Real-time porosity prediction method and system based on alkane gas and carbon isotope characteristics thereof and storage medium

Abstract

The invention discloses a porosity real-time prediction method and system based on alkane gas and carbon isotope characteristics thereof and a storage medium, relates to the technical field of oil and gas reservoir development, and aims to solve the problem of insufficient precision in a heterogeneous reservoir due to neglect of dynamic characteristics of alkane gas carbon isotope in an existing method. The method is characterized by comprising the following steps: S100, collecting alkane gas concentration and carbon isotope data thereof in real time, and performing time-frequency decomposition on the data to obtain characteristic data on different time and frequency scales; s200, performing multi-scale feature extraction on the data to obtain multi-scale features; s300, performing feature fusion on the multi-scale features; s400, constructing a GNN network model, introducing an adaptive attention mechanism, residual connection, jump aggregation and time sequence dynamic fusion, constructing a graph structure by taking a reservoir unit as a node, and mapping a fusion feature to a graph node attribute; and S500, constructing a reservoir space structure based on the GNN network model, and combining physical constraint optimization to realize real-time prediction of the porosity.

Description

Technical Field

[0001] This invention relates to the field of oil and gas reservoir development technology, and more specifically, to a method, system, and storage medium for real-time porosity prediction based on alkane gas and its carbon isotope characteristics. Background Technology

[0002] Reservoir porosity prediction is a core issue in oil and gas exploration and development, directly impacting reservoir evaluation and development efficiency. Traditional porosity prediction methods include empirical formulas, geophysical logging, and machine learning methods, but all have limitations:

[0003] 1. Empirical formulas rely on a large amount of core experimental data, which makes them difficult to adapt to heterogeneous reservoirs and cannot be updated in real time.

[0004] 2. Well logging methods are susceptible to noise interference, making it difficult to capture changes in reservoir microstructure, and data acquisition is delayed, making real-time prediction difficult.

[0005] 3. Although machine learning methods improve accuracy to a certain extent, they ignore the dynamic characteristics of carbon isotopes in alkane gases, resulting in insufficient accuracy in heterogeneous reservoirs and difficulty in processing dynamic data in real time and stably. Summary of the Invention

[0006] The technical problem to be solved by this invention is:

[0007] Existing machine learning methods ignore the dynamic characteristics of carbon isotopes in alkane gases, resulting in insufficient accuracy in heterogeneous reservoirs.

[0008] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0009] This invention provides a method for real-time prediction of porosity based on alkane gas and its carbon isotope characteristics, comprising the following steps:

[0010] S100, real-time acquired alkane gas concentration and carbon isotope data, time-frequency decomposition of the data to obtain characteristic data at different time and frequency scales;

[0011] S200: Perform multi-scale feature extraction on the feature data obtained in S100 to obtain multi-scale features;

[0012] S300: The multi-scale features obtained from S200 are fused to obtain fused features;

[0013] S400. Construct a GNN network model. The model is based on the Graph Convolutional Network (GCN) framework and introduces an adaptive attention mechanism, residual connections and skip aggregation, as well as temporal dynamic fusion. Construct a graph structure with reservoir units as nodes, map the fused features to graph node attributes, and train the GNN network model based on the graph structure data.

[0014] S500 constructs the reservoir space structure based on the trained GNN network model, and combines physical constraint optimization to achieve real-time prediction of porosity.

[0015] Furthermore, S100 includes the following steps:

[0016] S110: Real-time acquisition of well depth, drill bit depth, alkane gas concentration, carbon isotope value and peak integral area data, and processing of missing and outlier values ​​in the data;

[0017] S120. Perform noise reduction and signal curve smoothing on the data;

[0018] S130. Wavelet transform is performed using the Daubechies wavelet basis and the Mallat algorithm.

[0019] S140, decompose low-frequency and high-frequency components to obtain multi-scale component data.

[0020] Furthermore, S200 includes the following steps:

[0021] S210. Divide the data into low-frequency and high-frequency data;

[0022] S220, Extracting abrupt changes in carbon isotope values ​​or gas concentrations from high-frequency components;

[0023] S230. Extract the overall trend characteristics of gas concentration or porosity from low-frequency components;

[0024] S240. Calculate data statistics to quantify the data.

[0025] Furthermore, S300 includes the following steps:

[0026] S310. Concatenate feature vectors of different scales;

[0027] S320. Evaluate the correlation between characteristics and porosity through mutual information;

[0028] S330, Weight allocation based on mutual information;

[0029] S340. Dimensionality reduction is performed on the data to obtain the final fused features.

[0030] Furthermore, S400 includes the following steps:

[0031] S410. Divide reservoir units according to well depth and alkane gas characteristics;

[0032] S420. Construct an undirected graph based on spatial adjacency or feature similarity, using reservoir units as nodes.

[0033] S430, Map the fused features to graph node attributes;

[0034] S440. Construct a GNN network model. The model is based on the Graph Convolutional Network (GCN) framework and introduces adaptive attention mechanism, residual connection and skip aggregation, as well as temporal dynamic fusion. The GNN network model is trained based on graph structure data so that the model can learn spatial dependencies.

[0035] Furthermore, the process by which the GNN network model in S440 learns spatial dependencies is as follows:

[0036] First, a time-series update is performed, using a gated recurrent unit (GRU) to fuse real-time sequence information:

[0037]

[0038] Then, an adaptive attention mechanism is introduced to calculate attention weights and enhance the normalized adjacency matrix:

[0039]

[0040]

[0041] Finally, graph convolution updates are performed, incorporating residual connections, skip aggregation, and layer normalization:

[0042]

[0043] in, Let the (l+1)th layer node representation matrix be used; For activation functions; This is the normalized adjacency matrix; Here is the learnable weight matrix for the l-th layer; This is the feature representation matrix of all nodes in the l-th layer; Let i be the feature vector of node i in the l-th layer; This is the intermediate feature matrix of the l-th layer after being updated by the temporal GRU; This refers to the (l-1)th hidden state of the previous time step or the previous depth unit. Let be the set of the first-order neighborhood of node i; The activation function for leakage rectification; The attention weights of node j to i (Softmax normalized); This is the adjacency matrix after attention enhancement; The initial node feature representation matrix; LayerNorm is the layer normalization; This is a learnable attention vector.

[0044] Furthermore, the S500 includes the following steps:

[0045] S510. Generate preliminary porosity prediction results based on graph neural networks;

[0046] S520, limits the porosity range and embeds the porosity-permeability relationship;

[0047] S530. Construct a loss function by combining prediction bias and physical constraint penalties, and optimize model parameters:

[0048]

[0049] in, Total loss; To predict porosity; For true porosity, Here, N represents the weights, and N represents the number of samples. This represents the true penetration rate.

[0050] S540. Based on the optimized model, generate porosity results and smooth the results.

[0051] Furthermore, the porosity-permeability relationship described in S520 is as follows:

[0052]

[0053] in For penetration rate; It is a proportionality constant; Porosity; For exponents.

[0054] This invention provides a real-time porosity prediction system based on the characteristics of alkane gas and its carbon isotopes. The system has a program module corresponding to the steps of the method described in any of the above technical solutions, and executes the steps in the above-described real-time porosity prediction method based on the characteristics of alkane gas and its carbon isotopes when running.

[0055] The present invention also provides a computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps in the real-time porosity prediction method based on alkane gas and its carbon isotope characteristics as described in any of the above technical solutions.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] 1. Real-time prediction capability: By acquiring alkane gas concentration and carbon isotope values ​​in real time, combined with fast feature extraction and GNN modeling, dynamic prediction of porosity during drilling is achieved; an adaptive attention mechanism is introduced to dynamically adjust neighborhood weights, and residual connections and skip aggregation alleviate over-smoothing, which can capture the reservoir heterogeneity reflected by alkane gas and carbon isotope characteristics. Through spatial structure modeling, the accuracy of porosity prediction is significantly improved. >0.85).

[0058] 2. Introduce physical constraints to ensure the stability of the model under high noise and complex geological conditions (error <5% at a noise level of 20%).

[0059] 3. High computational efficiency: Through feature fusion and dimensionality reduction techniques, computational complexity is reduced to meet the needs of real-time prediction.

[0060] 4. High practicality: Generates smooth and continuous porosity distribution maps, providing reliable decision support for oil and gas exploration and reducing costs by approximately 30%.

[0061] 5. The method of this invention can be integrated into a cloud platform to support multi-well collaborative prediction and improve the overall oilfield development efficiency.

[0062] This invention is the first to integrate carbon isotopes as real-time signals with GNN spatial modeling, innovatively improving real-time decision-making efficiency by more than 30%. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the real-time porosity prediction process based on alkane gas and its carbon isotope characteristics in an embodiment of the present invention.

[0064] Figure 2 This is a diagram illustrating the data preprocessing process in an embodiment of the present invention;

[0065] Figure 3 This is a wavelet transform feature extraction diagram from an embodiment of the present invention;

[0066] Figure 4 This is a flowchart of the feature fusion and optimization process in an embodiment of the present invention;

[0067] Figure 5 This is the network graph for reservoir map structure construction and spatial relationship learning in this embodiment of the invention;

[0068] Figure 6 This is a line graph comparing the prediction results with core data in this embodiment of the invention;

[0069] Figure 7 This is a bar chart comparing the predictions and actual measurements in an embodiment of the present invention;

[0070] Figure 8This is an accuracy curve diagram under different noise levels in an embodiment of the present invention. Detailed Implementation

[0071] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0072] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0073] Specific Implementation Plan 1: (e.g.) Figure 1 As shown, this invention provides a real-time porosity prediction method based on alkane gas and its carbon isotope characteristics, comprising the following steps:

[0074] S100, real-time acquired alkane gas concentration and carbon isotope data, time-frequency decomposition of the data to obtain characteristic data at different time and frequency scales;

[0075] S200: Perform multi-scale feature extraction on the feature data obtained in S100 to obtain multi-scale features;

[0076] S300: The multi-scale features obtained from S200 are fused to obtain fused features;

[0077] S400. Construct a GNN network model. The model is based on the Graph Convolutional Network (GCN) framework and introduces an adaptive attention mechanism, residual connections and skip aggregation, as well as temporal dynamic fusion. Construct a graph structure with reservoir units as nodes, map the fused features to graph node attributes, and train the GNN network model based on the graph structure data.

[0078] S500 constructs the reservoir space structure based on the trained GNN network model, and combines physical constraint optimization to achieve real-time prediction of porosity.

[0079] Specific Implementation Plan Two: (e.g.) Figure 2 As shown, S100 includes the following steps:

[0080] S110: Real-time acquisition of well depth, drill bit depth, alkane gas concentration, carbon isotope value and peak integral area data, and removal of outliers and missing values;

[0081] Missing value handling: For single missing values, linear interpolation is used for imputation.

[0082]

[0083] in Indicates the current missing value. and For adjacent known values, the missing points are estimated by averaging to avoid data interruption.

[0084] For consecutive missing values, cubic spline interpolation is used:

[0085]

[0086] Where the coefficient Solving by boundary conditions at adjacent points (including continuous first derivatives) provides a smooth curve fit for continuous missing values.

[0087] Detecting outliers using Z-score:

[0088]

[0089] in For standardized scores, For the current data point, The mean, Let be the standard deviation, if Adjust according to the IQR methodology: Exceeding Values ​​with an anomalous value are considered outliers and replaced with the average of two adjacent values.

[0090]

[0091] Eliminating dimensional differences through data standardization:

[0092]

[0093] in For standardized values, The original value, The mean, The standard deviation is denoted as .

[0094] S120. Noise removal using moving average filtering:

[0095]

[0096] in The smoothed value, Half the width of the window For data points within a window, local averaging is used to reduce random fluctuations.

[0097] Wavelet thresholding denoising is performed using the soft thresholding method:

[0098]

[0099]

[0100] in Wavelet coefficients, threshold , The standard deviation of noise. This is the signal length.

[0101] By removing noise and smoothing signal curves, the signal-to-noise ratio of the data is improved, providing a cleaner input for time-frequency decomposition and ensuring the gradual optimization of data from coarse to refined.

[0102] S130. Select the Daubechies wavelet basis (including db4) and estimate the number of decomposition levels using the following formula. :

[0103]

[0104] in The number of decomposition layers, For signal length, Rounding down; automatically determining the number of levels based on the data size to avoid over-decomposition.

[0105] Fast wavelet transform using the Mallat algorithm:

[0106]

[0107] in and These are low-pass and high-pass filters, respectively.

[0108] The basis function db6 is selected to accommodate different noise levels. For low-frequency approximation coefficients, For high-frequency detail coefficients, It represents convolution; and enables efficient multi-resolution decomposition.

[0109] By selecting appropriate wavelet tools and parameters, decomposition efficiency and accuracy can be ensured.

[0110] S140, decomposes low-frequency (global trend, including gas concentration trend) and high-frequency (local variation, including carbon isotope mutation) components. Output: Low-frequency components after decomposition. and high frequency coefficient.

[0111]

[0112] in: Indicates the first The low-frequency approximation coefficients obtained from the level decomposition represent the global trend of the signal. Indicates the first The high-frequency detail coefficients obtained from the hierarchical decomposition capture the local changes in the signal; This represents the number of decomposition layers.

[0113] By separating the multi-scale components of the signal, it is easier to analyze trends (low frequency) and fluctuations (high frequency) separately, laying the foundation for subsequent feature extraction.

[0114] S150. The stationarity of the data was verified by the ADF test, and the correlation between the data and the core data was verified by the Pearson correlation coefficient:

[0115]

[0116] in This is the correlation coefficient, ranging from [-1, 1]. A value close to 1 indicates a strong positive correlation. For data pairs, The formula uses the mean as its metric to quantify the linear relationship between variables.

[0117] If r < 0.5, a re-collection warning is triggered. The preprocessing effect is evaluated by quantifying the linear relationship between variables.

[0118] This implementation plan is otherwise the same as Specific Implementation Plan 1.

[0119] Specific implementation plan three: S200 includes the following steps:

[0120] S210. Divide the data into low-frequency (global features, including gas concentration trends) and high-frequency (local features, including carbon isotope fluctuations) to provide a scale framework for feature extraction.

[0121] S220, such as Figure 3 As shown, abrupt changes in carbon isotope values ​​or gas concentrations are extracted from high-frequency components:

[0122]

[0123] in These are high-frequency detail components, representing local fluctuations; For detailed coefficients, quantify the fluctuation range; is a wavelet function used for the reconstruction of scale j and position k.

[0124] By reconstructing the local abrupt change characteristics of high-frequency signals using coefficients and wavelet basis functions, transient information (including carbon isotope mutations) can be captured, reflecting microscopic changes in the reservoir.

[0125] S230, Extracting the overall trend of gas concentration or porosity from low-frequency components:

[0126]

[0127] in This is a trend function, representing global changes; The slope; The intercept;

[0128] By fitting the overall trend of low-frequency components using linear regression, long-term patterns (including gradual changes in gas concentration) can be identified, supplementing the global perspective of local features.

[0129] S240: Quantify features by calculating statistical measures such as mean, variance, and peak value.

[0130]

[0131] Mean, central trend indicator; Variance is a measure of dispersion. Peak value, maximum value indicator; The number of samples; For data points.

[0132]

[0133] This is an index of skewness and asymmetry in distribution.

[0134] By transforming abstract features into computable metrics, subsequent fusion becomes easier.

[0135] S250. The effectiveness of the extracted features is ensured through Pearson correlation coefficient and cross-validation, guaranteeing both the validity and reliability of the extracted features.

[0136]

[0137] The coefficient of determination is the goodness-of-fit index. This is a predicted value; The actual value; This is the mean.

[0138] like Then the scale division is optimized.

[0139] This implementation plan is otherwise the same as Implementation Plan Two.

[0140] Specific Implementation Plan Four: (e.g.) Figure 4 As shown, S300 includes the following steps:

[0141] S310, Concatenating feature vectors of different scales:

[0142]

[0143] in This is the initial fusion vector; Let i be the eigenvector of the i-th scale.

[0144] Multi-source features are initially aggregated by vector concatenation to form a unified representation.

[0145] S320. Evaluate the correlation between characteristics and porosity through mutual information:

[0146]

[0147] in Mutual information value, a non-linear dependency index; For joint probability density; The marginal probability density; the formula quantifies the information sharing between feature X and target Y;

[0148] By identifying key features, redundant introductions can be avoided.

[0149] S330, Weighted fusion based on mutual information weight allocation:

[0150]

[0151]

[0152] For weighted fusion features; The weights of the i-th feature (normalization ensures) (Emphasis on relevant features through weighted summation); Improve sensitivity to porosity by optimizing the fusion representation.

[0153] S340. Reduce feature dimensionality using PCA:

[0154]

[0155] in The feature matrix after dimensionality reduction; This is the original fusion matrix; Principal component weight matrix;

[0156] By projecting features into a low-dimensional space, principal components with a contribution rate >90% are retained. This reduces computational complexity and improves real-time efficiency.

[0157] S350, via MSE and Evaluate the fusion effect:

[0158]

[0159]

[0160] in Mean squared error, a prediction bias index; The coefficient of determination is the goodness-of-fit index. For prediction The actual value; The average value is used; the formula quantifies the performance evaluation. This implementation plan is otherwise identical to specific implementation plan three.

[0161] Specific Implementation Plan Five: (e.g.) Figure 5 As shown, S400 includes the following steps:

[0162] S410. Reservoir units are divided based on well depth and alkane gas characteristics:

[0163]

[0164] in For the i-th geological unit; This represents the well depth. The starting and ending depths of the unit, including one unit every 10m;

[0165] By defining a set of spatially discrete units, the continuous reservoir is divided into modelable nodes, which facilitates the construction of graph structures.

[0166] S420. Construct an undirected graph using geological units as nodes, based on spatial adjacency or feature similarity:

[0167]

[0168] The cosine similarity is in the range [-1, 1]. These are the eigenvectors of units i and j; It is the Euclidean norm;

[0169] By quantifying similarity, edges are connected if the similarity is greater than 0.8, generating an adjacency matrix A. A spatial dependency graph is then constructed to simulate reservoir connectivity.

[0170] S430. Map the fused features to graph node attributes:

[0171]

[0172] Let i be the initial feature representation of node i; A subset of the fusion features;

[0173] Multi-level representation extensions are supported through direct assignment embedding.

[0174] S440. Construct a GNN network model. The model is trained based on the Graph Convolutional Network (GCN) framework, enabling the model to learn spatial dependencies. The GNN network model introduces an adaptive attention mechanism (dynamically adjusting neighborhood weights), residual connections and skip aggregation (alleviating oversmoothing), deep physical constraint integration (ensuring consistency of geological patterns), and temporal dynamic fusion (capturing real-time data sequences) on the basis of GCN.

[0175] By constructing a graph structure with reservoir units as nodes, and training the GNN network model based on the graph structure data, the model learns spatial dependencies.

[0176] First, a time-series update is performed, using a gated recurrent unit (GRU) to fuse real-time sequence information:

[0177]

[0178] Then, an adaptive attention mechanism is introduced to calculate attention weights and enhance the normalized adjacency matrix:

[0179]

[0180]

[0181] Finally, graph convolution updates are performed, incorporating residual connections, skip aggregation, and layer normalization:

[0182]

[0183] in, Let the (l+1)th layer node representation matrix be used; For activation functions; This is the normalized adjacency matrix; Here is the learnable weight matrix for the l-th layer; This is the feature representation matrix of all nodes in the l-th layer; Let i be the feature vector of node i in the l-th layer; This is the intermediate feature matrix of the l-th layer after being updated by the temporal GRU; This represents the hidden state of layer l-1 of the previous time step (or the previous depth unit); Let i be the first-order neighborhood set of node i (spatial adjacency or cosine similarity > 0.8); The activation function for leakage rectification has a slope of 0.2. The attention weights of node j to i (Softmax normalized); This is the adjacency matrix after attention enhancement; For learnable attention vectors; LayerNorm is for layer normalization; The weights for skip connections are initialized from 0.1. This is the initial node feature matrix;

[0184] Node features are updated through neighborhood aggregation, using the AdamW optimizer with a learning rate of 0.005-0.01 and 4 layers. This captures spatial dependencies and improves modeling capabilities.

[0185] S450, Model Validation: Validate the model's convergence and modeling capability using known porosity data.

[0186]

[0187] To verify the loss function; Predict porosity for the model; This represents the true porosity. The number of samples;

[0188] Performance is evaluated using mean squared error, and an early stopping mechanism is introduced to avoid overfitting, ensuring model reliability and generalization ability.

[0189] This implementation plan is otherwise the same as Specific Implementation Plan Four.

[0190] Specific implementation plan six: S500 includes the following steps:

[0191] S510. Preliminary porosity prediction based on GNN:

[0192]

[0193] Let i be the initial predicted porosity of node i; For GNN functions; Input features; It is an adjacency matrix;

[0194] S520, limiting the porosity range ([0%, 40%]) and embedding the porosity-permeability relationship (based on the Kozeny-Carman equation):

[0195]

[0196] in For penetration rate; It is a proportionality constant; Porosity; The index is calculated based on sandstone m=3, adjusted according to lithology.

[0197] By introducing physical relationships, we ensure that the predictions conform to geological laws and avoid purely data-driven, non-physical results.

[0198] S530. Construct the loss function by combining MSE and physical constraint penalty term:

[0199]

[0200] in Total loss; To predict porosity; For true porosity, As weight, N is the number of samples; This represents the true penetration rate. The first term is MSE (prediction bias); the second term is the range penalty: the portion exceeding 0.4 due to soft constraints; the third term is the penetration rate penalty. By balancing data fitting with physical constraints, the model parameters are optimized to improve prediction accuracy and stability.

[0201] S540: The model's real-time performance and robustness were verified through high noise (20%-30%) and low sampling rate tests. Performance indicators (e.g., MSE < 0.02) were evaluated based on simulation experiments; the applicability of the method under real-world conditions was ensured by simulating drilling environments.

[0202] S550. Continuous porosity distribution maps are generated using moving average or spline interpolation:

[0203]

[0204] in This is the porosity value after smoothing. The width is half the width of the window; These are the original predicted values;

[0205] A continuous curve is generated through local averaging, and a visual report, including a 3D porosity map, is output. The prediction results are smoothed for easy engineering applications. This implementation scheme is otherwise identical to Implementation Scheme Five.

[0206] This invention combines wavelet transform multi-scale feature extraction, mutual information weighted feature fusion, GNN spatial modeling, and physical constraint optimization to achieve high-precision and robust porosity prediction.

[0207] The meanings or conditions of the parameters in the above implementation plan are as follows:

[0208] - Well depth (unit: meters): Represents the total drilling depth, used as a spatial coordinate.

[0209] - Drill bit depth (unit: meters): Indicates the current depth of the drill bit, used for dynamic data alignment.

[0210] - Alkane gas concentration: The concentration (unit: %) is collected in real time by a gas analyzer to reflect the reservoir gas content.

[0211] - Carbon isotope values: the carbon isotope ratios of methane, ethane, and propane ( (Compared to the VPDB standard), it reflects the source of oil and gas and reservoir characteristics.

[0212] - Peak integral area: The peak area of ​​alkane gases in chromatographic analysis, which is directly proportional to the gas content.

[0213] Definition: In the above text Each represents a data point at a given location.

[0214] The data comes from real-time logging and gas analysis during the drilling process, providing direct or indirect characteristics of reservoir porosity.

[0215] The real-time porosity prediction method (algorithm) based on the characteristics of alkane gas and its carbon isotopes proposed in this invention is the underlying technical core of this invention, and various products can be derived based on the algorithm.

[0216] Based on the method proposed in this invention, a real-time porosity prediction system based on alkane gas and its carbon isotope characteristics is developed using a programming language. This system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned real-time porosity prediction method based on alkane gas and its carbon isotope characteristics when running.

[0217] The developed system (software) computer program is stored on a computer-readable storage medium. This computer program is configured to, when invoked by a processor, implement the steps of the aforementioned method for real-time porosity prediction based on alkane gas and its carbon isotope characteristics. In other words, the invention is materialized on a carrier, becoming a computer program product.

[0218] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0219] The computational programs (also referred to as programs, software, software applications, or code) of this invention include machine instructions of a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0220] The method of the present invention will be described below with reference to specific embodiments.

[0221] Example 1

[0222] Drilling data from an onshore oilfield was selected, with well depths ranging from 1000 to 2000 meters. Real-time data acquisition included: methane concentration (1.2%-2.5%), ethane concentration (0.5%-1.0%), propane concentration (0.3%-0.8%), and carbon isotope values. The db4 wavelet basis was used, with a decomposition level J=4. After data preprocessing, the extracted feature vectors had a dimension of 32. The GNN model was a GCN, with a learning rate of 0.01, epoch=100, trained on historical core data (porosity 10%-25%). The prediction results are shown in Table 1. Figure 6-7 As shown: MSE < 0.02, Compared with core data, the error is <5%. Calculation time is <1 minute per depth point, suitable for real-time drilling platforms.

[0223] Table 1

[0224]

[0225] To verify the real-time decision-making efficiency of this invention, the decision-making efficiency of this invention is compared with that of traditional well logging methods (mainly sonic logging) and traditional machine learning methods (based on support vector machines, SVM). The experimental setup is as follows:

[0226] Data input: Methane concentration (1.2%-2.5%), ethane (0.5%-1.0%), propane (0.3%-0.8%), carbon isotope value ( ), sampling frequency 1 time / minute.

[0227] The method of this invention: adopts db4 wavelet basis decomposition (J=4), GCN model (learning rate 0.01, epoch=100), and runs on a real-time drilling platform with 16GB memory and 2.5GHz CPU.

[0228] Traditional logging methods: based on sonic logging, processing time is 5 minutes per depth point.

[0229] Traditional SVM methods: based on RBF kernels, training time is about 2 hours, prediction time is 2 minutes per depth point.

[0230] Evaluation metrics: Decision time (total time from data acquisition to porosity prediction generation), exploration efficiency (number of depth points processed per unit time, depth points / hour), prediction accuracy ( and MSE).

[0231] The experiment selected 50 depth points within the well depth range of 1000-2000m for testing, and the results are shown in Table 2.

[0232] Table 2

[0233]

[0234] Prediction accuracy analysis: The method of this invention MSE=0.018, which is superior to traditional well logging methods. (MSE=0.050) and SVM method ( (MSE=0.035), error <5%.

[0235] Decision time: The average decision time of the method of this invention is 0.8 minutes per depth point, which is 84% ​​and 60% shorter than that of the traditional logging method (5.0 minutes) and the SVM method (2.0 minutes), respectively.

[0236] Exploration efficiency: The method of this invention processes 75 depth points per hour, which is 525% and 150% higher than the traditional well logging method (12 depth points / hour) and the SVM method (30 depth points / hour), respectively. In terms of the integrated decision-making process (data acquisition, processing, and prediction), the real-time decision-making efficiency of this invention is improved by approximately 33.3%, verifying the conclusion that "real-time decision-making efficiency is improved by more than 30%".

[0237] Example 2

[0238] For offshore drilling scenarios, noise levels are increased by 20%. A variant of GAT is used, with physical constraints. , .like Figure 8 As shown, the results demonstrate that the model is robust and has high prediction accuracy. Suitable for complex geological conditions. Additional testing: Accuracy remains >80% in highly saline environments.

[0239] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for real-time prediction of porosity based on alkane gas and its carbon isotope characteristics, characterized in that: Includes the following steps: S100, real-time acquired alkane gas concentration and carbon isotope data, time-frequency decomposition of the data to obtain characteristic data at different time and frequency scales; S200: Perform multi-scale feature extraction on the feature data obtained in S100 to obtain multi-scale features; S300: The multi-scale features obtained from S200 are fused to obtain fused features; S400. Construct a GNN network model. The model is based on the Graph Convolutional Network (GCN) framework and introduces an adaptive attention mechanism, residual connections and skip aggregation, as well as temporal dynamic fusion. Construct a graph structure with reservoir units as nodes, map the fused features to graph node attributes, and train the GNN network model based on the graph structure data. S500 constructs the reservoir space structure based on the trained GNN network model, and combines physical constraint optimization to achieve real-time prediction of porosity.

2. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 1, characterized in that: S100 includes the following steps: S110: Real-time acquisition of well depth, drill bit depth, alkane gas concentration, carbon isotope value and peak integral area data, and processing of missing and outlier values ​​in the data; S120. Perform noise reduction and signal curve smoothing on the data; S130. Wavelet transform is performed using the Daubechies wavelet basis and the Mallat algorithm. S140, decompose low-frequency and high-frequency components to obtain multi-scale component data.

3. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 2, characterized in that: S200 includes the following steps: S210. Divide the data into low-frequency and high-frequency data; S220, Extracting abrupt changes in carbon isotope values ​​or gas concentrations from high-frequency components; S230. Extract the overall trend characteristics of gas concentration or porosity from low-frequency components; S240. Calculate data statistics to quantify the data.

4. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 3, characterized in that: S300 includes the following steps: S310. Concatenate feature vectors of different scales; S320. Evaluate the correlation between characteristics and porosity through mutual information; S330, Weight allocation based on mutual information; S340. Dimensionality reduction is performed on the data to obtain the final fused features.

5. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 4, characterized in that: S400 includes the following steps: S410. Divide reservoir units according to well depth and alkane gas characteristics; S420. Construct an undirected graph based on spatial adjacency or feature similarity, using reservoir units as nodes. S430, Map the fused features to graph node attributes; S440. Construct a GNN network model. The model is based on the Graph Convolutional Network (GCN) framework and introduces adaptive attention mechanism, residual connection and skip aggregation, as well as temporal dynamic fusion. The GNN network model is trained based on graph structure data, enabling the model to learn spatial dependencies.

6. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 5, characterized in that: The process by which the GNN network model in S440 learns spatial dependencies is as follows: First, a time-series update is performed, using a gated recurrent unit (GRU) to fuse real-time sequence information: Then, an adaptive attention mechanism is introduced to calculate attention weights and enhance the normalized adjacency matrix: Finally, graph convolution updates are performed, incorporating residual connections, skip aggregation, and layer normalization: in, Let the (l+1)th layer node representation matrix be used; For activation functions; This is the normalized adjacency matrix; Here is the learnable weight matrix for the l-th layer; This is the feature representation matrix of all nodes in the l-th layer; Let i be the feature vector of node i in the l-th layer; This is the intermediate feature matrix of the l-th layer after being updated by the temporal GRU; This refers to the (l-1)th hidden state of the previous time step or the previous depth unit. Let be the set of the first-order neighborhood of node i; The activation function for leakage rectification; The attention weights of node j to i (Softmax normalized); This is the adjacency matrix after attention enhancement; The initial node feature representation matrix; LayerNorm is the layer normalization; This is a learnable attention vector.

7. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 6, characterized in that: S500 includes the following steps: S510. Generate preliminary porosity prediction results based on graph neural networks; S520, limits the porosity range and embeds the porosity-permeability relationship; S530. Construct a loss function by combining prediction bias and physical constraint penalties, and optimize model parameters: in, Total loss; To predict porosity; For true porosity, Here, N represents the weights, and N represents the number of samples. This represents the true penetration rate. S540. Based on the optimized model, generate porosity results and smooth the results.

8. The real-time porosity prediction method based on alkane gas and its carbon isotope characteristics according to claim 7, characterized in that: The porosity-permeability relationship described in S520 is as follows: in For penetration rate; It is a proportionality constant; Porosity; For exponents.

9. A real-time porosity prediction system based on alkane gas and its carbon isotope characteristics, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 8, and executes the steps in the above-described real-time porosity prediction method based on alkane gas and its carbon isotope characteristics when it is run.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the real-time porosity prediction method based on alkane gas and its carbon isotope characteristics as described in any one of claims 1 to 8.

Images