the steps of uncertainty analysis are:
[0093] S1: Use the formula H(X) = -∑ x∈χ P(x)logP(x) calculates the entropy of the prediction result, measures the uncertainty, and uses the formula Calculate the variance of the prediction results to measure the degree of dispersion of the prediction value, where H(X) is the entropy of the random variable $X$, χ is all possible states of X, P(x) is the probability that X is in state x, Var(X) is the variance of X, is the expectation of X, μ is the expected value of X; by calculating the entropy of the prediction results to measure uncertainty, the reliability of the model prediction results can be effectively evaluated. The calculation of the entropy value provides a quantitative indicator that reflects the degree of certainty of the prediction results, allowing users to intuitively understand the credibility of the prediction results.
[0094] S2: Evaluate the impact of different evidence variables on the prediction results, determine which variables contribute most to the prediction of fault activity, and finally evaluate the importance of the evidence variables. By evaluating the impact of different evidence variables on the prediction results, it is possible to determine which variables contribute most to the prediction of fault activity. This process not only helps to optimize the prediction model to make it more accurate and effective, but also reveals the key factors that affect fault activity.
[0095] A fault activity prediction system based on numerical simulation, comprising:
[0096] Data acquisition module, used to collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform pre-processing;
[0097] Numerical simulation module, used to simulate the stress distribution and deformation of the fault zone using the finite element method;
[0098] The model building module is used to build a prediction model using the acquired data, optimize the structure and parameters of the Bayesian network using Q-Learning, and combine the prediction results of multiple models using Bagging;
[0099] Data assimilation module, used to assimilate real-time monitoring data into the numerical simulation model and iteratively update the simulation model;
[0100] The data display and analysis module is used to use the optimized Bayesian network model to make probability predictions of fault activities and visualize the prediction results. Finally, uncertainty analysis is performed based on the prediction results.
[0101] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.
Claims
1. A method for predicting fault activity based on numerical simulation, characterized in that: The following steps are involved: (1): Collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform preprocessing; (2): Use the finite element method to simulate the stress distribution and deformation of the fault zone; (3): Use the acquired data to build a prediction model, use Q-Learning to optimize the structure and parameters of the Bayesian network, and use Bagging to combine the prediction results of multiple models; (4): Assimilate the real-time monitoring data into the numerical simulation model and iteratively update the simulation model; (5): Use the optimized Bayesian network model to predict the probability of fault activity and visualize the prediction results. Finally, perform uncertainty analysis based on the prediction results.
2. A method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps for preprocessing the collected data are: A1: Use the formula X clean =median(X i-k , ..., X i , ..., X i+k ) removes noise and fills missing values from the acquired data, where X clean is the cleaned data, X is the original data, and k is the filter window size; A2: Use the formula The cleaned data is standardized, where X norm is the standardized data, μ is the mean of the data, and σ is the standard deviation of the data A3: Extract features from the standardized data and reduce the data dimension.
3. A method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps of simulating the stress distribution and deformation of the fault zone using the finite element method are as follows: B1: Based on the collected geological data, a geometric model of the fault zone is established and discretized, a finite element mesh is generated, and boundary conditions and initial conditions are applied to the model. B2: Use the formula K·u=F to solve the stress distribution and deformation of the fracture zone, where K is the global stiffness matrix, u is the node displacement vector, and F is the external load vector; B3: According to the distribution of stress and deformation, the stress concentration area is adaptively refined to complete the simulation of the simulated fracture zone.
4. A method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps to build a predictive model using the acquired data are: C1: Use convolutional neural networks to extract features from image data and use long short-term memory networks (LSTM) to process time series data; C2: Construct a Bayesian network model to describe the conditional probability relationship of fault activity, and its expression is: Among them, P(X i |Parents(X i )) is node X i Conditional probability conditional on its parent node; C3: Use variational autoencoders for variational inference to estimate the parameters of the Bayesian network; C4: Construct a spatiotemporal Bayesian network, combine spatial and temporal data to capture the spatiotemporal dependencies of fault activities and use variational autoencoders for variational inference, estimate the parameters of the Bayesian network, extract features and make predictions through convolutional neural networks and long short-term memory networks, and generate a prediction model based on the prediction results.
5. The method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps to use Q-Learning to optimize the structure and parameters of the Bayesian network are: D1: Define the initial structure and parameters of the Bayesian network, initialize the Q table to generate Q values, use the greedy strategy to select actions, execute the selected actions in the current structural state, and obtain a new structural state and corresponding rewards; D2: Iterate the Q value until the Q value reaches the maximum number of iterations to obtain the optimized Bayesian network structure and parameters.
6. A method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps to use Bagging to combine the prediction results of multiple models are: E1: Generate B training subsets from the original data set by sampling with replacement, and train a Bayesian network model for each subset to form a model set f1, f2, ..., f B ; E2: For new input data x, the prediction results of multiple Bayesian network models are combined through the Bagging method, and its expression is: in, The prediction results of the combined model, B is the number of Bayesian network models trained, f b (x) is the prediction result of the b-th Bayesian network model for the input data x; E3: The prediction results of all models are averaged and integrated to obtain the final prediction result.
7. The method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps to assimilate real-time monitoring data into numerical simulation are: F1: Set the initial state vector C0 and the numerical simulation model M to predict the state at the next moment. The expression is: in, is the predicted state vector, w k is the model noise; F2: Obtain real-time monitoring data and use the Kalman filter to assimilate the real-time monitoring data and numerical simulation data; F3: Iterate steps F1-F2 at each time step, continuously update the state vector and error covariance matrix, and assimilate new monitoring data in real time.
8. The method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps for predicting the probability of fault activity using the optimized Bayesian network model are as follows: J1: Use Q-Learning to optimize the structure and parameters of the Bayesian network, perform probabilistic reasoning through the Bayesian network, calculate the probability distribution of fault activity, and use the belief propagation algorithm to perform probabilistic reasoning in the Bayesian network to obtain the inference prediction results; J2: Use visualization tools to display inference prediction results.
9. The method for predicting fault activity based on numerical simulation according to claim 1, characterized in that: The steps of uncertainty analysis are: S1: Use the formula Calculate the entropy of the prediction results, measure uncertainty, and use the formula Calculate the variance of the prediction results to measure the degree of dispersion of the prediction values, where H(X) is the entropy of the random variable $X$. are all possible states of X, P(x) is the probability of X being in state x, Var(X) is the variance of X, is the expectation of X, μ is the expected value of X; S2: Evaluate the impact of different evidence variables on the prediction results, determine which variables contribute the most to the prediction of fault activity, and finally evaluate the importance of the evidence variables.
10. A fault activity prediction system based on numerical simulation, characterized in that: include: Data acquisition module, used to collect geological data, seismic wave velocity, gravity anomaly, magnetic anomaly and image data and perform pre-processing; Numerical simulation module, used to simulate the stress distribution and deformation of the fault zone using the finite element method; The model building module is used to build a prediction model using the acquired data, optimize the structure and parameters of the Bayesian network using Q-Learning, and combine the prediction results of multiple models using Bagging; Data assimilation module, used to assimilate real-time monitoring data into the numerical simulation model and iteratively update the simulation model; The data display and analysis module is used to use the optimized Bayesian network model to make probability predictions of fault activities and visualize the prediction results. Finally, uncertainty analysis is performed based on the prediction results.
Citation Information
Cited By
Deposited reservoir prediction method and system based on three-dimensional seismic data
CN120335006A