Ancient and modern structure superposed oil and gas filling prediction method

Abstract

The invention discloses an ancient and today structure superimposed oil and gas filling prediction method. The method comprises the following steps: step 1, outputting a today structure contour map and an uncertainty thermodynamic diagram with both spatial precision and uncertainty quantization; 2, constructing a structural transformation index to quantify the damage / storage weight of each period to the trap; 3, integrating the fission track of apatite and (U-Th) / He thermal chronology data, and outputting a paleostructure contour map with error distribution and a paleowater system probability field; 4, a sand body probability distribution three-dimensional body is generated, the digital core pore network model is coupled to simulate an oil and gas filling critical threshold value, and a sand body connectivity index graph is generated to constrain high-quality reservoir distribution; 5, superposing multiple parameters in an ancient and modern structure high point superposition region, analyzing reservoir control element contribution degrees by applying an SHAP algorithm to focus main control factors, and locking a favorable target region of oil and gas enrichment; and step 6, generating a three-dimensional favorable target area probability body and generating an exploration priority sequence in combination with an economic evaluation module.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration technology, and in particular to a method for predicting oil and gas injection based on the superposition of ancient and modern structures. Background Technology

[0002] Against the backdrop of continuously growing global energy demand and increasingly depleted shallow oil and gas resources, complex tectonic zones—geological units that have undergone multiple tectonic movements, developed fault systems, dramatic sedimentary facies changes, complex lithological combinations, and present complex surface and subsurface conditions—are gradually becoming strategic replacement areas for oil and gas exploration. Their exploration background profoundly reflects the industry's transformation trend from "finding oil in simple structures" to "tackling complex geological bodies." Traditional exploration targets, such as simple tectonic oil and gas reservoirs like anticline traps, can no longer meet energy demands. Meanwhile, complex tectonic zones, due to multiple tectonic changes, result in diverse oil and gas reservoir types, complex formation processes, and variable preservation conditions. This leads to technical bottlenecks in conventional seismic imaging technology, such as low signal-to-noise ratio, difficulty in structural identification, and poor reservoir prediction accuracy. Simultaneously, drilling engineering faces challenges such as wellbore instability, high leakage risk, and difficulty in trajectory control. This forces exploration technology to innovate towards high-precision 3D seismic processing and interpretation, intelligent velocity field modeling, multiphysics coupling simulation, and big data-driven lithofacies prediction.

[0003] Traditional exploration focuses only on present structural highs. However, structural highs during the oil and gas charging period may migrate due to later tectonic alteration, leading to problems such as present highs being oil-free and ancient highs being difficult to find, causing the exploration target to deviate. Therefore, a method for predicting oil and gas charging by superimposing ancient and modern structures is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for predicting oil and gas injection based on the superposition of ancient and modern structures.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting hydrocarbon injection based on superimposed ancient and modern geological structures includes the following steps: Step 1: Construct a three-dimensional geological model base using high-precision seismic data, well formation data, and paleomagnetic dating results. Combine random forest algorithm to collaboratively constrain detrital zircon U-Pb dating data and optimize velocity field modeling with cosmogenic nuclide erosion rate. Then, generate structural surface error distribution through Monte Carlo simulation to mark the confidence interval of the current structural map. Output the current structural contour map and uncertainty heat map with both spatial accuracy and uncertainty quantification. Step 2: The FLAC³D+TOUGHREACT coupled model is used to reconstruct the thermal-fluid-solid coupled evolution history of the basin. The paleostress field is restored by calcite e-twin paleostress gauge to quantify the intensity of fracture activity. Then, the LSTM+Attention model is used to analyze the subsidence rate curve and magmatic activity record to automatically identify the tectonic transformation period. The tectonic transformation index is constructed to quantify the destruction / preservation weight of each period on the trap. Step 3: Integrate apatite fission track and (U-Th) / He thermochronological data, invert the spatiotemporal evolution of erosion based on the TecMod model, introduce random forest to optimize the weights of sediment flux and erosion rate to generate a probabilistic erosion distribution model, and couple Delft3D paleohydrodynamic simulation to invert the source direction and channel migration path, outputting paleotectonic contour maps with error distribution and paleoriver probability fields; Step 4: Construct the Transformer-CGAN model, using sparse drilling data, seismic attribute volume and paleogeographic elevation model as input to generate a three-dimensional volume of sand body probability distribution, coupled with a digital core pore network model to simulate the critical threshold of oil and gas charging, and generate a sand body connectivity index map to constrain the distribution of high-quality reservoirs. Step 5: Overlay multiple parameters in the overlapping area of ​​ancient and modern tectonic high points, apply the SHAP algorithm to analyze the contribution of reservoir-controlling factors to focus on the main controlling factors, lock in favorable target areas for oil and gas enrichment, and at the same time train the Geo-LLM (Geological Large Language Model) to deduce the probability distribution of reservoir formation scenarios. Step Six: Develop an active learning algorithm to prioritize and suggest drilling target areas, reduce uncertainty by updating the model using Bayesian methods, generate a 3D probability volume of favorable target areas, and combine it with the economic evaluation module to generate an exploration priority ranking.

[0006] The above further includes: Furthermore, determining the current structural high points using the current structural contour map generated in step one involves the following steps: Data Preparation and Basement Model Construction: Integrating high-precision seismic data, well stratigraphic data, and paleomagnetic dating results, a three-dimensional geological model basement was constructed. A random forest algorithm was used to couple the detrital zircon U-Pb age with the cosmogenic nuclide erosion rate, establishing a velocity field model constrained by erosion rate and dating data. This quantifies the impact of stratigraphic erosion processes on the basement morphology, represented as... ,in U-Pb age, For the erosion rate, The thickness of the eroded strata. For erosion time, For rock density, Porosity; Structural Surface Interpolation and Error Simulation: Using well point elevation and seismic horizon as hard data and paleomagnetic correction dip angle as soft constraints, the initial structural surface is generated using the Kriging interpolation method, represented as follows: In the formula, For Kriging weights, For hard data, To satisfy the second-order stationarity assumption, a random error term is introduced, and the interpolation error term is randomly sampled using Monte Carlo simulation to generate the structural surface error distribution and 95% confidence interval, denoted as follows: Quantify the uncertainty of spatial prediction; The current method for determining structural high points is as follows: contour lines are drawn using the interpolation result Z(x,y), and the structural high point region is marked by combining the error heat map. The confidence interval of the high point elevation is calculated using the error simulation results.

[0007] Furthermore, in step two, the dynamic coupling simulation process of construction-accumulation is as follows: Reconstruction of the Thermal-Fluid-Solid System Coupling Evolution History: The thermal-fluid-solid system coupling evolution history of the basin was reconstructed using the FLAC³D and TOUGHREACT coupled models. By solving the heat conduction equation, Darcy's fluid flow equation, and solid deformation equation, the dynamic coupling process of thermal history, fluid pressure, and tectonic deformation was quantified. The solved heat conduction equation is expressed as follows: In the formula, T is the temperature. Thermal conductivity, The Darcy fluid flow equation is expressed as follows, representing the terms of radioactive heat generation and frictional heat generation: In the formula, For Darcy velocity, For the permeability tensor, For fluid viscosity, Given the pressure gradient, the solid deformation equation is expressed as follows: In the formula, For effective stress, Bulk density; Paleostress Field Reconstruction and Fracture Activity Quantification: The paleostress field was inverted using a calcite e-twin paleostress meter. The fracture activity intensity was quantified using a quantitative formula relating twin width and differential stress. The quantitative formula is expressed as follows: In the formula, For differential stress, G is the surface energy, and G is the shear modulus. d is Poisson's ratio, and d is the calcite grain size; Tectonic change phase identification: Deploy a deep learning framework with LSTM and Attention mechanism to automatically extract tectonic change phases from subsidence rate curves and magmatic activity records; Construction of the Tectonic Transformation Index and Evaluation of Trap Effectiveness: A tectonic transformation index is constructed, which integrates trap height loss, caprock pressure breakthrough, and multi-stage weighting to quantify the destructive / preservation effects of tectonic transformation on hydrocarbon preservation, expressed as: In the formula, For the i-th period, For the loss of trap height, The initial trap height, This represents the amount of pressure breakthrough in the caprock. To overcome the pressure of the cap layer.

[0008] Furthermore, the specific steps for identifying the structural transformation period are as follows: Data preprocessing and feature encoding: The settlement rate curves and magmatic activity records of the target area are standardized into a unified feature tensor through data preprocessing; LSTM-Attention Model Construction: This model utilizes an LSTM network to capture long-term dependencies in time series data. Its cell state update mechanism learns temporal patterns of geological events in the target region through dynamic regulation of the forget gate, input gate, and output gate. Subsequently, an attention mechanism is introduced, using an attention weight allocation formula... Quantifying the importance of each time step, in the formula, , This is the weight matrix. For bias terms; Model training and validation: The model is optimized using a weighted cross-entropy loss function, where the attention weighting coefficient enhances the recognition accuracy of key periods.

[0009] Furthermore, in step three, the process for constructing the probability field of paleotectonic high points and paleoriver systems is as follows: Thermochronological data integration and spatiotemporal evolution inversion of erosion: Integrating apatite fission tracks and (U-Th) / He age data, the spatiotemporal evolution of erosion was inverted based on the TecMod model. Age was simulated using the age-erosion relationship, which is expressed as follows: In the formula, For thermochronological age, The decay constant is This refers to the amount of erosion. The sample tilt angle, The empirical coefficients for the diameter of the etch pits and the track density are used. The erosion field D(x,y,t) is optimized by the Markov chain Monte Carlo algorithm to minimize the error between the simulated age and the measured AFT age and (U-Th) / He age. Deposition-erosion weight optimization and probabilistic erosion quantity modeling: The random forest algorithm is introduced to optimize the weight allocation of deposition flux and erosion rate, construct a probabilistic erosion quantity distribution model, generate the probability distribution of erosion quantity, and quantify the contribution of deposition flux and erosion rate to the erosion process. Paleohydrodynamic simulation and source-channel inversion: Coupled with the Delft3D paleohydrodynamic model, the paleosource direction and channel migration path are inverted by solving the shallow water equations and parameterizing the Manning coefficient and boundary flow. Paleotectonic high point determination and error quantification: By integrating erosion data and paleoriver data, paleogeographic reconstruction formulas and error propagation Monte Carlo simulations are used to generate paleotectonic contour maps with error distribution.

[0010] Furthermore, in step four, the sand body connectivity prediction and high-quality reservoir constraint process is as follows: Transformer-CGAN Sandbody Probability Volume Generation: Construct a Transformer-CGAN model, using sparse drilling data, seismic attribute volume and paleogeographic elevation model as input, and use the multi-head self-attention mechanism of the Transformer encoder to capture the long-distance spatial correlation of the input data, and combine the adversarial training strategy of conditional generative adversarial network to generate a three-dimensional volume of sandbody probability distribution. SedFlux Synthetic Data Augmentation: A SedFlux sedimentary process simulator is deployed to generate a synthetic seismic attribute-sandbody probability pair dataset by solving the diffusion equation and sedimentation rate formula. The diffusion equation is expressed as follows: In the formula, h is the water depth, D is the diffusion coefficient, Q is the source-sink term, and the deposition rate formula is expressed as: In the formula, For the inflow volume, The concentration of suspended solids. For the deposition area, Porosity; Digital core pore network model simulation: The digital core pore network model is used to quantify the critical threshold for oil and gas injection and to construct the sand body connectivity index; Sandbody connectivity index map generation: Integrating sandbody probabilistic volume and critical threshold results, the connectivity index formula is used to quantify sandbody size, permeability-weighted average, and the proportion of isolated sandbody bodies, generating a connectivity index map. The connectivity index formula is expressed as follows: In the formula, For the area of ​​the sand body grid, For penetration rate, The number of isolated sand bodies.

[0011] Furthermore, in step five, the specific steps for multi-parameter constraints and favorable area prediction in the overlapping ancient and modern tectonic zones are as follows: Analysis of the superposition of ancient and modern tectonic high points: Spatial superposition of ancient and modern tectonic high points identifies regions that possess both tectonic inheritance and preservation conditions. The formula for determining the superposition zone is expressed as follows: In the formula, The elevation of the ancient tectonic high point This represents the lower limit of the elevation of the ancient tectonic high point. To construct the current high point elevation, This represents the lower limit of the current high point's altitude. Multi-parameter synergistic constraints and comprehensive evaluation: Within the overlapping area, sand body connectivity index (CI), tectonic transformation index (TDI), and source rock maturity (Ro) are superimposed to quantify hydrocarbon accumulation conditions. The comprehensive evaluation formula is expressed as follows: In the formula, , , =0.2 is the weight. , The threshold for hydrocarbon generation from source rocks; The SHAP algorithm analyzes the contribution of hydrocarbon accumulation control factors: it quantifies the marginal contribution of each parameter to hydrocarbon accumulation prediction, focuses on the main control factors, and the SHAP value calculation formula is as follows: In the formula, z is the feature subset. This is the prediction function for the XGBoost model. The total number of features; Favorable target area identification and verification: Based on comprehensive score and main control factor analysis, favorable oil and gas enrichment areas are identified; Geological Big Language Model for Hydrocarbon Accumulation Scenario Inference: Based on geological knowledge graphs and historical hydrocarbon accumulation cases, this study infers the probability of hydrocarbon accumulation under different parameter combinations. It adopts a BERT pre-training + fine-tuning strategy, with geological parameter sequences as input and hydrocarbon accumulation probabilities as output.

[0012] Furthermore, in step six, the active learning optimization and exploration economic evaluation process is as follows: Active learning algorithms are prioritized for recommending encrypted drilling target areas: Active learning algorithms are deployed to quantify the prediction uncertainty of high-risk areas based on an uncertainty sampling strategy, using formulas... Identify the in-place drilling target area most sensitive to model optimization, where, For the prediction uncertainty of position x, This represents the predicted probability of the current model for category c). Based on existing drilling data, regions with U(x) > 0.3 are selected as priority target areas for infill drilling; Bayesian update model reduces prediction uncertainty: A Bayesian update model is used to fuse new drilling data, and Gaussian process regression is used for posterior correction of the sand body probability volume, expressed as follows: In the formula, and The updated mean and variance are obtained using the formula. Dynamically reduce prediction variance, where, For model parameters, For new drilling data, Prior data; Generation of 3D favorable target area probability volume: Integrating the updated sand body probability volume, tectonic transformation index (TDI), and source rock maturity, and using a comprehensive probability formula. Generate a three-dimensional probability field of favorable target areas to quantify spatial exploration priorities; The economic evaluation module generates an exploration priority ranking: This is coupled with the economic evaluation module and based on the net present value formula. Calculate the economic benefits of each target area, where, For oil prices, Annual oil production For operating costs, Let r be the geological success rate, and r be the discount rate. For capital expenditures, an exploration priority sequence is generated by sorting them in descending order of NPV / risk coefficient.

[0013] The present invention has the following beneficial effects: 8. In this invention, a confidence interval for the current structural map is generated through Monte Carlo simulation, and an uncertainty heat map is output to clarify the distribution range of current structural high points. Apatite fission tracks and (U-Th) / He data are integrated, and the spatiotemporal evolution of erosion is inverted based on the TecMod model. The probability field of the paleo-river system is inverted by coupling Delft3D paleohydrodynamic simulation, and the paleostructural contour map of the hydrocarbon charging period is reconstructed. The SHAP algorithm is applied to analyze the contribution of reservoir-controlling factors, with the superposition degree of structural high points as the main controlling factor. The spatial superposition area between structural high points during the hydrocarbon charging period and current structural high points is locked. Multiple parameters such as sand body connectivity index, tectonic transformation index, and source rock maturity are superimposed in the superposition area. Uncertainty is reduced through a Bayesian update model, generating a three-dimensional probability volume of favorable target areas. An exploration priority ranking is generated by combining this with an economic evaluation module. This achieves accurate spatiotemporal matching of ancient and modern structural high points, solves the problem of traditional exploration target deviation, and significantly improves the success rate of oil and gas drilling.

[0014] 9. In this invention, the thermal-fluid-solid coupled evolution history of the basin is reconstructed using the FLAC³D+TOUGHREACT coupled model. The paleostress field is restored by combining calcite e-twin paleostress gauges, and the change in fracture activity intensity over time is quantified. The subsidence rate curve and magmatic activity record are analyzed using the LSTM+Attention model to automatically identify tectonic transformation periods, construct a tectonic transformation index, quantify the destruction / preservation weight of each period on the trap, and input the tectonic transformation index as a constraint condition into the Transformer-CGAN model to generate a three-dimensional volume of sand body probability distribution. A digital core pore network model is coupled to simulate the critical threshold of hydrocarbon charging, dynamically evaluate the impact of tectonic transformation on hydrocarbon charging efficiency, and extrapolate the probability distribution of hydrocarbon accumulation scenarios using Geo-LLM to clarify the control role of multiple tectonic transformations on hydrocarbon enrichment patterns, thereby achieving spatiotemporal synchronous coupling of tectonic activity and hydrocarbon charging. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a method for predicting oil and gas injection based on the superposition of ancient and modern structures proposed in this invention. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 As shown, this invention is a method for predicting hydrocarbon injection based on the superposition of ancient and modern structures, comprising the following steps: Step 1: Construct a three-dimensional geological model base using high-precision seismic data, well formation data, and paleomagnetic dating results. Combine random forest algorithm to collaboratively constrain detrital zircon U-Pb dating data and optimize velocity field modeling with cosmogenic nuclide erosion rate. Generate structural surface error distribution through Monte Carlo simulation to mark the confidence interval of the current structural map. Output a high-precision current structural contour map and uncertainty heat map with both spatial accuracy and uncertainty quantification. The current structural high points are determined by the high-precision current structural contour map. Step 2: The FLAC³D+TOUGHREACT coupled model is used to reconstruct the basin's thermo-fluid-solid coupled evolution history. The paleostress field is restored and the intensity of fracture activity is quantified by combining calcite e-twin paleostress gauge. Then, the LSTM+Attention model is used to analyze the subsidence rate curve and magmatic activity record to automatically identify the tectonic transformation period. The Tectonic Transformation Index (TDI) is constructed to quantify the destruction / preservation weight of each period on the trap. This achieves a leap from static tectonic analysis to dynamic transformation period division, ensuring that the "paleotectonic high point" still has effectiveness after the hydrocarbon charging period (i.e., it has not been destroyed by later tectonic activity). Step 3: Integrate apatite fission track and (U-Th) / He thermochronological data, invert the spatiotemporal evolution of erosion based on the TecMod model, introduce random forest to optimize the weights of sediment flux and erosion rate to generate a probabilistic erosion distribution model, and simultaneously couple Delft3D paleohydrodynamic simulation to invert the source direction and channel migration path, outputting paleotectonic contour maps with error distribution and paleoriver probability fields. Paleotectonic high points (referring to tectonic high-level areas in geological history (including hydrocarbon charging periods)) are determined by paleotectonic contour maps (inverting erosion evolution and paleogeomorphological reconstruction); Step 4: Construct the Transformer-CGAN model, using sparse drilling data, seismic attribute volume, and paleogeographic elevation model as inputs to generate a 3D volume of sand body probability distribution. Enhance training by generating a synthetic dataset through SedFlux numerical simulation, coupled with a digital core pore network model to simulate the critical threshold for oil and gas charging, and generate a sand body connectivity index map to constrain the distribution of high-quality reservoirs, predict sand body connectivity, and ensure that reservoirs have good conduction and storage capacity in areas of ancient and modern tectonic high points. Step 5: Overlay multiple parameters such as sand body connectivity index (Step 4), TDI index (structural preservation conditions, Step 2), and source rock maturity in the overlapping area of ​​ancient and modern tectonic high points. Apply the SHAP algorithm to analyze the contribution of reservoir-controlling factors to focus on the main controlling factors, lock in favorable target areas for oil and gas enrichment, and train the Geo-LLM geological language model to deduce the probability distribution of reservoir formation scenarios. Step Six: Develop an active learning algorithm to prioritize and suggest drilling target areas (such as disputed sand body boundary areas), reduce uncertainty by updating the model using Bayesian methods, generate a 3D probability volume of favorable target areas, and combine it with the economic evaluation module to generate an exploration priority ranking.

[0018] In one embodiment, determining the current structural high points using the current structural contour map generated in step one includes the following steps: Data Preparation and Basement Model Construction: Integrating high-precision seismic data, well stratigraphic data, and paleomagnetic dating results, a three-dimensional geological model basement was constructed. A random forest algorithm was used to couple the detrital zircon U-Pb age with the cosmogenic nuclide erosion rate, establishing a velocity field model constrained by erosion rate and dating data. This quantifies the impact of stratigraphic erosion processes on the basement morphology, represented as... ,in U-Pb age, For the erosion rate, The thickness of the eroded strata. For erosion time, For rock density, To account for porosity, the model couples the erosion rate with dating data to constrain the basement uplift rate field; Structural Surface Interpolation and Error Simulation: Using well point elevation and seismic horizon as hard data and paleomagnetic correction dip angle as soft constraints, the initial structural surface is generated using the Kriging interpolation method, represented as follows: In the formula, For Kriging weights, For hard data, To satisfy the second-order stationarity assumption, a random error term is introduced, and the interpolation error term is randomly sampled using Monte Carlo simulation to generate the structural surface error distribution and 95% confidence interval, denoted as follows: Quantify the uncertainty of spatial prediction; The current method for determining structural high points is as follows: Contour lines (intervals of 10m) are plotted using the interpolation result Z(x,y). Combined with an error heatmap, the area of ​​structural high points is marked. The confidence interval for the elevation of the high points is calculated using error simulation results, and is expressed as follows: Combined with drilling verification (well C measured an elevation of 3245 m, which is within the confidence interval), the reliability of the high point location was confirmed.

[0019] In one embodiment, the construction-accumulation dynamic coupling simulation process in step two is as follows: Reconstruction of the Thermal-Fluid-Solid System Coupling Evolution History: The thermal-fluid-solid system coupling evolution history of the basin was reconstructed using the FLAC³D and TOUGHREACT coupled models. By solving the heat conduction equation, Darcy's fluid flow equation, and solid deformation equation, the dynamic coupling process of thermal history, fluid pressure, and tectonic deformation was quantified. The solved heat conduction equation is expressed as follows: In the formula, T is the temperature. Thermal conductivity, The Darcy fluid flow equation is expressed as follows, representing the terms of radioactive heat generation and frictional heat generation: In the formula, For Darcy velocity, For the permeability tensor, For fluid viscosity, Given the pressure gradient, the solid deformation equation is expressed as follows: In the formula, For effective stress, For bulk density, such as the simulation of the evolution of geothermal gradient during the Caledonian period in the Tarim Basin and the initial migration of hydrocarbons driven by fluid overpressure; Paleostress Field Reconstruction and Fracture Activity Quantification: The paleostress field was inverted using a calcite e-twin paleostress meter. The fracture activity intensity was quantified using a quantitative formula relating twin width and differential stress. The quantitative formula is expressed as follows: In the formula, For differential stress, G is the surface energy, and G is the shear modulus. Here, d represents Poisson's ratio, and d represents the calcite grain size. For example, in the calcite vein of fracture zone I in the tower, the twin width 2L = 15 μm is obtained. The maximum principal stress during the Hercynian period (320-280 Ma) is then calculated. =85MPa, fracture activity intensity index (FAI) reaches 0.72 (threshold >0.5 is strong activity); Tectonic transformation phase identification: Deploy a deep learning framework with LSTM and Attention mechanism to automatically extract tectonic transformation phases from subsidence rate curves and magmatic activity records, such as the intelligent identification of the late Caledonian subsidence surge and the early Hercynian magmatic peak in the Tazhong area. Construction of the Tectonic Change Index (TDI) and Assessment of Trap Effectiveness: The Tectonic Change Index (TDI) is constructed by integrating trap height loss, caprock pressure breakthrough, and multi-stage weighting to quantify the destructive / preservation effects of tectonic changes on hydrocarbon preservation. This is expressed as: In the formula, The weight for the i-th period is obtained by normalizing the LSTM attention weights. For the loss of trap height, The initial trap height, This represents the amount of pressure breakthrough in the caprock. To ensure the preservation of paleotectonic high points after the hydrocarbon inflow period, the evaluation results show that the Ordovician traps in the Tarim Basin retain 78% of their effectiveness due to Hercynian magma compensation.

[0020] In one embodiment, the specific steps for identifying the period of structural change are as follows: Data preprocessing and feature encoding: The settlement rate curves and magmatic activity records of the target area are standardized into a unified feature tensor through data preprocessing; LSTM-Attention Model Construction: This model utilizes an LSTM network to capture long-term dependencies in time series data. Its cell state update mechanism learns temporal patterns of geological events in the target region through dynamic regulation of the forget gate, input gate, and output gate. Subsequently, an attention mechanism is introduced, using an attention weight allocation formula... Quantifying the importance of each time step, in the formula, , This is the weight matrix. For bias terms; Model training and validation: The model is optimized using a weighted cross-entropy loss function, where the attention weighting coefficient enhances the recognition accuracy of key periods.

[0021] In this embodiment: Input data visualization: The subsidence rate curve shows a sudden increase to -120 m / Ma during the Late Caledonian period (420 Ma), while the magma frequency peaks at 0.08 Ma during the Early Hercynian period (300 Ma). -1 .

[0022] Attention weight analysis: The model outputs attention weights, highlighting two key periods: Late Caledonian period (420 Ma): =0.92, corresponding to basal fracture activation and rapid uplift.

[0023] Early Hercynian period (300 Ma) =0.88, associated with magmatic diapirs and thermal anomaly events.

[0024] Tectonic change explanation: Late Caledonian: High subsidence rate reflects intense basement faulting activity, leading to the migration of paleotectonic high points, but not completely destroying the traps (TDI=0.45).

[0025] Early Hercynian period: The peak of magmatic activity triggered regional thermal anomalies, forming a high-pressure storage box that compensated for the earlier tectonic damage (TDI=0.32).

[0026] Compared with independent observations: Late Caledonian fracture activation is consistent with the results of calcite e-twin paleostressing. =85 MPa).

[0027] The early magmatic activity in Hercynian cratons is consistent with zircon U-Pb ages (305±2 Ma), validating the reliability of the model.

[0028] Using the LSTM-Attention framework, the model automatically extracted the tectonic transformation periods of the late Caledonian and early Hercynian periods in the central Tarim Basin from multi-parameter time series, and the attention weights quantitatively indicated the impact of key events on hydrocarbon accumulation.

[0029] In one embodiment, the process for constructing the probability field of paleotectonic high points and paleoriver systems in step three is as follows: Thermochronological data integration and spatiotemporal evolution inversion of erosion: Integrating apatite fission track (AFT) and (U-Th) / He age data, the spatiotemporal evolution of erosion was inverted based on the TecMod model. Age was simulated using the age-erosion relationship, which is expressed as follows: In the formula, For thermochronological age, The decay constant is This refers to the amount of erosion. The sample tilt angle, The empirical coefficients for etch pit diameter and track density are used to optimize the erosion field D(x,y,t) using the Markov chain Monte Carlo (MCMC) algorithm. This minimizes the errors between the simulated age and the measured AFT age and (U-Th) / He age (50-90 Ma), and is expressed as follows: ; For example, samples from the northern Tarim Basin have ages of 80-120 Ma. Inversion shows that the Mesozoic erosion in the northern Tarim Basin is characterized by a pattern of "high in the east and low in the west". The erosion in the east reaches 2500 m (AFT closure depth ~6 km), while the erosion in the west is only 1200 m, which is consistent with the distribution of paleotectonic high points. Deposition-erosion weight optimization and probabilistic erosion quantity modeling: The random forest algorithm is introduced to optimize the weight allocation of deposition flux and erosion rate, construct a probabilistic erosion quantity distribution model, generate the probability distribution of erosion quantity, and quantify the contribution of deposition flux and erosion rate to the erosion process. Paleohydrodynamic simulation and source-channel inversion: Coupled with the Delft3D paleohydrodynamic model, the paleosource direction and channel migration path are inverted by solving the shallow water equations and parameterizing the Manning coefficient and boundary flow. The shallow water equations are expressed as follows: ; ; In the formula, Because of the water depth, For flow velocity vectors, The elevation of the subgrade. This refers to the shear stress of the subbed.

[0030] Parameterization scheme: Manning coefficient n: n=0.03 for sandstone channels and n=0.05 for mudstone floodplains.

[0031] Boundary conditions: Inlet flow =5000 (Based on paleoclimate simulation).

[0032] Paleotectonic high point determination and error quantification: By integrating erosion data and paleoriver data, paleogeographic reconstruction formulas and error propagation Monte Carlo simulations are used to generate paleotectonic contour maps with error distribution.

[0033] In one embodiment, the sand body connectivity prediction and high-quality reservoir constraint process in step four is as follows: Transformer-CGAN Sandbody Probability Volume Generation: A Transformer-CGAN model is constructed, taking sparse drilling data, seismic attribute volume and paleogeographic elevation model as input. The multi-head self-attention mechanism of the Transformer encoder is used to capture the long-range spatial correlation of the input data. Combined with the adversarial training strategy of Conditional Generative Adversarial Network (CGAN), a three-dimensional volume of sandbody probability distribution is generated (generator G generates sandbody probability volume with noise z and condition c (seismic + paleogeographic). SedFlux Synthetic Data Augmentation: A SedFlux sedimentary process simulator is deployed to generate a synthetic seismic attribute-sandbody probability pair dataset by solving the diffusion equation and sedimentation rate formula. The diffusion equation is expressed as follows: In the formula, h is the water depth, D is the diffusion coefficient, and Q is the source-sink term (such as river input). The deposition rate formula is expressed as: In the formula, For the inflow volume, The concentration of suspended solids. For the deposition area, To increase porosity and expand the diversity of training samples; Digital core pore network model (PNM) simulation: The digital core pore network model (PNM) is used to quantify the critical threshold for oil and gas charging (quantify the critical threshold for oil and gas charging (0.5-1.2MPa) corresponding to the effective pore throat radius (5-15μm) and clarify the lower limit of reservoir permeability) and construct the sand body connectivity index; Sandbody Connectivity Index Map Generation: Integrating sandbody probabilistic volume and critical threshold results, the connectivity index (CI) formula is used to quantify sandbody size, permeability-weighted average, and the proportion of isolated sandbody bodies, generating a connectivity index map. The connectivity index (CI) formula is expressed as follows: In the formula, For the area of ​​the sand body grid, For penetration rate, The number of isolated sand bodies.

[0034] In one embodiment, the specific steps for multi-parameter constraints and favorable area prediction in step five of the ancient and modern tectonic overlapping region are as follows: Analysis of the superposition of ancient and modern tectonic high points: Spatial superposition of ancient and modern tectonic high points identifies regions that possess both tectonic inheritance and preservation conditions. The formula for determining the superposition zone is expressed as follows: In the formula, The elevation of the ancient tectonic high point This represents the lower limit of the elevation of the ancient tectonic high point. To construct the current high point elevation, This represents the lower limit of the current highest point's altitude.

[0035] Example: The paleohigh point of a certain tectonic zone has an elevation of 1800 m (result of inversion in step 3), and the current high point has an elevation of 3250 m (result of interpolation in step 1). It meets the superposition condition and is marked as a candidate area.

[0036] Multi-parameter synergistic constraints and comprehensive evaluation: Within the overlapping area, sand body connectivity index (CI), tectonic transformation index (TDI), and source rock maturity (Ro) are superimposed to quantify hydrocarbon accumulation conditions. The comprehensive evaluation formula is expressed as follows: In the formula, =0.5、 =0.3、 =0.2 is the weight (optimized by the SHAP algorithm). =0.7%, =1.5% is the threshold for hydrocarbon generation from source rocks.

[0037] Example: Parameters within the overlapping region: CI=0.72 (predicted in step four), TDI=0.28 (calculated in step two), Ro=1.2% (measured), overall score. =0.81, significantly higher than the non-overlapping region (mean 0.56).

[0038] The SHAP algorithm analyzes the contribution of hydrocarbon accumulation control factors: it quantifies the marginal contribution of each parameter to hydrocarbon accumulation prediction, focuses on the main control factors, and the SHAP value calculation formula is as follows: In the formula, z is the feature subset. This is the prediction function for the XGBoost model. The total number of features; Favorable target area identification and verification: Based on comprehensive score and main control factor analysis, favorable oil and gas enrichment areas are identified; Geo-LLM (Geological Big Language Model) for hydrocarbon accumulation scenario inference: Based on geological knowledge graphs and historical hydrocarbon accumulation cases, it infers the probability of hydrocarbon accumulation under different parameter combinations. It adopts BERT pre-training + fine-tuning strategy, with geological parameter sequences as input and hydrocarbon accumulation probability as output.

[0039] In one embodiment, in step six, the active learning optimization and exploration economic evaluation process is as follows: Active learning algorithms prioritize the densification of drilling target areas: Active learning algorithms are deployed to quantify the prediction uncertainty of high-risk areas such as disputed sand body boundaries based on an uncertainty sampling strategy, using formulas... Identify the in-place drilling target area most sensitive to model optimization, where, For the prediction uncertainty of position x, This represents the predicted probability of the current model for category c (e.g., sandstone / mudstone). Based on existing drilling data, regions with U(x) > 0.3 are selected as priority target areas for infill drilling; Bayesian update model reduces prediction uncertainty: A Bayesian update model is used to fuse new drilling data, and Gaussian process regression is used for posterior correction of the sand body probability volume, expressed as follows: In the formula, and The updated mean and variance are obtained using the formula. Dynamically reduce prediction variance, where, For model parameters, For new drilling data, Prior data; Generation of 3D favorable target area probability volume: Integrating the updated sand body probability volume, tectonic transformation index (TDI), and source rock maturity, and using a comprehensive probability formula. Generate a three-dimensional probability field of favorable target areas to quantify spatial exploration priorities; The economic evaluation module generates an exploration priority ranking: This is coupled with the economic evaluation module and based on the net present value (NPV) formula. Calculate the economic benefits of each target area, where, For oil prices, Annual oil production For operating costs, Let r be the geological success rate, and r be the discount rate (10%). For capital expenditures, an exploration priority sequence is generated by sorting them in descending order of NPV / risk coefficient.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting hydrocarbon injection based on superimposed ancient and modern geological structures, characterized in that, Includes the following steps: Step 1: Construct a three-dimensional geological model base using seismic data, well formation data, and paleomagnetic dating results. Combine random forest algorithm to collaboratively constrain detrital zircon U-Pb dating data and optimize velocity field modeling with cosmogenic nuclide erosion rate. Then, generate structural surface error distribution through Monte Carlo simulation to mark the confidence interval of the current structural map. Output current structural contour map and uncertainty heat map with both spatial accuracy and uncertainty quantification. Step 2: The FLAC³D+TOUGHREACT coupled model is used to reconstruct the thermal-fluid-solid coupled evolution history of the basin. The paleostress field is restored by calcite e-twin paleostress gauge to quantify the intensity of fracture activity. Then, the LSTM+Attention model is used to analyze the subsidence rate curve and magmatic activity record to automatically identify the tectonic transformation period. The tectonic transformation index is constructed to quantify the destruction / preservation weight of each period on the trap. Step 3: Integrate apatite fission track and (U-Th) / He thermochronological data, invert the spatiotemporal evolution of erosion based on the TecMod model, introduce random forest to optimize the weights of sediment flux and erosion rate to generate a probabilistic erosion distribution model, and couple Delft3D paleohydrodynamic simulation to invert the source direction and channel migration path, outputting paleotectonic contour maps with error distribution and paleoriver probability fields; Step 4: Construct the Transformer-CGAN model, using sparse drilling data, seismic attribute volume and paleogeographic elevation model as input to generate a three-dimensional volume of sand body probability distribution, coupled with a digital core pore network model to simulate the critical threshold of oil and gas charging, and generate a sand body connectivity index map to constrain the distribution of high-quality reservoirs. Step 5: Overlay multiple parameters in the overlapping area of ​​ancient and modern tectonic high points, apply the SHAP algorithm to analyze the contribution of reservoir-controlling factors to focus on the main controlling factors, lock in favorable target areas for oil and gas enrichment, and at the same time train the geological big data model to deduce the probability distribution of reservoir formation scenarios. Step Six: Develop an active learning algorithm to prioritize and suggest drilling target areas, reduce uncertainty by updating the model using Bayesian methods, generate a 3D probability volume of favorable target areas, and combine it with the economic evaluation module to generate an exploration priority ranking.

2. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 1, characterized in that, Determining current structural high points using the current structural contour map generated in step one involves the following steps: Data Preparation and Basement Model Construction: Integrating high-precision seismic data, well stratigraphic data, and paleomagnetic dating results, a three-dimensional geological model basement was constructed. A random forest algorithm was used to couple the detrital zircon U-Pb age with the cosmogenic nuclide erosion rate, establishing a velocity field model constrained by erosion rate and dating data. This quantifies the impact of stratigraphic erosion processes on the basement morphology, represented as... ,in U-Pb age, For the erosion rate, The thickness of the eroded strata. For erosion time, For rock density, Porosity; Structural Surface Interpolation and Error Simulation: Using well point elevation and seismic horizon as hard data and paleomagnetic correction dip angle as soft constraints, the initial structural surface is generated using the Kriging interpolation method, represented as follows: In the formula, For Kriging weights, For hard data, To satisfy the second-order stationarity assumption, a random error term is introduced, and the interpolation error term is randomly sampled using Monte Carlo simulation to generate the structural surface error distribution and 95% confidence interval, denoted as follows: Quantify the uncertainty of spatial prediction; The current method for determining structural high points is as follows: contour lines are drawn using the interpolation result Z(x,y), and the structural high point region is marked by combining the error heat map. The confidence interval of the high point elevation is calculated using the error simulation results.

3. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 1, characterized in that, In step two, the dynamic coupling simulation process of structure-accumulation is as follows: Reconstruction of the Thermal-Fluid-Solid System Coupling Evolution History: The thermal-fluid-solid system coupling evolution history of the basin was reconstructed using the FLAC³D and TOUGHREACT coupled models. By solving the heat conduction equation, Darcy's fluid flow equation, and solid deformation equation, the dynamic coupling process of thermal history, fluid pressure, and tectonic deformation was quantified. The solved heat conduction equation is expressed as follows: In the formula, T is the temperature. Thermal conductivity, The Darcy fluid flow equation is expressed as follows, representing the terms of radioactive heat generation and frictional heat generation: In the formula, For Darcy velocity, For the permeability tensor, For fluid viscosity, Given the pressure gradient, the solid deformation equation is expressed as follows: In the formula, For effective stress, Bulk density; Paleostress Field Reconstruction and Fracture Activity Quantification: The paleostress field was inverted using a calcite e-twin paleostress meter. The fracture activity intensity was quantified using a quantitative formula relating twin width and differential stress. The quantitative formula is expressed as follows: In the formula, For differential stress, G is the surface energy, and G is the shear modulus. d is Poisson's ratio, and d is the calcite grain size; Tectonic change phase identification: Deploy a deep learning framework with LSTM and Attention mechanism to automatically extract tectonic change phases from subsidence rate curves and magmatic activity records; Construction of the Tectonic Transformation Index and Evaluation of Trap Effectiveness: A tectonic transformation index is constructed, which integrates trap height loss, caprock pressure breakthrough, and multi-stage weighting to quantify the destructive / preservation effects of tectonic transformation on hydrocarbon preservation, expressed as: In the formula, For the i-th period, For the loss of trap height, The initial trap height, This represents the amount of pressure breakthrough in the caprock. This is to overcome the pressure of the cap layer.

4. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 3, characterized in that, The specific steps for identifying the period of structural change are as follows: Data preprocessing and feature encoding: The settlement rate curves and magmatic activity records of the target area are standardized into a unified feature tensor through data preprocessing; LSTM-Attention Model Construction: This model utilizes an LSTM network to capture long-term dependencies in time series data. Its cell state update mechanism learns temporal patterns of geological events in the target region through dynamic regulation of the forget gate, input gate, and output gate. Subsequently, an attention mechanism is introduced, using an attention weight allocation formula... Quantifying the importance of each time step, in the formula, , This is the weight matrix. For bias terms; Model training and validation: The model is optimized using a weighted cross-entropy loss function, where the attention weighting coefficient enhances the recognition accuracy of key periods.

5. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 1, characterized in that, In step three, the process for constructing the probability field of paleotectonic high points and paleoriver systems is as follows: Thermochronological data integration and spatiotemporal evolution inversion of erosion: Integrating apatite fission tracks and (U-Th) / He age data, the spatiotemporal evolution of erosion was inverted based on the TecMod model. Age was simulated using the age-erosion relationship, which is expressed as follows: In the formula, For thermochronological age, The decay constant is Erosion amount, The sample tilt angle, The empirical coefficients for the diameter of the etch pits and the track density are used. The erosion field D(x,y,t) is optimized by the Markov chain Monte Carlo algorithm to minimize the error between the simulated age and the measured AFT age and (U-Th) / He age. Deposition-erosion weight optimization and probabilistic erosion quantity modeling: The random forest algorithm is introduced to optimize the weight allocation of deposition flux and erosion rate, construct a probabilistic erosion quantity distribution model, generate the probability distribution of erosion quantity, and quantify the contribution of deposition flux and erosion rate to the erosion process. Paleohydrodynamic simulation and source-channel inversion: Coupled with the Delft3D paleohydrodynamic model, the paleosource direction and channel migration path are inverted by solving the shallow water equations and parameterizing the Manning coefficient and boundary flow. Paleotectonic high point determination and error quantification: By integrating erosion data and paleoriver data, paleogeographic reconstruction formulas and error propagation Monte Carlo simulations are used to generate paleotectonic contour maps with error distribution.

6. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 1, characterized in that, In step four, the sand body connectivity prediction and high-quality reservoir constraint process is as follows: Transformer-CGAN Sandbody Probability Volume Generation: Construct a Transformer-CGAN model, using sparse drilling data, seismic attribute volume and paleogeographic elevation model as input, and use the multi-head self-attention mechanism of the Transformer encoder to capture the long-distance spatial correlation of the input data, and combine the adversarial training strategy of conditional generative adversarial network to generate a three-dimensional volume of sandbody probability distribution. SedFlux Synthetic Data Augmentation: A SedFlux sedimentary process simulator is deployed to generate a synthetic seismic attribute-sandbody probability pair dataset by solving the diffusion equation and sedimentation rate formula. The diffusion equation is expressed as follows: In the formula, h is the water depth, D is the diffusion coefficient, Q is the source-sink term, and the deposition rate formula is expressed as: In the formula, For the inflow volume, The concentration of suspended solids. For the deposition area, Porosity; Digital core pore network model simulation: The digital core pore network model is used to quantify the critical threshold for oil and gas injection and to construct the sand body connectivity index; Sandbody connectivity index map generation: Integrating sandbody probabilistic volume and critical threshold results, the connectivity index formula is used to quantify sandbody size, permeability-weighted average, and the proportion of isolated sandbody bodies, generating a connectivity index map. The connectivity index formula is expressed as follows: In the formula, For the area of ​​the sand body grid, For penetration rate, The number of isolated sand bodies.

7. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 1, characterized in that, In step five, the specific steps for multi-parameter constraints and favorable area prediction in the overlapping ancient and modern tectonic zones are as follows: Analysis of the superposition of ancient and modern tectonic high points: Spatial superposition of ancient and modern tectonic high points identifies regions that possess both tectonic inheritance and preservation conditions. The formula for determining the superposition zone is expressed as follows: In the formula, The elevation of the ancient tectonic high point This represents the lower limit of the elevation of the ancient tectonic high point. To construct the current high point elevation, This represents the lower limit of the current high point's altitude. Multi-parameter synergistic constraints and comprehensive evaluation: Within the overlapping area, sand body connectivity index (CI), tectonic transformation index (TDI), and source rock maturity (Ro) are superimposed to quantify hydrocarbon accumulation conditions. The comprehensive evaluation formula is expressed as follows: In the formula, , , =0.2 is the weight. , The threshold for hydrocarbon generation from source rocks; The SHAP algorithm analyzes the contribution of hydrocarbon accumulation control factors: it quantifies the marginal contribution of each parameter to hydrocarbon accumulation prediction, focuses on the main control factors, and the SHAP value calculation formula is as follows: In the formula, z is the feature subset. This is the prediction function for the XGBoost model. The total number of features; Favorable target area identification and verification: Based on comprehensive score and main control factor analysis, favorable oil and gas enrichment areas are identified; Geological Big Language Model for Hydrocarbon Accumulation Scenario Inference: Based on geological knowledge graphs and historical hydrocarbon accumulation cases, this study infers the probability of hydrocarbon accumulation under different parameter combinations. It adopts a BERT pre-training + fine-tuning strategy, with geological parameter sequences as input and hydrocarbon accumulation probabilities as output.

8. The method for predicting hydrocarbon injection based on superimposed ancient and modern structures according to claim 7, characterized in that, In step six, the active learning optimization and exploration economic evaluation process is as follows: Active learning algorithms are prioritized for recommending encrypted drilling target areas: Active learning algorithms are deployed to quantify the prediction uncertainty of high-risk areas based on an uncertainty sampling strategy, using formulas... Identify the in-place drilling target area most sensitive to model optimization, where, For the prediction uncertainty of position x, This represents the predicted probability of the current model for category c). Based on existing drilling data, regions with U(x) > 0.3 are selected as priority target areas for infill drilling; Bayesian update model reduces prediction uncertainty: A Bayesian update model is used to fuse new drilling data, and Gaussian process regression is used for posterior correction of the sand body probability volume, expressed as follows: In the formula, and The updated mean and variance are obtained using the formula. Dynamically reduce prediction variance, where, For model parameters, For new drilling data, Prior data; Generation of 3D favorable target area probability volume: Integrating the updated sand body probability volume, tectonic transformation index (TDI), and source rock maturity, and using a comprehensive probability formula. Generate a three-dimensional probability field of favorable target areas to quantify spatial exploration priorities; The economic evaluation module generates an exploration priority ranking: This is coupled with the economic evaluation module and based on the net present value formula. Calculate the economic benefits of each target area, where, For oil prices, Annual oil production For operating costs, Let r be the geological success rate, and r be the discount rate. For capital expenditures, an exploration priority sequence is generated by sorting them in descending order of NPV / risk coefficient.

Images