Reservoir porosity real-time prediction method and system based on alkane gas curvature and storage medium

Abstract

The invention discloses a reservoir porosity real-time prediction method and system based on alkane gas curvature and a storage medium, relates to the technical field of oil and gas reservoir development, and aims to solve the problems of insufficient utilization of dynamic characteristics of alkane gas curvature, poor real-time performance and limited heterogeneous reservoir modeling capability of an existing method. According to the technical key points, the method comprises the following steps: S100, acquiring alkane gas concentration and carbon isotope ratio data in a drilling process in real time, and calculating a curvature characteristic of the alkane gas concentration; s200, performing dynamic feature extraction on the alkane gas curvature; s300, performing dynamic weighted feature fusion on the curvature dynamic feature and the carbon isotope ratio to obtain fusion data; s400, constructing a BiLSTM model, training the BiLSTM model based on the fusion data, generating a hidden state, introducing an attention mechanism to perform dynamic weighting, and combining with an adjacent matrix of a reservoir geological unit to realize space enhancement; and S500, on the basis of the BiLSTM model, in combination with physical constraints, performing real-time prediction on the porosity of the reservoir.

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 prediction of reservoir porosity based on alkane gas curvature. Background Technology

[0002] Reservoir porosity prediction is a key technology in oil and gas exploration and development, directly impacting reservoir evaluation and development efficiency. Traditional methods include empirical formulas (such as the Archie formula), geophysical logging (such as sonic and resistivity logging), and machine learning methods (such as random forests or support vector machines), but they have significant limitations:

[0003] 1. Empirical formulas: rely on a large amount of core experimental data, are difficult to capture reservoir heterogeneity, cannot update prediction results in real time, and are greatly affected by geological variability.

[0004] 2. Logging methods: Due to downhole noise interference, data acquisition is delayed (usually requiring several hours), making it difficult to use the curvature characteristics of alkane gas concentration changes to reflect reservoir micro-changes.

[0005] 3. Machine learning methods: Although they improve prediction accuracy, they do not make full use of the dynamic characteristics of alkane gas curvature, have poor real-time performance, and have limited ability to model heterogeneous reservoirs (e.g., ignoring time-series dependencies).

[0006] While existing technologies attempt to predict porosity using gas concentration, they fail to consider tortuosity as a core feature and do not integrate carbon isotope ratios and time-series models, resulting in low accuracy under complex geological conditions. To address these issues, this invention proposes a real-time reservoir porosity prediction method based on alkane gas tortuosity. This method acquires alkane gas concentrations during drilling in real time, calculates its tortuosity characteristics (based on the second derivative of concentration with well depth), combines it with carbon isotope ratios as an auxiliary feature, and utilizes time-varying feature extraction algorithms and BiLSTM to model the dynamic changes in tortuosity. Combined with reservoir spatial structure analysis, this achieves high-precision, real-time porosity prediction. This method innovatively introduces time-varying tortuosity analysis, overcoming the static limitations of traditional methods and providing dynamic and reliable technical support for oil and gas development. Summary of the Invention

[0007] The technical problem to be solved by this invention is:

[0008] Existing methods do not adequately utilize the dynamic characteristics of alkane gas curvature, have poor real-time performance, and have limited ability to model heterogeneous reservoirs.

[0009] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0010] This invention aims to provide a high-precision, real-time, and highly adaptable porosity prediction method for oil and gas reservoir development, thereby improving exploration efficiency and the reliability of development decisions. It provides a real-time reservoir porosity prediction method based on alkane gas curvature, comprising the following steps:

[0011] S100: Real-time acquisition of alkane gas concentration and carbon isotope ratio data during drilling process, and calculation of the curvature characteristics of alkane gas concentration.

[0012] S200: Dynamic feature extraction of alkane gas curvature;

[0013] S300. Dynamically weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio values ​​is performed to obtain fused data;

[0014] S400. Construct a BiLSTM model, train the BiLSTM model based on fused data, generate hidden states, introduce an attention mechanism to dynamically weight the hidden states at different time steps, and combine the weighted hidden states with the adjacency matrix of reservoir geological units to perform spatial augmentation.

[0015] S500, based on a trained BiLSTM model and combined with physical constraints, performs real-time prediction of reservoir porosity.

[0016] Furthermore, S100 includes the following steps:

[0017] S110: Real-time acquisition of well depth, drill bit depth, alkane gas concentration and carbon isotope ratio, and removal of outliers and missing values ​​in the data;

[0018] S120. Calculation of the curvature characteristics of alkane gas concentration based on the high-order finite difference method;

[0019] S130. The curvature data is smoothed by dynamic weighted moving average, and noise is removed by wavelet transform using an adaptive denoising threshold.

[0020] Furthermore, the removal of outliers and missing values ​​from the data described in S110 includes the following steps:

[0021] S111. Dynamic linear interpolation and timestamp-weighted interpolation are used to interpolate missing values, while cubic spline interpolation is used for consecutive missing values.

[0022] S112. Detecting outliers using local anomaly factors:

[0023]

[0024] in Data points Local anomaly factors, representing points The degree of anomaly relative to its neighborhood;

[0025] Data points of A set of nearest neighbors;

[0026] : point The locally reachable density is defined as:

[0027]

[0028] in It is a point arrive The reachable distance;

[0029] The number of points in the neighborhood;

[0030] Detection of outliers based on local anomaly factors;

[0031] S113. Standardize the data.

[0032] Furthermore, the dynamic feature extraction of alkane gas curvature described in S200 includes the following steps:

[0033] S210. Extract time-varying dynamic statistical features of curvature and isotope ratio, including: mean, standard deviation and slope;

[0034] S220. Extract the instantaneous features of curvature through Hilbert transform and enhance the curvature features;

[0035] S230, Quantify the statistics of curvature characteristics, including peak value and frequency, to extract key dynamic indicators.

[0036] Furthermore, the dynamic weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio described in S300 includes the following steps;

[0037] S310. Integrate the curvature and isotope ratio characteristics to obtain comprehensive characteristics;

[0038] S320. Evaluate the dynamic correlation between comprehensive characteristics and porosity;

[0039] S330. Based on dynamic correlation, the comprehensive features are weighted and fused, and the features are dimensionality reduced to obtain the fused features.

[0040] Furthermore, S400 includes the following steps:

[0041] S410. Sequence the fused features into time-series data;

[0042] S420. Construct a bidirectional long short-term memory (BiLSTM) network model, train the BiLSTM model based on time-series data, and generate forward and backward hidden states.

[0043] S430. Introduce an attention mechanism to dynamically weight the hidden states at different time steps to generate weighted hidden states;

[0044] S440. The weighted hidden state is combined with the adjacency matrix of the reservoir geological unit through a graph attention network (GAT) to generate a spatially enhanced state.

[0045] S450. To verify the convergence and temporal modeling capability of the BiLSTM model, the Adam optimizer and early stopping mechanism were used. The initial learning rate was 0.001, and the number of hidden layer units was optimized through grid search with candidate values ​​of [64, 128, 256] until the model converged, thus obtaining the trained BiLSTM model.

[0046] Furthermore, the S500 includes the following steps:

[0047] S510. Based on the spatial enhancement state output by GAT, preliminary porosity prediction results are generated through the fully connected layer.

[0048] S520, limits the range of porosity and incorporates the tortuosity-porosity relationship:

[0049] ;

[0050] in Porosity at well depth z;

[0051] Curvature characteristics;

[0052] Linear fitting coefficients are determined using the least squares method.

[0053] S530. Optimize model parameters using the loss function:

[0054]

[0055] in Total loss function;

[0056] Predicted and actual porosity;

[0057] Penalty item weight;

[0058] S540. The real-time performance and robustness of the optimized model obtained from S530 are tested in a high-noise environment.

[0059] S550. Smooth the prediction results generated by the model to obtain the real-time prediction results of reservoir porosity.

[0060] This invention provides a real-time reservoir porosity prediction system based on alkane gas curvature. The system has a program module corresponding to the steps of the method described in one of the above technical solutions, and executes the steps in the above-described real-time reservoir porosity prediction method based on alkane gas curvature when running.

[0061] The present invention provides a computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps in the method for real-time prediction of reservoir porosity based on alkane gas curvature as described in any of the above technical solutions.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] 1. Real-time dynamic prediction: By collecting alkane gas concentration in real time and calculating tortuosity characteristics, dynamic prediction of porosity during drilling can be achieved. It supports updates every minute and has strong real-time performance.

[0064] 2. High-precision prediction: By extracting the temporal features of curvature and combining them with carbon isotope ratios, the heterogeneity and complex geological features of the reservoir are captured, with a target accuracy of >95% and a prediction accuracy improvement of more than 20%.

[0065] 3. Strong adaptability: BiLSTM models the temporal characteristics of curvature, introduces an attention mechanism to dynamically weight the hidden state, focuses on the key temporal information of curvature features, and fuses the weighted hidden state with the adjacency matrix of reservoir geological units to achieve spatial enhancement. It innovatively integrates spatial-temporal features, which significantly improves the model's adaptability to heterogeneous reservoirs and enables the model to adapt to dynamic changes under complex geological conditions (such as fracture or bedding changes).

[0066] 4. Overcoming traditional limitations: Effectively utilizing curvature features to address the shortcomings of traditional methods in real-time and heterogeneous modeling.

[0067] 5. High-efficiency computing: Through dynamic feature fusion and dimensionality reduction techniques, computational complexity is reduced, supporting edge computing devices.

[0068] 6. High practicality: Generates smooth and continuous porosity distribution maps, providing reliable decision support for oil and gas exploration, with potential economic benefits including a 20% reduction in ineffective drilling. Attached Figure Description

[0069] Figure 1 This is a flowchart illustrating the real-time porosity prediction process based on alkane gas tortuosity in an embodiment of the present invention.

[0070] Figure 2 This is a flowchart of the data preprocessing process in an embodiment of the present invention;

[0071] Figure 3 This is a flowchart of the curvature feature extraction process in an embodiment of the present invention;

[0072] Figure 4 This is a flowchart of the feature fusion and optimization process in an embodiment of the present invention;

[0073] Figure 5 This is a flowchart of the BiLSTM modeling process in an embodiment of the present invention;

[0074] Figure 6 This is a comparison chart of the prediction results and core data results in the embodiments of the present invention. Detailed Implementation

[0075] 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.

[0076] 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.

[0077] Specific Implementation Plan 1: (e.g.) Figure 1 As shown, this invention provides a method for real-time prediction of reservoir porosity based on alkane gas tortuosity, comprising the following steps:

[0078] S100: Real-time acquisition of alkane gas concentration and carbon isotope ratio data during drilling process, and calculation of the curvature characteristics of alkane gas concentration.

[0079] S200: Dynamic feature extraction of alkane gas curvature;

[0080] S300. Dynamically weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio values ​​is performed to obtain fused data;

[0081] S400. Construct a BiLSTM model, train the BiLSTM model based on fused data, generate hidden states, introduce an attention mechanism to dynamically weight the hidden states at different time steps, and combine the weighted hidden states with the adjacency matrix of reservoir geological units to perform spatial augmentation.

[0082] S500, based on a trained BiLSTM model and combined with physical constraints, performs real-time prediction of reservoir porosity.

[0083] Specific Implementation Plan Two: (e.g.) Figure 2 As shown, S100 includes the following steps:

[0084] S110. Real-time acquisition of alkane gas concentration and carbon isotope ratio data during the drilling process, removing outliers and missing values; including the following steps:

[0085] S111. Data Collection and Cleaning: Real-time acquisition of well depth, drill bit depth, alkane gas concentration, and carbon isotope ratio; removal of outliers and missing values ​​to avoid noise interference with subsequent analysis.

[0086] Missing value handling: Dynamic linear interpolation is used, weighted based on timestamps.

[0087]

[0088] in : The interpolation result at the i-th time point, representing the missing data point value (including alkane gas concentration or carbon isotope ratio).

[0089] : Timestamps, representing the time of the current point, the (i+1)th point, and the (i-1)th point, respectively, used for time alignment of real-time data acquisition.

[0090] : Known data values ​​of neighboring points before and after.

[0091] By using time-weighted interpolation, missing values ​​are estimated based on the distance ratio between consecutive time points, ensuring data continuity.

[0092] For consecutive missing values, use cubic spline interpolation: The coefficients a, b, c, and d are solved by boundary conditions to ensure the continuity of the second derivative.

[0093] S112. Outlier Detection: Local outlier (LOF) detection is used.

[0094]

[0095] in Data points Local anomaly factors, representing points The degree of anomaly relative to its neighborhood.

[0096] Data points of A set of nearest neighbors (based on Euclidean distance).

[0097] : point j / The locally reachable density is defined as ,in It is a point arrive The reachable distance.

[0098] : The number of points in the neighborhood.

[0099] Detect outliers (such as sudden changes in gas concentration) using LOF. Values ​​that are considered outliers are replaced with the local median. The threshold can be adjusted according to the noise level of the data.

[0100] S113. Data Standardization: Eliminating Differences in Dimensions. , where μ and σ are the mean and standard deviation of the dynamic window.

[0101] S120, such as Figure 2 As shown, the curvature characteristics of alkane gas concentration are calculated based on a high-order finite difference method to capture the dynamic changes in reservoir heterogeneity, including the following steps:

[0102] Based on the calculation of curvature characteristics using alkane gas concentration, a higher-order finite difference method is used to improve accuracy. The first-order derivative approximation formula is as follows:

[0103]

[0104] in Gas concentration As well depth The first derivative represents the rate of change of concentration (unit: ).

[0105] : Nearest well depth (unit: meters).

[0106] : Well depth sampling interval (unit: meters, taken as 0.1-1 meters).

[0107] Using the higher-order finite difference method (accuracy) The rate of change of concentration with well depth is calculated to provide a basis for the calculation of curvature.

[0108] Second derivative approximation formula:

[0109]

[0110] in gas concentration The second derivative with well depth represents the acceleration of concentration change (unit: % / ).

[0111] The second derivative is a core component reflecting the tortuosity characteristics and the nonlinear variation of gas concentration with well depth, used to capture the microscopic characteristics of reservoir porosity.

[0112] S130. The curvature data obtained in S120 is smoothed by dynamic weighted moving average, and noise is removed by wavelet transform using an adaptive denoising threshold to enhance the stability of the curvature features. This includes the following steps:

[0113] Dynamic weighted moving average:

[0114]

[0115] Smoothed data values ​​(including gas concentration, unit: %).

[0116] : Original data value.

[0117] Weighted coefficients, dimensionless, based on Gaussian kernel function.

[0118] Data point index difference, representing temporal or spatial distance.

[0119] Adaptive bandwidth (unit: index or meter), determined based on signal variance.

[0120] Data is smoothed using weighted averaging to reduce noise and preserve signal trends. Ensure that values ​​closer to the current point have a greater impact.

[0121] Adaptive denoising threshold: Noise is removed using wavelet transform (Daubechies wavelet).

[0122] in:

[0123] : Noise reduction threshold.

[0124] Standard deviation of the signal (in %).

[0125] Total number of data points.

[0126] Denoising is achieved through wavelet transform (Daubechies wavelet) and thresholding. It adaptively adjusts based on the signal noise level, retains key features, and removes high-frequency noise.

[0127] S140 verifies the data quality obtained in S130 by checking data stationarity and correlation with core data to ensure the validity of the features, including the following steps:

[0128] The stationarity of the data was verified using the KPSS test (p-value > 0.05 is considered stationary), and the correlation with the core data was verified using the Kendall correlation coefficient.

[0129]

[0130] in Kendall correlation coefficient, dimensionless, range [-1, 1], indicates the consistency between the ordering of data (including gas concentration) and core data.

[0131] Number of data points.

[0132] : Gas concentration or other characteristic value.

[0133] Core porosity value (unit: %).

[0134] : A sign function that returns 1, -1, or 0.

[0135] Data quality was verified by calculating the correlation between the processed data and the core data. This indicates insufficient data quality, requiring re-collection. All other aspects of this implementation plan are the same as in Specific Implementation Plan One.

[0136] Specific Implementation Plan Three: (e.g.) Figure 3 As shown, the dynamic feature extraction of alkane gas curvature described in S200 includes the following steps:

[0137] S210. Extract time-varying statistical features of tortuosity and isotope ratios to capture reservoir dynamics and provide rich information for model input. Use tortuosity data to generate statistical features.

[0138] Extracting curvature κ(z) and carbon isotope ratios Time-varying statistical characteristics (including: mean, standard deviation, slope): , , , where W is the sliding window.

[0139] in Time window The mean within (including the average of gas concentration or curvature, in % or) ).

[0140] The standard deviation within the time window reflects the volatility of the data.

[0141] The slope within the time window represents the rate of change of the feature over time (unit: % / second or...). / Second).

[0142] : Sliding window size (unit: number of data points, default 20).

[0143] Characteristic values ​​(such as curvature) Or isotope ratio, unit: or‰).

[0144] Time interval (unit: seconds).

[0145] Reservoir heterogeneity can be captured by extracting dynamic statistical features of curvature and isotope ratios.

[0146] S220. Extract the instantaneous features of curvature through Hilbert transform to enhance the dynamic expressive ability of time-series signals. Enhance the curvature features to generate instantaneous features.

[0147] Extracting instantaneous features using Hilbert transform: (Approximate calculation using Fast Fourier Transform).

[0148] in Curvature The result of the Hilbert transform, in complex form, includes instantaneous amplitude and phase.

[0149] Original curvature characteristics (unit: ).

[0150] Well depth variable (unit: meters).

[0151] :time.

[0152] Imaginary unit.

[0153] The instantaneous features of curvature (such as instantaneous frequency) are extracted by Hilbert transform to enhance the dynamic characteristics of time-varying signals, and the calculation is approximated by Fast Fourier Transform (FFT).

[0154] S230 quantifies the peak value and frequency of curvature features, extracts key dynamic indicators, and improves the model's prediction accuracy. The instantaneous features of S220 are quantified into statistical quantities.

[0155] Calculate the statistics of dynamic characteristics: , .

[0156] The maximum value of a feature within a time window (e.g., peak curvature, unit: ).

[0157] Instantaneous frequency (unit: Phase from Hilbert transform extract.

[0158] The feature sequence within the window.

[0159] By quantifying the dynamic characteristics of curvature, the peak value reflects drastic changes in the reservoir, while the frequency reflects periodic changes.

[0160] S240. Ensure the extracted features are highly correlated with porosity and optimize the effectiveness of the feature set. Verify the features from S230; if successful, input them into S300 for fusion; otherwise, adjust the window size and re-extract.

[0161] Feature validity is ensured through Kendall correlation coefficient and time-series cross-validation (K=5 fold). If cross-validation... <0.8, optimized window size.

[0162] Specific Implementation Plan Four: (e.g.) Figure 4 As shown, the dynamic weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio described in S300 includes the following steps;

[0163] S310: Integrate the curvature and isotope ratio features to construct a comprehensive feature set, enhancing the diversity of model inputs. Concatenate the features from S230 into a vector and input it into S320 for correlation analysis.

[0164]

[0165] S320. Evaluate the dynamic correlation between characteristics and porosity using time-varying correlation coefficients:

[0166]

[0167] in Pearson correlation coefficient within the time window, dimensionless, range .

[0168] Curvature characteristics or isotopic ratios (unit: or‰).

[0169] Porosity or other reference values ​​(unit: %).

[0170] : Inside the window The mean.

[0171] By evaluating the dynamic correlation between curvature characteristics and porosity, a weighting basis is provided for feature fusion.

[0172] S330. Optimize the feature set based on dynamic correlation weighted fusion features to improve prediction accuracy:

[0173] The weight w is based on Normalization.

[0174] in The fused feature vector is dimensionless (after standardization).

[0175] Curvature eigenvector (unit: (Dimensionless after standardization).

[0176] Carbon isotope ratio eigenvector (unit: (Dimensionless after standardization).

[0177] Time-varying weights, dimensionless, based on Normalization .

[0178] The feature set is optimized by weighting and fusing curvature and isotopic features according to their correlation.

[0179] S340 reduces the dimensionality of the fusion features of S330, thereby reducing computational complexity and retaining key information.

[0180] Through time-varying principal component analysis (T-PCA): ,in The eigenvectors of the dynamic covariance matrix (retaining 95% variance).

[0181] in : The dimensionless eigenvectors after dimensionality reduction.

[0182] Original feature matrix (containing curvature and isotopic features).

[0183] The eigenvector matrix of the dynamic covariance matrix, retaining 95% of the variance.

[0184] T-PCA reduces feature dimensionality, decreases computational complexity, and preserves key information.

[0185] S350. Evaluate the predictive performance of the fused features to ensure that the feature set is suitable for real-time prediction.

[0186] Through time-varying MSE and : , .like Iteratively optimize the weights.

[0187] in Mean squared error (unit: ), which measures the deviation between predicted porosity and the actual value.

[0188] : Coefficient of determination, dimensionless, range [0, 1], representing the proportion of variance explained by the model.

[0189] Predicted porosity (unit: %).

[0190] True porosity (unit: %).

[0191] : The average value of the actual porosity within the window.

[0192] pass To evaluate the feature fusion effect, if The weights are optimized and the components are re-integrated. Otherwise, this implementation plan is the same as specific implementation plan three.

[0193] Specific implementation plan five: S400 includes the following steps:

[0194] S410. Serialize the dimensionality reduction and fusion features of S340 into time-series data:

[0195]

[0196] in The temporal feature matrix input to BiLSTM contains fused features with a sliding window size of W.

[0197] : The fused feature vector at time step t.

[0198] The fused features are serialized into temporal inputs for BiLSTM modeling.

[0199] S420. Construct a Bidirectional Long Short-Term Memory (BiLSTM) network model. Train the BiLSTM model using the time-series data from S410 to generate hidden states: Using the Adam optimizer with a learning rate of 0.001, GPU acceleration is supported to ensure real-time training (batch size 32).

[0200] in The hidden state vector at the current time step t (fusing forward and backward information).

[0201] Input features at the current time step.

[0202] The hidden state and memory unit of the previous time step.

[0203] BiLSTM models temporal dependencies through forget gates, input gates, and output gates, while capturing both forward and backward information to optimize the dynamic relationship between curvature and porosity.

[0204] S430. An attention mechanism is introduced after the hidden state to dynamically weight the hidden states at different time steps:

[0205]

[0206] Q, K, and V are generated from the hidden state h_t through a linear transformation;

[0207] : The dimension of the query vector, used to scale the dot product to prevent gradient vanishing;

[0208] Attention weight distribution ;

[0209] Obtain the weighted hidden state:

[0210]

[0211] By focusing on key temporal features (such as curvature peaks) through an attention mechanism, weighted hidden states are generated. This improves the model's robustness to noise. By focusing on key time steps (such as curvature peaks, abrupt changes, or abnormal fluctuations), noise interference is suppressed, enhancing the model's sensitivity to reservoir micro-changes and prediction accuracy.

[0212] S440. The weighted hidden state obtained in S430 is combined with the adjacency matrix using GAT to generate a spatially enhanced state, which enhances the model's adaptability to geological structures.

[0213]

[0214] Where A is the adjacency matrix of reservoir geological units (constructed based on well distance and geological connectivity, dimensionless);

[0215] Node i enhances its hidden state in space at time t;

[0216] The robustness of predictions is enhanced by incorporating reservoir spatial relationships (such as geological unit connectivity) into the model.

[0217] S450 verifies the convergence and temporal modeling capability of the BiLSTM model obtained in S430 to ensure prediction reliability. The model's convergence (using the Adam optimizer, early stopping mechanism, patience = 10 epochs, initial learning rate of 0.001, and the number of hidden layer units optimized through grid search with candidate values ​​of [64, 128, 256]) and temporal modeling capability (using the BiLSTM forget gate to handle long dependencies) are verified using real-time porosity data. Other aspects of this implementation scheme are the same as in specific implementation scheme four.

[0218] Specific Implementation Plan Six: (e.g.) Figure 5 As shown, S500 includes the following steps:

[0219] Preliminary porosity prediction results are generated based on the enhanced state of S510 and S440:

[0220]

[0221] in : time Predicted porosity (unit: %).

[0222] Fully connected layer function, mapping BiLSTM output to porosity.

[0223] Input features and hidden states.

[0224] S520. Physical constraints are imposed on the predicted values ​​of S510 to ensure that the predictions conform to geological laws and porosity ranges.

[0225] Limit the porosity range [0%, 40%] and embed the tortuosity relationship:

[0226] in Porosity at well depth z (unit: %).

[0227] The linear fitting coefficients are determined using the least squares method.

[0228] By embedding the physical relationship between curvature and porosity, we ensure that the predictions conform to geological laws.

[0229] S530. Optimize the model using the loss function:

[0230]

[0231] Total loss function (unit: ).

[0232] Predicted and actual porosity (in %).

[0233] Penalty weight, dimensionless, default 0.1.

[0234] : Penalize items with porosity exceeding 40%.

[0235] : Penalize terms that deviate from the curvature relationship.

[0236] By optimizing the model, we ensure that the predictions conform to physical constraints (porosity range [0%, 40%]) and tortuosity relationships.

[0237] The S540 and S530 optimized models were tested for real-time performance and robustness in high-noise environments to ensure feasibility for field applications.

[0238] Simulate high noise: σ=0.05. The model is updated using online learning.

[0239] in Data with noise added (including gas concentration, unit: %).

[0240] : Raw data.

[0241] The mean is 0 and the variance is 0. Normally distributed noise ( ).

[0242] The robustness of the model was tested by simulating a high-noise downhole environment.

[0243] The prediction results of S550 and S540 are smoothed to generate a continuous porosity distribution, which facilitates geological interpretation and decision support.

[0244] .

[0245] in Porosity after smoothing (unit: %).

[0246] Original predicted porosity.

[0247] Weighted coefficients, dimensionless, based on Gaussian kernels.

[0248] Bandwidth parameter controls the smoothness.

[0249] By generating a continuous and smooth porosity distribution, prediction fluctuations are reduced, facilitating geological interpretation.

[0250] This implementation plan is otherwise the same as Implementation Plan Five.

[0251] The meanings or conditions of the parameters in the above implementation scheme are as follows:

[0252] - Well depth (unit: meters): Represents the total drilling depth, used as a spatial coordinate.

[0253] - Drill bit depth (unit: meters): Indicates the current depth of the drill bit, used for dynamic data alignment.

[0254] - Alkane gas concentration: Methane ( ), ethane ( ), propane ( Concentration of ) (unit: The gas content in the reservoir is collected in real time by a gas analyzer (such as a mud gas recorder).

[0255] - Carbon isotope ratios: Carbon isotope ratios of methane, ethane, and propane ( ,unit: As an auxiliary feature, it reflects the source of oil and gas and reservoir characteristics.

[0256] - Curvature characteristics: Based on the variation of alkane gas concentration with well depth, the curvature characteristics are calculated.

[0257]

[0258] in, Let z be the alkane gas concentration and z be the well depth. This refers to the curvature of the gas.

[0259] The data comes from real-time logging and gas analysis during the drilling process, providing direct or indirect characteristics of reservoir porosity.

[0260] In the above text Both refer to the input features at the given location.

[0261] The real-time prediction method (algorithm) of reservoir porosity based on alkane gas curvature proposed in this invention is the underlying technical core of this invention, and various products can be derived based on the algorithm.

[0262] Based on the method proposed in this invention, a real-time reservoir porosity prediction system based on alkane gas curvature 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 reservoir porosity prediction method based on alkane gas curvature when running.

[0263] The developed system (software) computer program is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-described method for real-time prediction of reservoir porosity based on alkane gas tortuosity when called by a processor. In other words, the invention is materialized on a carrier, becoming a computer program product.

[0264] 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.

[0265] 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.

[0266] The present invention will be described below with reference to specific embodiments.

[0267] Example 1

[0268] Data was collected at a drilling site in an oilfield (well depth 2000m, heterogeneous carbonate reservoir) using a mud gas recorder. The experimental steps and results are as follows:

[0269] 1. Data Acquisition and Preprocessing

[0270] Methane was collected in real time using a mud gas recorder. Concentration sequences (examples: 1.2%, 1.5%, 1.3%), ethane ( Concentration (examples: 0.3%, 0.4%, 0.35%) and carbon isotope ratios ( (Examples: -45‰, -44.5‰, -45.2‰), sampling interval 0.1 meters, simultaneously recording well depth and drill bit depth data.

[0271] Clean the data and handle missing values ​​(e.g.) Concentration missing points) and outliers (e.g.) (Concentration abruptly changed to 2.5%), and data integrity was ensured through interpolation and detection of local outliers.

[0272] calculate Concentration curvature characteristics (example: And a stable sequence is obtained through smoothing and denoising techniques (wavelet transform) (example: (This is used to capture the dynamic changes in reservoir heterogeneity).

[0273] 2. Dynamic Feature Extraction

[0274] Time-varying statistical features of curvature and carbon isotope ratios were extracted within a sliding window (window size 20 data points) to obtain... The mean curvature was 1.33%, and the standard deviation was 0.12%. Mean -44.8‰, standard deviation 0.3‰.

[0275] Instantaneous features of curvature (such as instantaneous frequency 0.05 Hz, peak value) are extracted using a transformation method. This enhances the ability to characterize dynamic changes in reservoirs.

[0276] The correlation between verification characteristics and core porosity reached 0.85, indicating that tortuosity and carbon isotope ratio can effectively reflect reservoir heterogeneity and complex geological features (such as fractures or bedding variations).

[0277] 3. Feature Fusion and BiLSTM Modeling

[0278] By integrating the characteristics of curvature and carbon isotope ratio, and assigning weights through dynamic correlation analysis (curvature weight 0.6, carbon isotope ratio weight 0.4), a comprehensive feature set is formed.

[0279] The comprehensive features are dimensionality reduced, retaining 95% of the information, and then serialized into time-series data.

[0280] A BiLSTM model (128 hidden units, learning rate 0.001, batch size 32) was constructed and trained based on the input window (size 30). Spatial augmentation was performed using GAT by combining the adjacency matrix of reservoir geological units (based on well distance) to ensure that the model captures the spatial structure of the reservoir and was trained until convergence.

[0281] 4. Porosity Prediction and Optimization

[0282] Generate a preliminary porosity prediction (example: 15.2%) and embed physical constraints to limit porosity to between 0% and 40%, and correct it by incorporating the linear relationship between tortuosity and porosity.

[0283] After optimizing the model parameters, the final mean squared error (MSE) reached 0.005, indicating that the prediction results are highly stable.

[0284] Tested in a high-noise environment (noise level 10%), the model still maintains high accuracy after adding simulated noise, and the prediction results are smoothed to generate a continuous porosity distribution curve.

[0285] 5. High-precision prediction verification:

[0286] On a test dataset (1000 sample points) at a well depth of 2000 meters, the predicted porosity was compared with the core data (example values: 15.0%, 15.3%, 14.9%), with a mean absolute error (MAE) of 0.15% and a coefficient of determination ( ). The accuracy was 0.96, meeting the target accuracy of >95%. Especially in the fractured zone (well depth 1500-1600 meters), the porosity mutation (from 14% to 18%) was successfully captured, verifying the sensitivity of tortuosity and carbon isotope ratio to heterogeneous reservoirs.

[0287] Compared with existing methods:

[0288] Traditional empirical formulas (such as Archie's formula): The value was 0.78, and the MAE was 1.2%. Due to the lack of utilization of time series characteristics, it was greatly affected by reservoir heterogeneity.

[0289] Random Forest Method: The accuracy is 0.80, and the MAE is 1.0%, which is limited due to the lack of dynamic modeling and physical constraints.

[0290] Method of the present invention: The accuracy was 0.96, the MAE was 0.15%, and the accuracy was improved by about 20% (relative error reduction: (0.80-0.96) / (1-0.80)=80%), which is attributed to the synergistic effect of curvature temporal feature extraction and carbon isotope ratio fusion.

[0291] At a 10% noise level, the model The accuracy still reached 0.92, demonstrating the robustness of the high-precision prediction.

[0292] 6. High practicality verification

[0293] By smoothing the data, a continuous porosity distribution map is generated (example: at a well depth of 1000-2000 meters, porosity smoothly transitions from 12% to 18%), reducing noise-induced fluctuations (fluctuation amplitude is reduced from 0.5% to 0.1%). The distribution map clearly identifies high-porosity areas (e.g., 1500-1600 meters, porosity approximately 18%) and low-porosity areas (e.g., 1800-1900 meters, porosity below 10%), facilitating geological interpretation and drilling decisions.

[0294] In the experiment, the drilling path was adjusted three times based on the real-time predicted porosity distribution map, successfully avoiding two low-porosity zones (porosity <10%) and improving the drilling hit rate. Compared with traditional logging methods (data lag of several hours, requiring multiple core verifications), the method of this invention updates the prediction every minute, with a single prediction taking less than 1 second, significantly improving decision-making efficiency.

[0295] Economic benefit analysis: Traditional methods generated 10 invalid wells (in low porosity areas or non-reservoir areas) during the experiment, while the method of this invention reduced this to 8, reducing the invalid well rate by 20% (2 / 10).

[0296] Porosity distribution maps are output to geological modeling software (such as Petrel) via API interface to generate 3D reservoir models, supporting reservoir connectivity analysis and subsequent development planning, and further enhancing the reliability of exploration decisions.

[0297] Data privacy protection: AES-256 encryption is used to transmit methane concentration, ethane concentration and carbon isotope ratio data to ensure the security of well site data.

[0298] Hardware compatibility: The model is deployed on edge computing devices, supports GPU acceleration and API interfaces, and integrates seamlessly with existing logging systems, with real-time prediction taking less than 1 second per sample.

[0299] Experimental repeatability: Through 5-fold time-series cross-validation, the average... The accuracy reached 0.95, verifying the prediction accuracy and the stability of the porosity distribution map.

[0300] 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 reservoir porosity based on alkane gas tortuosity, characterized in that: Includes the following steps: S100: Real-time acquisition of alkane gas concentration and carbon isotope ratio data during drilling process, and calculation of the curvature characteristics of alkane gas concentration. S200: Dynamic feature extraction of alkane gas curvature; S300. Dynamically weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio values ​​is performed to obtain fused data; S400. Construct a BiLSTM model, train the BiLSTM model based on fused data, generate hidden states, introduce an attention mechanism to dynamically weight the hidden states at different time steps, and combine the weighted hidden states with the adjacency matrix of reservoir geological units to perform spatial augmentation. S500, based on a trained BiLSTM model and combined with physical constraints, performs real-time prediction of reservoir porosity.

2. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity 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 and carbon isotope ratio, and removal of outliers and missing values ​​in the data; S120. Calculation of the curvature characteristics of alkane gas concentration based on the high-order finite difference method; S130. The curvature data is smoothed by dynamic weighted moving average, and noise is removed by wavelet transform using an adaptive denoising threshold.

3. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity according to claim 2, characterized in that: The removal of outliers and missing values ​​from the data described in S110 includes the following steps: S111. Dynamic linear interpolation and timestamp-weighted interpolation are used to interpolate missing values, while cubic spline interpolation is used for consecutive missing values. S112. Detecting outliers using local anomaly factors: in Data points Local anomaly factors, representing points The degree of anomaly relative to its neighborhood; Data points of A set of nearest neighbors; : point The locally reachable density is defined as: in It is a point arrive The reachable distance; The number of points in the neighborhood; Detection of outliers based on local anomaly factors; S113. Standardize the data.

4. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity according to claim 3, characterized in that: The dynamic feature extraction of alkane gas curvature described in S200 includes the following steps: S210. Extract time-varying dynamic statistical features of curvature and isotope ratio, including: mean, standard deviation and slope; S220. Extract the instantaneous features of curvature through Hilbert transform and enhance the curvature features; S230, Quantify the statistics of curvature characteristics, including peak value and frequency, to extract key dynamic indicators.

5. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity according to claim 4, characterized in that: The dynamic weighted feature fusion of curvature dynamic characteristics and carbon isotope ratio as described in S300 includes the following steps; S310. Integrate the curvature and isotope ratio characteristics to obtain comprehensive characteristics; S320. Evaluate the dynamic correlation between comprehensive characteristics and porosity; S330. Based on dynamic correlation, the comprehensive features are weighted and fused, and the features are dimensionality reduced to obtain the fused features.

6. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity according to claim 5, characterized in that: S400 includes the following steps: S410. Sequence the fused features into time-series data; S420. Construct a BiLSTM model, train the BiLSTM model based on time series data, and generate forward and backward hidden states; S430. An attention mechanism is introduced to dynamically weight the hidden states at different time steps, generating weighted hidden states. S440, the weighted hidden state and the adjacency matrix of the reservoir geological unit are combined through a graph attention network (GAT) to generate a spatially enhanced state; S450. Verify the convergence and temporal modeling capability of the BiLSTM model until the model converges, and obtain the trained BiLSTM model.

7. The method for real-time prediction of reservoir porosity based on alkane gas tortuosity according to claim 6, characterized in that: S500 includes the following steps: S510. Based on the spatial enhancement state output by GAT, preliminary porosity prediction results are generated through the fully connected layer. S520, limits the range of porosity and incorporates the tortuosity-porosity relationship: ； in Porosity at well depth z; Curvature characteristics; Linear fitting coefficients are determined using the least squares method. S530. Optimize model parameters using the loss function: in Total loss function; Predicted and actual porosity; Penalty item weight; S540. The real-time performance and robustness of the optimized model obtained from S530 are tested in a high-noise environment. S550. Smooth the prediction results generated by the model to obtain the real-time prediction results of reservoir porosity.

8. A real-time reservoir porosity prediction system based on alkane gas tortuosity, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 7, and executes the steps in the above-described method for real-time prediction of reservoir porosity based on alkane gas curvature when it is run.

9. 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 in the real-time reservoir porosity prediction method based on alkane gas curvature as described in any one of claims 1 to 7.

Images