Method and system for constructing reservoir evaluation model based on deep learning

Abstract

The invention relates to the technical field of reservoir evaluation, in particular to a deep learning-based reservoir evaluation model construction method and system. The method comprises the following steps: acquiring logging data and core analysis data of a target work area, and respectively carrying out abnormal value processing and missing value filling to obtain complete logging data and complete core data; calculating reservoir pore structure parameters based on the complete core data; and inverting the macroscopic porosity of the reservoir based on the complete logging data. According to the method, the accuracy, the stability and the robustness of reservoir evaluation are remarkably improved by fusing the core data and the logging data and combining the pore structure parameters, the macroscopic porosity and the geological attention mechanism, the problems of data missing and small samples are solved, and the interpretability and the geological consistency of the model are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir evaluation, and in particular, to a method and system for constructing a reservoir evaluation model based on deep learning. Background Art

[0002] In the process of oil and gas exploration and development, reservoir evaluation is an important means to identify and quantify oil and gas reservoirs, and its accuracy is directly related to resource utilization efficiency and development benefits. Traditional reservoir evaluation technologies have gradually evolved from core experiments, empirical formula methods for well logging interpretation, to statistical regression and machine learning models. Early methods such as Archie's formula, Wyllie time-average equation, etc., were based on simplified physical assumptions and were applicable to specific lithologies, but had poor adaptability under complex geological conditions. With the development of data mining technologies, methods such as support vector machine (SVM), random forest (RF), etc. have been applied to reservoir evaluation, improving the processing ability for multi-variables. However, these methods generally ignore the depth sequence characteristics and geological context information of well logging data in the modeling process, limiting the prediction accuracy and generalization ability of the models.

[0003] One-dimensional convolutional neural network (1D-CNN) can effectively capture local patterns in well logging curves, but it is still difficult to incorporate geological prior knowledge. Introducing an attention mechanism can enhance the model's attention to key features. However, existing attention mechanisms are mostly data-driven, lacking geological knowledge constraints, and prone to attention deviation; at the same time, the model training data for small-sample reservoirs is insufficient, affecting prediction stability. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for constructing a reservoir evaluation model based on deep learning to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for constructing a reservoir evaluation model based on deep learning includes the following steps: Step S1: Obtain well logging data and core analysis data of the target work area, and perform outlier processing and missing value filling respectively to obtain complete well logging data and complete core data; Step S2: Calculate reservoir pore structure parameters based on the complete core data; invert the reservoir macroscopic porosity based on the complete well logging data; Step S3: Analyze the physical properties of reservoir lithofacies based on the reservoir pore structure parameters and the reservoir macroscopic porosity, determine the pore connectivity index, reservoir quality factor, lithology control coefficient, and heterogeneity index, and generate a reservoir evaluation index data set; Step S4: Divide the reservoir evaluation index dataset into a training set, a validation set, and a test set according to intervals; use the training set to construct the basic structure of a one-dimensional convolutional neural network model and introduce a geological attention mechanism to obtain a geologically attention-enhanced neural network; Step S5: Guide the geologically attention-enhanced neural network with weak supervision signals to obtain a geologically knowledge-constrained attention network; perform data synthesis and augmentation on the training set for small-sample reservoir accumulation through a generative adversarial network to obtain a balanced training dataset; use the geologically knowledge-constrained attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; Step S6: Use the test set to evaluate and optimize the performance of the candidate model to obtain an initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

[0006] The present invention systematically integrates core data and well logging data to improve the accuracy and data integrity of reservoir evaluation, and solves the problem of error accumulation caused by improper handling of data missing and anomalies in traditional methods. The introduction of multi-scale fusion of pore structure parameters and macroscopic porosity strengthens the comprehensive analysis ability of reservoir physical properties and helps to improve the accuracy of reservoir quality identification. The use of a geological attention mechanism combined with a one-dimensional convolutional neural network improves the model's perception ability of key geological features and the expression ability of geological context, and overcomes the problem of attention drift in pure data-driven attention mechanisms. At the same time, through weak supervision signal guidance and geological knowledge embedding, the interpretability and geological consistency of the model are effectively improved. For the small-sample reservoir problem, a generative adversarial network is used for data augmentation, enhancing the training stability and generalization ability of the model in the case of insufficient samples. The final model has stronger robustness, adaptability, and prediction accuracy, significantly improving the reservoir evaluation efficiency and development decision-making support ability under complex geological conditions.

[0007] Preferably, the present invention also provides a construction system for a reservoir evaluation model based on deep learning for executing the above-mentioned construction method of a reservoir evaluation model based on deep learning. The construction system for a reservoir evaluation model based on deep learning includes: A data preprocessing module, configured to obtain well logging data and core analysis data of a target work area, and perform outlier processing and missing value filling respectively to obtain complete well logging data and complete core data; A pore parameter inversion module, configured to calculate reservoir pore structure parameters based on the complete core data; invert the macroscopic porosity of the reservoir based on the complete well logging data; A reservoir property analysis module, which is used to analyze the physical properties of reservoir lithofacies based on reservoir pore structure parameters and reservoir macroscopic porosity, determine the pore connectivity index, reservoir quality factor, lithology control coefficient and heterogeneity index, and generate a reservoir evaluation index dataset; A geological attention model construction module, which is used to divide the reservoir evaluation index dataset into a training set, a validation set and a test set according to intervals; construct the basic structure of a one-dimensional convolutional neural network model by using the training set, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; A model training and enhancement module, which is used to perform weak supervision signal guidance on the geological attention enhanced neural network to obtain a geological knowledge constrained attention network; perform data synthesis and augmentation of small-sample reservoir accumulation on the training set through a generative adversarial network to obtain a balanced training dataset; use the geological knowledge constrained attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; A model evaluation and integration module, which is used to perform performance evaluation and tuning on the candidate model by using the test set to obtain an initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

[0008] Through the collaborative work of multiple modules, the present invention effectively improves the accuracy and reliability of reservoir evaluation. In the data preprocessing stage, through outlier processing and missing value filling, the integrity of logging data and core data is ensured, providing high-quality input data for subsequent analysis. The pore parameter inversion module combines core data and logging data to achieve accurate inversion of reservoir pore structure and macroscopic porosity, laying a foundation for reservoir physical property analysis. The reservoir property analysis module further deeply analyzes pore connectivity, reservoir quality, lithology control and heterogeneity, providing a comprehensive reservoir index dataset for comprehensive evaluation. The geological attention model construction module optimizes the model's attention to key geological features by introducing a geological attention mechanism, making the model more interpretable and accurate when processing complex geological data. The model training and enhancement module enhances the model's learning ability and generalization ability through weak supervision guidance and small-sample data augmentation, ensuring that the model can still be efficiently trained and optimized in the case of incomplete or unbalanced data. Finally, the model evaluation and integration module generates an accurate and geologically realistic final reservoir evaluation model through test set evaluation and tuning, combined with geological knowledge constraints, comprehensively improving the intelligent level and application effect of reservoir evaluation, and providing strong technical support for oil and gas exploration and development. Description of the Drawings

[0009] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of the non-restrictive embodiments with reference to the following drawings: Figure 1 Schematic diagram of the step process for a method of constructing a reservoir evaluation model based on deep learning according to the present invention; Figure 2 is Figure 1 detailed schematic diagram of step S1 in Figure 3 is Figure 1 detailed schematic diagram of step S3 in Specific embodiments

[0010] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative work based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0011] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although terms such as "first" and "second" may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0013] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides a method for constructing a reservoir evaluation model based on deep learning, and the method includes the following steps: Step S1: Obtain well logging data and core analysis data of the target work area, and perform outlier processing and missing value filling respectively to obtain complete well logging data and complete core data; Step S2: Calculate reservoir pore structure parameters based on the complete core data; invert the macroscopic porosity of the reservoir based on the complete well logging data; Step S3: Analyze the physical properties of reservoir lithofacies based on reservoir pore structure parameters and reservoir macroscopic porosity, determine the pore connectivity index, reservoir quality factor, lithology control coefficient and heterogeneity index, and generate a reservoir evaluation index dataset; Step S4: Divide the reservoir evaluation index dataset into a training set, a validation set and a test set according to intervals; use the training set to construct the basic structure of a one-dimensional convolutional neural network model, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; Step S5: Guide the geological attention enhanced neural network with weak supervision signals to obtain a geological knowledge constrained attention network; perform data synthesis and augmentation on the training set for small sample reservoir accumulation through a generative adversarial network to obtain a balanced training dataset; use the geological knowledge constrained attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; Step S6: Use the test set to evaluate and optimize the performance of the candidate model to obtain an initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

[0014] In the embodiment of the present invention, refer to Figure 1 As shown, it is a schematic diagram of the step flow of a method for constructing a reservoir evaluation model based on deep learning according to the present invention. In this example, the method for constructing a reservoir evaluation model based on deep learning includes the following steps: Step S1: Obtain the logging data and core analysis data of the target work area, and perform outlier processing and missing value filling respectively to obtain complete logging data and complete core data; In the embodiment of the present invention, first, use a sodium iodide crystal detector to collect natural gamma data. Its working voltage is 1200V, the current is 10mA, the crystal size of the detector is 5cm×5cm×5cm, and the sampling interval is 0.1s. At the same time, use a comprehensive logging tool to collect acoustic travel time, density, neutron porosity and resistivity data. Among them, the working frequency of the acoustic travel time sensor is 20kHz, and the measurement range is 30μs / m to 200μs / m; the source distance of the density sensor is 35cm, and the measurement accuracy is ±0.01g / cm³; the neutron source intensity of the neutron porosity sensor is , the measurement range is from 0% to 50%; the operating frequency of the resistivity sensor is 1 MHz, and the measurement range is from 0.1 Ω·m to 2000 Ω·m. The acquisition depth interval of all logging data is 0.1524 m. Then, porosity, permeability, and water saturation data are collected by a core porosity and permeability determination instrument. The instrument uses the gas permeation method, with an operating pressure range of 0 MPa to 70 MPa, a temperature control range of 20 °C to 300 °C, a porosity measurement accuracy of ±0.5%, a permeability measurement accuracy of ±10%, and a water saturation measurement accuracy of ±3%. The core sample size is 3.8 cm in diameter and 10 cm in length. Then, the collected logging data and core analysis data are depth-matched using the bicubic interpolation algorithm, with a matching accuracy of ±0.1 m. For the logging data, the 3σ principle is used for outlier processing, that is, data outside the range of the mean plus or minus 3 times the standard deviation is determined as an outlier, and missing values are filled using linear interpolation; for the core analysis data, outliers are determined by the box-and-whisker plot method, and missing values are filled using the mean method to obtain complete logging data and complete core data.

[0015] Step S2: Calculate reservoir pore structure parameters based on the complete core data; invert the macroscopic porosity of the reservoir based on the complete logging data; In the embodiments of the present invention, a complete core data is scanned by an X-ray computed tomography (CT) device. The X-ray tube voltage of the CT device is 120 kV, the current is 100 μA, the detector pixel size is 100 μm×100 μm, the scanning layer thickness is 0.5 mm, and the scanning range covers the entire length of the core. The scanned image data is imported into image processing software, and the adaptive threshold segmentation algorithm is used to process the image, where the threshold is calculated by the Otsu method. The three-dimensional distribution data of the pore space after segmentation is stored in the form of voxels. Using a pore geometry-topology feature extraction tool based on the Marching Cubes algorithm, geometric features such as the volume, surface area, and connectivity of the pores are calculated, and the K-means clustering analysis is performed on the extraction results. The value of K is set to 3 to obtain the pore type classification result. Using a lithology-dependent scale box dimension calculation tool based on the fractal theory, the fractal dimension feature data of the pore space is calculated, and the calculation range is from 2 to 3. Based on the pore type classification result, the integral calculation method is used to quantify the volume of each type of pore to obtain the microporosity distribution data. Using the maximum inscribed sphere algorithm to calculate the pore type classification result to obtain the pore throat radius distribution data. Using a tensor field analysis tool to analyze the pore geometry to obtain the reservoir pore geometry parameters. At the same time, the multi-sensor acquisition accuracy calibration and unified standardization processing are performed on the complete logging data. The least squares method is used to fit the calibration curve to uniformly convert the data collected by different sensors into standard logging curve data. Using the principal component analysis method to perform multi-variable non-linear covariant analysis on the standardized logging curve data to extract the logging response characteristic parameters. Based on the logging response characteristic parameters, a spectral analysis method based on the Fast Fourier Transform (FFT) is used to analyze the reservoir fluid properties, and the reservoir fluid phase separation is performed based on the spectral analysis result to obtain the fluid type distribution data. Combining the standardized logging curve data with the fluid type distribution data, a joint inversion method of neutron-density porosity based on Bayesian inversion is used to obtain the initial effective porosity data. Based on the initial effective porosity data, an X-ray diffractometer is used to measure the clay mineral content in the reservoir. The diffraction angle range is 2θ = 3° to 60°, the scanning step is 0.02°, the diffraction intensity is recorded in the counting rate mode, the clay mineral content is calculated based on the diffraction peak intensity and position, and the initial effective porosity data is corrected by combining the correction coefficient to obtain the reservoir macroscopic porosity data.

[0016] Step S3: Analyze the physical properties of the reservoir lithofacies based on the reservoir pore structure parameters and the reservoir macroscopic porosity, determine the pore connectivity index, reservoir quality factor, lithology control coefficient, and heterogeneity index, and generate a reservoir evaluation index dataset; In the embodiments of the present invention, first, the pore geometric morphology parameters and the pore throat radius distribution data are input into an analysis tool based on graph theory algorithms. This tool uses the breadth-first search algorithm and takes the pore connectivity condition as the search rule to construct a pore network topology relationship graph. Based on this graph, the Dijkstra shortest path algorithm is used to trace the seepage path, and effective fluid channel data such as path length, quantity, and connectivity probability are extracted. These data are input into a numerical integration tool, and the Simpson integration method is used for numerical integration. The result is processed by statistical normalization based on the maximum-minimum normalization method to obtain the pore connectivity index. Secondly, using the pore fractal dimension characteristic data, it is analyzed by a scale qualitative analysis tool based on fractal geometry theory to obtain the initial score of pore complexity. This score and the micro-porosity distribution data are input into a weighted fusion tool, and the weighted average method is used for fusion, with weight coefficients of 0.6 and 0.4 respectively, to obtain the initial pore structure complexity index. This index is input into the experimental verification and calibration module of a mercury intrusion porosimeter, and the pressure calibration method is used, with a calibration pressure range of 0 MPa to 70 MPa, to obtain the corrected pore structure complexity index. Then, using the reservoir macroscopic porosity data and the pore connectivity index, they are input into an evaluation tool based on the reservoir physical model, and the linear regression method is used to construct a reservoir fluid storage capacity evaluation model to generate a reservoir capacity factor. At the same time, the permeability data and the pore connectivity index in the complete core data are input into a correlation analysis tool, and the Pearson correlation coefficient method is used for correlation analysis to obtain a seepage capacity factor. The reservoir capacity factor and the seepage capacity factor are input into a multi-objective optimization tool, with the optimization objectives being to maximize the reservoir capacity factor and to maximize the seepage capacity factor respectively, and the weighted sum method is used for integration, with weight coefficients of 0.7 and 0.3 respectively, to obtain a reservoir quality factor. Then, based on the fluid type distribution data and the reservoir macroscopic porosity data, they are input into a rock-fluid interaction analysis tool, and an analysis method based on fluid mechanics and thermodynamic equilibrium is used to obtain a wettability evaluation result. This result and the corrected pore structure complexity index are input into a multi-variable regression tool, and the multiple linear regression method is used to obtain a lithology control coefficient. Using the reservoir pore geometric morphology parameters and the pore type classification results, they are input into a spatial variability identification tool, and the semi-variogram analysis method is used to identify the spatial variability and generate a spatial variability coefficient. This coefficient is input into a sequence constraint processing tool, and a constraint algorithm based on sequence stratigraphy is used to obtain a heterogeneity index. Finally, the reservoir quality factor, the lithology control coefficient, and the heterogeneity index are input into an analytic hierarchy process weight assignment tool, and the calculation process of the analytic hierarchy process (AHP) is used to construct a judgment matrix, and the eigenvalue method is used to calculate the index weight matrix. This matrix and each evaluation index are input into a weighted fusion tool, and the weighted summation method is used to obtain a reservoir comprehensive evaluation score. This score is input into a quantization grading tool, and the equal-distance segmentation method is used to divide the score range into five levels to obtain a reservoir evaluation index data set.

[0017] Step S4: Divide the reservoir evaluation index data set into a training set, a validation set, and a test set according to layers; use the training set to construct the basic structure of a one-dimensional convolutional neural network model, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; In the embodiment of the present invention, first, the reservoir evaluation index data set is input into a feature correlation analysis tool. This tool uses the Pearson correlation coefficient method with a threshold set to 0.7 for multicollinearity detection, and uses the variance inflation factor (VIF) method with a threshold set to 10 to generate a standardized feature data set. Then, based on the preset geological stratification information, using a geological stratification identification tool and adopting a convolutional neural network (CNN) algorithm based on deep learning, with a training set ratio of 75%, a validation set ratio of 15%, and a test set ratio of 15%, the standardized feature data set is divided into reservoir layers to obtain a labeled layer data set. Use the training set to construct the basic structure of a one-dimensional convolutional neural network model. Adopting the TensorFlow framework, the number of neurons in the input layer of the model is 128, the convolution kernel size of the convolution layer group is 3×3, the stride is 1, the padding method is same, the activation function is ReLU, in the combination design of max pooling and average pooling, the pooling window size is 2×2, the stride is 2, the number of neurons in the fully connected layer is 64, the number of neurons in the output layer is 8, and the activation function is softmax for non-linear transformation configuration to obtain the basic network structure. Perform Dropout and L2 weight regularization on the basic network structure, with a Dropout probability of 0.5 and an L2 regularization coefficient of 0.01 to obtain an overfitting prevention enhanced network structure. Introduce a self-attention calculation unit, adopt the Scaled Dot-ProductAttention mechanism, with 8 attention heads and a hidden layer dimension of 512, perform weight initialization to obtain a geological sensitive feature enhancement component. Perform weight allocation on the geological sensitive feature enhancement component, adopt an optimization algorithm based on gradient descent with a learning rate of 0.001 to generate geological key features. Based on the geological key features, perform multi-scale integration of the original convolutional features, adopt a feature pyramid network (FPN) structure, and integrate convolution kernels with scales of 1×1, 3×3, and 5×5 to obtain geological enhanced features. Based on the geological enhanced features, construct skip connections, adopt the skip connection method in the U-Net architecture, with 4 connection layers to obtain a gradient transfer optimization network structure. Perform an attention-guided loss combination design, adopt a combination of the cross-entropy loss function and the Dice loss function, with weight coefficients of 0.7 and 0.3 respectively, to obtain a geological perception comprehensive loss function. Integrate the basic network structure, the geological sensitive feature enhancement component, the gradient transfer optimization network structure, and the geological perception comprehensive loss function into a geological attention enhanced neural network through the distributed deep learning framework Horovod.

[0018] Step S5: Perform weak supervision signal guidance on the geological attention enhancement neural network to obtain a geological knowledge-constrained attention network; perform data synthesis and augmentation of small-sample reservoir accumulation on the training set through a generative adversarial network to obtain a balanced training data set; use the geological knowledge-constrained attention network and the balanced training data set to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; In the embodiments of the present invention, first, a geological constraint rule set is obtained. The rule set includes 10 geological constraint rules such as reservoir type, pore structure, and permeability. Each rule is set with a clear threshold. For example, the porosity threshold is 15%-30%, and the permeability threshold is 0.1-1000 mD. Based on this rule set, each rule in the training set is matched one by one. An exact matching algorithm is used, and the matching accuracy is 100% to obtain weakly supervised signal labels. The weakly supervised signal labels are applied to the geological attention enhanced neural network. An attention weight optimization algorithm is used, and the optimization goal is to minimize the prediction error. The learning rate is 0.001, and the number of iterations is 1000 times to generate a weakly supervised guidance architecture. Based on the weakly supervised guidance architecture, a geological rule consistency loss term is designed. The mean square error method is used to calculate the loss value, and the weight coefficient is 0.8 to generate a multi-objective loss function. The multi-objective loss function is constrained and optimized. The Lagrange multiplier method is used, and the constraint condition is that the loss value is less than 0.01 to obtain a geological knowledge constraint optimizer. The geological attention enhanced neural network and the geological knowledge constraint optimizer are integrated and reconstructed. A modular integration method is used, and the interface protocol is TCP / IP to obtain a geological knowledge constraint attention network. The sample distribution statistics and clustering analysis of each reservoir type in the training set are performed. The K-means clustering algorithm is used, the K value is set to 5, and the clustering accuracy is 95% to identify small sample paragraphs and obtain a small sample paragraph label set. The feature-preserving data generation of the small sample paragraph label set is performed through a generative adversarial network (GAN). Both the generator and discriminator of the GAN are 5-layer neural networks. The activation function is ReLU, the optimization algorithm is Adam, the learning rate is 0.002, and the batch size is 32 to generate synthetic reservoir small sample data. The distribution resampling of the training set and the synthetic reservoir small sample data is performed. The stratified sampling method is used, and the sampling rate is 1.5 times. Based on the distribution resampling result, the weight ratio of each type of reservoir is corrected, and the correction coefficient is 0.8-1.2 to obtain a balanced training data set. The forward propagation and backpropagation training of the balanced training data set are performed based on the geological knowledge constraint attention network. The stochastic gradient descent method is used, the learning rate is 0.01, the momentum coefficient is 0.9, the batch size is 64, and the number of iterations is 2000 times to obtain an initial network weight parameter set. The loss function value, prediction error of the validation set, and gradient change trend during the training process are statistically analyzed. The moving average method is used to calculate the dynamic performance index, and the window size is 50 to generate a dynamic performance index data set. The automatic hyperparameter search of the geological knowledge constraint attention network is performed based on the dynamic performance index data set. The Bayesian optimization algorithm is used, and the search space is a learning rate of 0.001-0.1, a batch size of 16-128, and the number of iterations of 1000-3000 to obtain an optimal hyperparameter combination set. The batch training iteration of the balanced training data set is performed based on the optimal hyperparameter combination set. A parallel computing method is used, and GPU acceleration is performed. The number of iterations is 3000 times to obtain a candidate model parameter set.Perform early stopping and Dropout validation on the candidate model parameter set based on the validation set. The early stopping condition is that the validation loss does not decrease for 10 epochs, and the Dropout probability is 0.5, to obtain the candidate model.

[0019] Step S6: Use the test set to evaluate and optimize the performance of the candidate model to obtain the initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate the final reservoir evaluation model.

[0020] In the embodiment of the present invention, perform a prediction task on the candidate model using the test set. The test set contains 1,000 reservoir samples, each sample having 10 features and 1 true reservoir parameter label. Input the test set data into the candidate model and perform calculations using the forward propagation algorithm to obtain a set of reservoir parameter prediction values. The prediction values include 5 reservoir parameters such as porosity and permeability, and the prediction accuracy is 95%. Based on the set of reservoir parameter prediction values and the true core labels, conduct an error comparison analysis, and use the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) as error evaluation indicators. Based on the model error evaluation indicator set, perform parameter fine-tuning and structure fine-tuning on the candidate model. Parameter fine-tuning uses the gradient descent method, with a learning rate of 0.01, 100 iterations, and a batch size of 32. Structure fine-tuning includes adding 1 hidden layer, with 64 neurons in the hidden layer and the activation function being ReLU. Obtain the initial reservoir evaluation model, and the prediction accuracy of this model for the test set reaches 97%. Perform integrated nesting on the initial reservoir evaluation model and the preset SZ3 framework, using a modular integration method, with the interface protocol being TCP / IP and the data transmission format being JSON. The SZ3 framework provides data preprocessing, feature extraction, and model optimization functions. The initial reservoir evaluation model sends the prediction results to the SZ3 framework in JSON format, and the SZ3 framework performs post-processing on the results, including error correction and visualization display, to finally obtain the final reservoir evaluation model.

[0021] The present invention enhances the accuracy and data integrity of reservoir evaluation by systematically integrating core data and well logging data, and solves the problem of error accumulation caused by improper handling of data missing and anomalies in traditional methods. The introduction of multi-scale fusion of pore structure parameters and macroscopic porosity strengthens the comprehensive analysis ability of reservoir physical properties and helps to improve the accuracy of reservoir quality identification. By using a geological attention mechanism combined with a one-dimensional convolutional neural network, the model's perception ability of key geological features and the expression ability of geological context are enhanced, overcoming the problem of attention drift in pure data-driven attention mechanisms. At the same time, through weak supervision signal guidance and geological knowledge embedding, the interpretability and geological consistency of the model are effectively improved. For the problem of small-sample reservoirs, a generative adversarial network is used for data augmentation, enhancing the training stability and generalization ability of the model in the case of insufficient samples. The final model has stronger robustness, adaptability and prediction accuracy, significantly improving the reservoir evaluation efficiency and development decision-making support ability under complex geological conditions.

[0022] Preferably, step S1 includes the following steps: Step S11: Collect natural gamma through a sodium iodide crystal detector, and collect acoustic travel time, density, neutron porosity and resistivity through a comprehensive well logging tool to generate well logging data; Step S12: Collect porosity, permeability and water saturation through a core porosity and permeability measuring instrument to generate core analysis data; Step S13: Perform depth matching on the well logging data and core analysis data, and respectively perform outlier processing and missing value filling to obtain complete well logging data and complete core data.

[0023] In the embodiment of the present invention, first, natural gamma data is collected through a sodium iodide crystal detector. The working voltage of the detector is 1200V, the current is 10mA, the crystal size is 5cm×5cm×5cm, and the sampling interval is 0.1s. At the same time, acoustic travel time, density, neutron porosity and resistivity data are collected by using a comprehensive well logging tool. The working frequency of the acoustic travel time sensor is 20kHz, and the measurement range is 30μs / m to 200μs / m; the source spacing of the density sensor is 35cm, and the measurement accuracy is ±0.01g / cm³; the neutron source intensity of the neutron porosity sensor is , the measurement range is from 0% to 50%; the operating frequency of the resistivity sensor is 1 MHz, and the measurement range is from 0.1 Ω·m to 2000 Ω·m. The acquisition depth interval for all logging data is 0.1524 m. Then, porosity, permeability, and water saturation data are collected through a core porosity and permeability measuring instrument. This measuring instrument uses the gas permeability method, with a working pressure range of 0 MPa to 70 MPa, a temperature control range of 20 °C to 300 °C, a porosity measurement accuracy of ±0.5%, a permeability measurement accuracy of ±10%, a water saturation measurement accuracy of ±3%, and the core sample size is 3.8 cm in diameter and 10 cm in length. Then, the collected logging data and core analysis data are depth-matched using the bicubic interpolation algorithm, and the matching accuracy is ±0.1 m. For the logging data, the 3σ principle is used for outlier processing, that is, data outside the range of the mean plus or minus 3 times the standard deviation is determined as an outlier, and missing values are filled using linear interpolation; for the core analysis data, outliers are determined by the box-and-whisker plot method, and missing values are filled using the mean method, so as to obtain complete logging data and complete core data.

[0024] The present invention comprehensively collects multi-physical quantity logging data such as natural gamma, acoustic travel time, density, neutron porosity, resistivity, etc., and key physical property parameters such as porosity, permeability, and water saturation of the core by integrating a variety of logging and experimental equipment, realizing multi-source complementarity of reservoir information. By depth-matching the logging and core data, the consistency and spatial correspondence accuracy of the data are improved, laying a high-quality data foundation for subsequent attribute inversion and model training. Outlier processing and missing value filling further enhance the integrity and usability of the data, reduce the impact of noise interference and information loss on the analysis results, ensure the stability and credibility of the model input data, and thus effectively improve the accuracy and reliability of reservoir evaluation.

[0025] Preferably, calculating the reservoir pore structure parameters based on the complete core data in step S2 includes: Performing a CT scan on the complete core data to generate a three-dimensional digital model of the core; Adaptive threshold segmentation of the reservoir pore space based on the three-dimensional digital model of the core to obtain three-dimensional distribution data of the pore space; Extracting pore geometry-topology features from the three-dimensional distribution data of the pore space, and performing cluster analysis on the results of the pore geometry-topology feature extraction to obtain a pore type classification result; Calculating the lithology-dependent scale box dimension of the three-dimensional distribution data of the pore space to obtain pore fractal dimension feature data; Quantitatively integrating and calculating the volume of each type of pore based on the pore type classification result to obtain micro-porosity distribution data; Performing the calculation of the maximum inscribed sphere on the pore type classification result to obtain pore throat radius distribution data; Perform a tensor field analysis on the pore geometry based on the pore type classification results to obtain the reservoir pore geometry parameters.

[0026] In the embodiments of the present invention, an X-ray computed tomography (CT) device is used to scan the complete core data. The X-ray tube voltage of the device is set to 120 kV, the current is 100 μA, the detector pixel size is 100 μm × 100 μm, the scan layer thickness is 0.5 mm, and the scan range covers the entire length of the core. The scanned image data is imported into image processing software and processed using an adaptive threshold segmentation algorithm, where the threshold is calculated by the Otsu method. The three-dimensional distribution data of the pore space after segmentation is stored in the form of voxels. A pore geometry-topology feature extraction tool based on the Marching Cubes algorithm is used to calculate geometric features such as the volume, surface area, and connectivity of the pores, and a K-means clustering analysis is performed on the extraction results with the K value set to 3 to obtain the pore type classification results. A lithology-dependent scale box dimension calculation tool based on the fractal theory is used to calculate the fractal dimension feature data of the pore space, and the calculation range is from 2 to 3. Based on the pore type classification results, an integral calculation method is used to quantify the volume of each type of pore to obtain the microporosity distribution data. The maximum inscribed sphere algorithm is used to calculate the pore type classification results to obtain the pore throat radius distribution data. A tensor field analysis tool is used to analyze the pore geometry to obtain the reservoir pore geometry parameters. In specific operations, after the image data generated by CT scanning is subjected to adaptive threshold segmentation, a three-dimensional geometric model of the pores is constructed by the Marching Cubes algorithm, and characteristic parameters such as the volume, surface area, and connectivity of the pores are calculated. The K-means clustering analysis classifies the pores into three categories, and the characteristic parameters of each category of pores are extracted and recorded. The fractal dimension calculation determines the complexity of the pore space within the range of 2 to 3 by the box dimension method. The acquisition of the microporosity distribution data is based on integral calculation to quantify and statistically analyze the volume of different types of pores. The calculation of the pore throat radius distribution data relies on the maximum inscribed sphere algorithm to determine the radius of the maximum inscribed sphere in the pores, reflecting the size distribution of the pore throats. The tensor field analysis tool obtains the geometric morphology parameters of the reservoir pores by analyzing the anisotropic characteristics of the pore geometry, and these parameters together constitute the reservoir pore structure parameters.

[0027] Through core CT scanning and 3D reconstruction, the present invention realizes high-resolution visualization and digital characterization of pore structures, significantly enhancing the perception ability of reservoir microstructures. Adaptive threshold segmentation and geometric-topological feature extraction accurately separate and quantify pore space information, enhancing the description accuracy of complex pore systems. Automatic identification of pore types is achieved through clustering analysis, providing a clear grouping basis for subsequent quantitative analysis; calculation of fractal dimensions reveals the multi-scale non-linear characteristics of pore structures, enriching the characterization dimensions of pore space structures. Precise acquisition of key parameters such as microporosity distribution and pore throat radius helps to more comprehensively reflect the seepage capacity and reservoir space characteristics of the reservoir. Tensor field analysis further captures the anisotropy and spatial orientation characteristics of pore geometries, providing important support for constructing high-fidelity reservoir physical models and overall enhancing the scientificity and accuracy of reservoir fine evaluation and modeling.

[0028] Preferably, the inversion of reservoir macroscopic porosity based on complete logging data in step S2 includes: Perform multi-sensor acquisition accuracy calibration and unified standardization processing on the complete logging data to obtain standardized logging curve data; Perform multi-variable non-linear covariant analysis on the standardized logging curve data to generate logging response characteristic parameters; Analyze the reservoir fluid properties based on the logging response characteristic parameter spectrum, and perform reservoir fluid phase separation based on the spectrum analysis results to obtain fluid type distribution data; Perform fluid-constrained neutron-density porosity joint inversion on the standardized logging curve data and the fluid type distribution data to obtain initial effective porosity data; Perform X-ray diffraction on the clay mineral content in the reservoir based on the initial effective porosity data and correct it to obtain a clay correction coefficient; Perform numerical integration on the initial effective porosity data and the clay correction coefficient, and perform volume averaging on the numerical integration results to obtain reservoir macroscopic porosity data.

[0029] In the embodiments of the present invention, first, multi-sensor acquisition accuracy calibration and unified standardization processing are performed on the complete logging data. The least squares method is used to fit the calibration curve, and the data collected by different sensors are uniformly converted into standard logging curve data. The principal component analysis method is used to perform multi-variable non-linear co-variance analysis on the standardized logging curve data, and the logging response characteristic parameters are extracted. Based on the logging response characteristic parameters, a spectral analysis method based on Fast Fourier Transform (FFT) is used to analyze the reservoir fluid properties, and reservoir fluid phase separation is performed based on the spectral analysis results to obtain fluid type distribution data. The standardized logging curve data and the fluid type distribution data are combined, and a joint inversion method of neutron-density porosity based on Bayesian inversion is used to obtain the initial effective porosity data. Based on the initial effective porosity data, an X-ray diffractometer is used to measure the clay mineral content in the reservoir. The diffraction angle range is 2θ = 3° to 60°, the scanning step is 0.02°, the diffraction intensity is recorded in the counting rate mode, the clay mineral content is calculated based on the diffraction peak intensity and position, and the initial effective porosity data is corrected in combination with the correction coefficient to obtain the reservoir macroscopic porosity data. In specific operations, first, the logging data is subjected to accuracy calibration and standardization processing, and the calibration curve is fitted by the least squares method to uniformly convert the data collected by different sensors into standard logging curve data. Then, the principal component analysis method is used to extract the logging response characteristic parameters, the FFT spectral analysis method is used to analyze the reservoir fluid properties and perform fluid phase separation to obtain the fluid type distribution data. Next, the standardized logging curve data and the fluid type distribution data are combined, and the Bayesian inversion method is used for joint inversion of neutron-density porosity to obtain the initial effective porosity data. Finally, an X-ray diffractometer is used to measure the clay mineral content, and the initial effective porosity data is corrected in combination with the correction coefficient to obtain the reservoir macroscopic porosity data.

[0030] By standardizing and calibrating the accuracy of the logging data, the present invention improves the consistency and comparability of multi-source logging information and provides high-quality input for subsequent inversion. The multi-variable non-linear co-variance analysis strengthens the ability to model the non-linear relationship between logging responses and geological attributes, which helps to accurately extract the key parameters characterizing the pore structure and fluid characteristics. The spectral analysis and fluid phase separation effectively identify the distribution of different types of fluids in the reservoir, improving the adaptability and constraint ability of the inversion process to the influence of fluids. The joint inversion of neutron-density porosity combines fluid type information, greatly improving the accuracy and physical consistency of porosity inversion. Introducing X-ray diffraction for clay influence correction solves the interference of clay minerals on porosity calculation and enhances the geological authenticity of the inversion results. Finally, the macroscopic porosity data obtained through numerical integration and volume averaging has higher representativeness and engineering applicability, providing scientific and reliable basic parameters for reservoir quantitative characterization and development.

[0031] Preferably, step S3 includes the following steps: Step S31: Conduct a three-dimensional network connectivity assessment on the reservoir pore geometric morphology parameters and pore throat radius distribution data to obtain a pore network topology relationship diagram; trace the seepage path based on the pore network topology relationship diagram to obtain effective fluid channel data; perform numerical integration and statistical normalization processing on the effective fluid channel data to obtain a pore connectivity index; Step S32: Qualitatively analyze the microscopic pore structure based on the pore fractal dimension characteristic data scale to obtain an initial pore complexity score; perform weighted fusion on the initial pore complexity score and the microporosity distribution data to obtain an initial pore structure complexity index; verify and calibrate the initial pore structure complexity index through a mercury intrusion porosimeter for porous media to obtain a corrected pore structure complexity index; Step S33: Construct a reservoir fluid storage capacity assessment model based on the reservoir macroscopic porosity data and the pore connectivity index to generate a reservoir capacity factor; conduct a correlation analysis on the permeability data and the pore connectivity index in the complete core data to obtain a seepage capacity factor; perform multi-objective optimization integration on the reservoir capacity factor and the seepage capacity factor to obtain a reservoir quality factor; Step S34: Analyze the rock-fluid interaction based on the fluid type distribution data and the reservoir macroscopic porosity data to obtain a wettability assessment result; perform multivariate regression on the wettability assessment result and the corrected pore structure complexity index to obtain a lithology control coefficient; identify the spatial variability based on the reservoir pore geometric morphology parameters and the pore type classification result to generate a spatial variability coefficient; perform sequence constraint processing on the spatial variability coefficient to obtain a heterogeneity index; Step S35: Perform weight assignment for the reservoir quality factor, the lithology control coefficient, and the heterogeneity index by the analytic hierarchy process to obtain an index weight matrix; perform weighted fusion on each evaluation index based on the index weight matrix to obtain a reservoir comprehensive evaluation score; perform quantitative grading on the reservoir comprehensive evaluation score to obtain a reservoir evaluation index dataset.

[0032] In the embodiments of the present invention, first, the pore connectivity index is calculated. The pore geometric morphology parameters and the pore throat radius distribution data are input into an analysis tool based on the graph theory algorithm. This tool uses the breadth-first search algorithm and constructs a pore network topology relationship graph with the pore connectivity condition as the search rule. The connectivity threshold between pore nodes is set such that the pore throat radius ratio is greater than 0.7. Based on the constructed pore network topology relationship graph, the Dijkstra shortest path algorithm is used to trace the seepage path, and effective fluid channel data such as path length, quantity, and connectivity probability are extracted. These data are input into a numerical integration tool, and the Simpson integration method is used to perform numerical integration on the effective fluid channel data to obtain the original quantization value of pore connectivity. Then, the integration result is subjected to statistical normalization processing based on the maximum-minimum normalization method, and the value range of the pore connectivity index is constrained between 0 and 1, finally obtaining the pore connectivity index. The pore fractal dimension characteristic data are input into a scale qualitative analysis tool based on the fractal geometry theory, and after scale qualitative analysis, the initial score of pore complexity is obtained, with the score range from 1 to 10. This score and the microporosity distribution data obtained in step S2 are input into a weighted fusion tool, and the weighted average method is used for fusion, with the weight coefficients being 0.6 and 0.4 respectively, to obtain the initial pore structure complexity index. This index is input into the experimental verification and calibration module of a mercury intrusion porosimeter for porous media. The pressure calibration method is adopted, and the calibration pressure range is from 0 MPa to 70 MPa. By comparing the pore structure data obtained from the mercury intrusion experiment with the initial pore structure complexity index, a calibration model is constructed to correct the initial index, obtaining the corrected pore structure complexity index. The reservoir macroscopic porosity data and the pore connectivity index are input into an evaluation tool based on the reservoir physical model, and the linear regression method is used to construct a reservoir fluid storage capacity evaluation model to generate the reservoir capacity factor. At the same time, the permeability data in the complete core data and the pore connectivity index are input into a correlation analysis tool, and the Pearson correlation coefficient method is used for correlation analysis to obtain the seepage capacity factor. The reservoir capacity factor and the seepage capacity factor are input into a multi-objective optimization tool, with the optimization objectives being to maximize the reservoir capacity factor and to maximize the seepage capacity factor respectively. The weighted sum method is used for integration, with the weight coefficients being 0.7 and 0.3 respectively, to obtain the reservoir quality factor. Based on the fluid type distribution data and the reservoir macroscopic porosity data, they are input into a rock-fluid interaction analysis tool, and an analysis method based on fluid mechanics and thermodynamic equilibrium is used to obtain the wettability evaluation result. The evaluation result is expressed by the wettability index, with the range from 0 to 1. This result and the corrected pore structure complexity index are input into a multi-variable regression tool, and the multiple linear regression method is used to obtain the lithology control coefficient. The reservoir pore geometric morphology parameters and the pore type classification result are used to input into a spatial variability identification tool, and the semi-variogram analysis method is used to identify the spatial variability and generate the spatial variability coefficient.Perform sequence constraint processing on the spatial variation coefficient, adopt a constraint algorithm based on sequence stratigraphy to obtain the heterogeneity index, and the index value ranges from 0 to 1. The larger the value, the stronger the heterogeneity. Input the reservoir quality factor, lithology control coefficient and heterogeneity index into the analytic hierarchy process (AHP) weight allocation tool, adopt the calculation process of the analytic hierarchy process (AHP) to construct a judgment matrix, and calculate the index weight matrix by the eigenvalue method. Based on the index weight matrix, use the weighted summation method to perform weighted fusion on each evaluation index to obtain the comprehensive reservoir evaluation score, and the score range is from 0 to 100 points. Finally, perform quantitative grading on the comprehensive reservoir evaluation score. Using the equal-distance segmentation method, divide the score range into five levels: 0-20 points is the extremely poor reservoir, 21-40 points is the poor reservoir, 41-60 points is the medium reservoir, 61-80 points is the good reservoir, and 81-100 points is the excellent reservoir to obtain the reservoir evaluation index data set.

[0033] The present invention accurately depicts the reservoir seepage path and connectivity characteristics through three-dimensional pore network modeling and fluid channel tracing, improving the cognitive depth of the reservoir's microscopic seepage ability. Integrate the fractal dimension and microporosity to quantitatively evaluate the pore structure complexity, and calibrate it through mercury intrusion experiments to ensure the physical consistency and engineering feasibility of the evaluation results. The comprehensive modeling of reservoir capacity and seepage ability provides a quality factor that comprehensively characterizes the reservoir quality, reflecting the collaborative characteristics of the reservoir in terms of storage and conduction capabilities. The rock-fluid interaction analysis reveals the regulation effect of wettability on the pore structure, and the establishment of the lithology control coefficient enhances the model's ability to explain geological attribute variations. The heterogeneity index captures the internal structural differences of geological bodies through spatial variation coefficient and sequence constraint modeling, providing support for fine reservoir division. Finally, through the analytic hierarchy process, scientific weight allocation of each index is realized, and the comprehensively weighted generated reservoir evaluation index system has high accuracy, systematicness and scalability, providing a strong decision-making basis for subsequent reservoir classification, development optimization and risk prediction.

[0034] Preferably, step S4 includes the following steps: Step S41: Conduct feature correlation analysis on the complete data set and perform multicollinearity detection to generate a standardized feature data set; Step S42: Identify reservoir layers based on the preset geological stratification information for the standardized feature data set, and divide the reservoir layers based on the identification results to obtain the labeled layer data set; divide the labeled layer data set according to a ratio, where 75% is the training set, 15% is the validation set, and 15% is the test set; Step S43: Use the training set to construct a one-dimensional convolutional neural network model and perform non-linear transformation configuration to obtain the basic network structure; Step S44: Perform Dropout and L2 weight regularization on the basic network structure to obtain an overfitting prevention enhanced network structure; Step S45: Introduce a self-attention calculation unit into the anti-overfitting enhanced network structure and perform weight initialization to obtain a geological sensitive feature enhancement component; Step S46: Perform weight allocation on the geological sensitive feature enhancement component to generate geological key features; based on the geological key features, perform multi-scale integration of the original convolutional feature application features to obtain geological enhanced features; Step S47: Construct a skip connection based on the geological enhanced features to obtain a gradient transfer optimization network structure; Step S48: Design an attention-guided loss combination for the gradient transfer optimization network structure to obtain a geological perception comprehensive loss function; Step S49: Integrate the basic network structure, the geological sensitive feature enhancement component, the gradient transfer optimization network structure, and the geological perception comprehensive loss function into a geological attention enhanced neural network through a distributed deep learning framework.

[0035] In the embodiments of the present invention, first, a feature correlation analysis is performed on the complete data set. Using the Pearson correlation coefficient method and setting the threshold to 0.7, features with high correlation are screened out. Then, a multicollinearity detection is carried out. Using the variance inflation factor (VIF) method and setting the threshold to 10, features with multicollinearity are removed to generate a standardized feature data set. Then, based on the preset geological stratification information, using a geological stratification identification tool and adopting the convolutional neural network (CNN) algorithm, with the training set ratio set to 75%, the validation set ratio set to 15%, and the test set ratio set to 15%, the standardized feature data set is divided into reservoir layers to obtain a labeled layer data set. The labeled layer data set is divided according to the ratio, where 75% is the training set, 15% is the validation set, and 15% is the test set. Next, based on the analysis of the reservoir evaluation model architecture requirements for the complete data set, a basic architecture diagram is drawn to clarify the model input, output, and intermediate layer structures. The input layer structure is designed based on the optimal feature subset, with the parameters set to 128 neurons and the data type to 64-bit floating-point numbers, obtaining the model input layer. A convolutional layer group is designed using multi-scale convolutional kernels (sizes of 1×1, 3×3, and 5×5) and a 64-bit floating-point precision calculation unit. The number of convolutional kernels in each layer is 64, and the activation function is ReLU. A combination design of max pooling and average pooling is performed on the convolutional layer group, with the pooling window size of 2×2 and the stride of 2, obtaining a dimensionality-reduced feature map pooling layer. Based on the dimensionality-reduced feature map pooling layer, a fully connected layer architecture is designed, with the number of neurons set to 256 and adopting the Dropout regularization configuration with a dropout probability of 0.5, obtaining a fully connected feature expression layer. Based on the preset reservoir parameter prediction task, the output layer structure is designed, with the number of neurons set to 8 and the activation function to softmax, obtaining a multi-task learning output layer. Nonlinear transformation configuration is performed on the model input layer, convolutional layer group, dimensionality-reduced feature map pooling layer, fully connected feature expression layer, and multi-task learning output layer, using the ReLU activation function, obtaining a basic network structure. Dropout and L2 weight regularization are performed on the basic network structure, with the Dropout probability set to 0.5 and the L2 regularization coefficient set to 0.01, obtaining an overfitting-preventing enhanced network structure. A self-attention calculation unit is introduced, adopting the Scaled Dot-Product Attention mechanism, with the number of attention heads set to 8 and the hidden layer dimension to 512, and weight initialization is performed, with the initialization method being He regularization, obtaining a geological sensitivity feature enhancement component. Weight allocation is performed on the geological sensitivity feature enhancement component, using an optimization algorithm based on gradient descent, with the learning rate set to 0.001, generating geological key features. Based on the geological key features, multi-scale integration of the original convolutional features is performed, using the Feature Pyramid Network (FPN) structure, integrating convolutional kernels with scales of 1×1, 3×3, and 5×5, obtaining geological enhanced features.Based on the geological enhancement features, skip connections are constructed, and the skip connection method in the U-Net architecture is adopted. The number of connection layers is 4, and the gradient transfer optimization network structure is obtained. The attention-guided loss combination design is carried out, and the combination of the cross entropy loss function and the Dice loss function is adopted, with weight coefficients of 0.7 and 0.3 respectively, to obtain the geological perception comprehensive loss function. Through the distributed deep learning framework Horovod, the basic network structure, geological sensitive feature enhancement components, gradient transfer optimization network structure and geological perception comprehensive loss function are integrated into a geological attention enhancement neural network. The data parallel strategy is set, and each GPU is responsible for the calculation of a part of the data. Finally, the results of each part are summarized to complete the construction of the geological attention enhancement neural network.

[0036] The present invention effectively improves the independence and expression ability of the model input features through feature correlation analysis and multicollinearity detection, and provides a high-quality data foundation for subsequent modeling. Combined with geological stratification, the reservoir layer segment can be accurately identified and marked, and the model's perception of geological distribution laws can be enhanced. The construction and nonlinear configuration of the one-dimensional convolutional neural network enable the model to have the ability to extract deep features of logging sequence data. The introduction of Dropout and L2 regularization mechanisms can effectively alleviate the overfitting problem and improve the stability and generalization ability of the model. The fusion of the self-attention mechanism and the geological sensitive feature enhancement component enables the model to focus more on key geological features and enhance the ability to capture important information. Multi-scale feature integration combined with jump connections optimizes the information flow and gradient propagation path, and improves the training efficiency and robustness of deep networks. The attention-guided loss function design enhances the sensitivity of the model to geological changes and helps capture the complex relationship between the nonlinear characteristics of the reservoir and the geological background. Finally, the geological attention-enhanced neural network constructed through the distributed deep learning framework has stronger geological adaptability, feature expression and prediction accuracy, providing a high-performance intelligent solution for complex reservoir evaluation.

[0037] It is particularly important that step S43 includes the following steps: Step S431: Analyze the reservoir evaluation model architecture requirements based on the complete data set to obtain a basic architecture diagram; Step S432: designing an input layer structure based on the optimal feature subset, and performing parameter configuration to obtain a model input layer; Step S433: designing a convolution layer based on the model input layer using a multi-scale convolution kernel and a 64-bit floating-point precision computing unit to obtain a convolution layer group; Step S434: performing a maximum pooling and an average pooling combination design on the convolution layer group to obtain a dimension reduction feature map pooling layer; Step S435: performing a fully connected layer architecture design and a dropout regularization configuration based on the dimension reduction feature map pooling layer to obtain a fully connected feature expression layer; Step S436: Design an output layer structure based on a preset reservoir parameter prediction task to obtain a multi-task learning output layer; Step S437: Perform non-linear transformation configuration on the model input layer, convolutional layer group, dimensionality-reduced feature mapping pooling layer, fully-connected feature expression layer, and multi-task learning output layer to obtain a basic network structure.

[0038] In the embodiment of the present invention, first, analyze the requirements of the reservoir evaluation model architecture based on the complete data set, draw a basic architecture diagram, and clarify the model input, output, and intermediate layer structures. Then, design the input layer structure based on the optimal feature subset, set the parameters to 128 neurons, and the data type to 64-bit floating-point numbers to obtain the model input layer. Next, design the convolutional layer using multi-scale convolutional kernels (sizes of 1×1, 3×3, and 5×5) and a 64-bit floating-point precision calculation unit, with 64 convolutional kernels in each layer and the activation function ReLU to obtain the convolutional layer group. Perform a combination design of max pooling and average pooling on the convolutional layer group, with a pooling window size of 2×2 and a stride of 2 to obtain the dimensionality-reduced feature mapping pooling layer. Design the fully-connected layer architecture based on the dimensionality-reduced feature mapping pooling layer, set the number of neurons to 256, and adopt Dropout regularization configuration with a dropout probability of 0.5 to obtain the fully-connected feature expression layer. After that, design the output layer structure based on the preset reservoir parameter prediction task, set the number of neurons to 8, and the activation function to softmax to obtain the multi-task learning output layer. Finally, perform non-linear transformation configuration on the model input layer, convolutional layer group, dimensionality-reduced feature mapping pooling layer, fully-connected feature expression layer, and multi-task learning output layer, and use the ReLU activation function to obtain the basic network structure. In specific operations, first analyze the requirements of the reservoir evaluation model architecture, draw a basic architecture diagram, and determine the model input, output, and intermediate layer structures. Design the input layer structure based on the optimal feature subset, set the number of neurons and the data type to obtain the model input layer. Design the convolutional layer group using multi-scale convolutional kernels and a 64-bit floating-point precision calculation unit, and set the number of convolutional kernels and the activation function in each layer to obtain the convolutional layer group. Perform pooling operations on the convolutional layer group, set the pooling window size and stride to obtain the dimensionality-reduced feature mapping pooling layer. Design the fully-connected layer architecture based on the dimensionality-reduced feature mapping pooling layer, set the number of neurons and adopt Dropout regularization configuration to obtain the fully-connected feature expression layer. Design the output layer structure according to the reservoir parameter prediction task, set the number of neurons and the activation function to obtain the multi-task learning output layer. Perform non-linear transformation configuration on each layer, use the ReLU activation function, and finally obtain the basic network structure.

[0039] Through feature correlation analysis and collinearity detection, the present invention screens and standardizes high-quality input features, effectively reducing redundant information and improving the model stability and learning efficiency. The formation identification guided by geological stratification enhances the adaptability of the model to geological structures, achieving accurate annotation and scientific data partitioning. The network architecture constructed based on task requirements integrates multi-scale convolution and high-precision computing units in design, strengthening the model's ability to identify reservoir features at different scales. The combined design of pooling layers retains key information while reducing dimensions, enhancing the model's ability to express local and global features. The fully connected layer and dropout configuration further improve the feature fusion effect and effectively prevent overfitting. The output layer, based on the concept of multi-task learning, takes into account the prediction requirements of multi-dimensional reservoir attributes, enhancing the model's generalization ability. Through the non-linear transformation and regularization mechanism running through each layer of the model structure, it ensures the efficient transmission of information flow and the in-depth extraction of features. The overall architecture design is scientific and reasonable, with good scalability, learning ability, and prediction accuracy, providing strong support for complex reservoir modeling and multi-parameter joint evaluation.

[0040] Preferably, the weak supervision signal guidance for the geological attention-enhanced neural network in step S5 includes: Obtaining a geological constraint rule set; performing rule matching on the training set based on the geological constraint rule set to obtain weak supervision signal labels; Applying the weak supervision signal labels to the geological attention-enhanced neural network and optimizing the attention weights to generate a weak supervision guidance architecture; Designing a geological rule consistency loss term based on the weak supervision guidance architecture to generate a multi-objective loss function; Performing constraint optimization on the multi-objective loss function to obtain a geological knowledge constraint optimizer; Integrating and reconstructing the geological attention-enhanced neural network and the geological knowledge constraint optimizer to obtain a geological knowledge constraint attention network.

[0041] In the embodiments of the present invention, first, a predefined set of geological constraint rules is extracted from the reservoir geological expert knowledge base. This rule set contains 120 explicit geological rules, covering 5 categories: porosity-permeability relationship rules (e.g., when the porosity < 5%, the permeability must < 0.1 mD; when the porosity > 15%, the permeability must > 10 mD), pore throat structure constraint rules (e.g., when the average throat radius < 2 microns, the connectivity index must < 0.4; when the throat-pore ratio < 0.1, the reservoir heterogeneity index must > 0.6), mineral composition influence rules (e.g., when the clay content > 12%, the lithology control coefficient must > 0.7; when the quartz content > 85%, the lithology control coefficient must < 0.3), sedimentation-related rules (e.g., in the delta front facies area, the change rate of the vertical heterogeneity index must < 0.05 / m; in the shoreface sandbar facies area, the quality factor must > 0.65), and fluid property constraint rules (e.g., when the water saturation > 65%, the reservoir quality factor must < 0.5; when the crude oil viscosity > 50 mPa·s, the pore connectivity index must < 0.6); then, rule matching is performed on the training set based on these rules. The specific operation is as follows: Each rule is converted into a deterministic condition-result statement. The condition part is expressed as a parameter range constraint (e.g., 5% ≤ porosity ≤ 10%), and the result part is expressed as a value range constraint (e.g., 0.1 mD ≤ permeability ≤ 1 mD). The parameters in the condition part are accurate to two decimal places, and the parameter range in the result part is accurate to the same precision level as the measured physical quantity; for each sample point in the training set, the probability value P_cond that meets the rule condition is calculated. P_cond follows a Gaussian function distribution on the interval [0, 1] (centered at the midpoint of the condition interval, with a standard deviation of 1 / 6 of the interval width). When the sample is completely inside the condition interval, P_cond = 1; when it is completely outside the condition interval, P_cond = 0; when it is near the boundary, 0 < P_cond < 1; at the same time, the probability value P_result that meets the rule result is calculated, and the calculation method is the same as that of P_cond; the rule matching degree M = P_cond × P_result is defined. For each sample point, the matching degrees M_i of all applicable rules are calculated, and the highest 5 M_i values are retained; these 5 M_i values are weighted and averaged with weights 1.0, 0.8, 0.6, 0.4, and 0.2 to obtain the comprehensive rule confidence level C of this sample point; according to the C value, the training set samples are divided into strong rule samples (C ≥ 0.85), medium rule samples (0.5 ≤ C < 0.85), weak rule samples (0.2 ≤ C < 0.5), and no rule samples (C < 0.2) Four categories; for each category of samples, generate corresponding weakly supervised signal labels: strongly regular samples are labeled as [1,0,0,0], moderately regular samples are labeled as [0,1,0,0], weakly regular samples are labeled as [0,0,1,0], and irregular samples are labeled as [0,0,0,1]; add these weakly supervised signal labels as auxiliary labels to the training set to obtain an enhanced training set with rule confidence labels; subsequently, apply the weakly supervised signal labels to the geological attention enhanced neural network by inserting a rule perception layer in the middle layer of the network (after the convolutional layer and before the fully connected layer). The rule perception layer sets an attention channel for each rule category, and each channel contains 64 attention units. Each attention unit is implemented as a W·tanh(Ux + b) structure, where W is a 1×64 weight vector, U is a 64×n transformation matrix (n is the feature dimension), b is a 64-dimensional bias vector, and tanh is the hyperbolic tangent activation function; apply the attention mechanism to the intermediate feature map F of the network, calculate the attention weight α = softmax(rule perception layer(F)), and then obtain the rule-enhanced feature representation through α·F; apply L1 regularization constraint to the attention weight, and set the regularization coefficient to 0.01 to prevent over-concentration of attention; introduce an attention update step during training, update the attention weight every 100 batches, and set the update step size to 0.003; generate a weakly supervised guidance architecture based on the updated attention structure, combine the original classification or regression task with the rule category prediction task to form a multi-task learning framework; based on the multi-task learning framework, design a geological rule consistency loss term, defined as Lrule = λ1·BCE(ypred_rule, ytrue_rule)+λ2·KL(αpred, αtarget), where BCE is the binary cross-entropy loss used to evaluate the accuracy of rule category prediction, KL is the KL divergence used to measure the difference between the predicted attention distribution and the target attention distribution, and λ1 and λ2 are weight coefficients, set to 0.2 and 0.15 respectively; combine the geological rule consistency loss term with the original task loss (mean squared error loss) to generate a multi-objective loss function Ltotal = Lorigin + Lrule; perform constrained optimization on the multi-objective loss function, use the Adam optimizer with momentum, set the initial learning rate to 0.001, the weight decay coefficient to 0.0005, the momentum parameter to 0.9, adopt the cosine annealing learning rate scheduling strategy, the minimum learning rate is 1 / 10 of the initial learning rate, and the period is 30 rounds; add a gradient clipping mechanism, and set the clipping threshold to 5.0. Prevent gradient explosion; implement an early stopping mechanism to monitor the rule consistency index on the validation set. Stop training if there is no improvement for 5 consecutive rounds and save the best model parameters. Finally, modularly integrate the backbone structure of the geological attention-enhanced neural network with the geological knowledge constraint optimizer, reconstruct the network computational graph, and optimize the memory layout to obtain the geological knowledge constraint attention network, which includes a rule embedding layer, an attention enhancement layer, and a rule consistency checking layer.

[0042] The present invention realizes weakly supervised labeling of training data by introducing a geological constraint rule set, effectively improves the learning ability of the model under the condition of incomplete sample labels, and enhances its response ability to geological prior knowledge. The optimization of attention weights guides the model to focus on geological key feature regions, improving the accuracy and geological consistency of feature selection. By introducing a geological rule consistency loss term, geological knowledge is explicitly incorporated into the model training objective, enabling the model to not only rely on data-driven but also have geological logical constraints, thereby improving the generalization ability and prediction reliability. The design of the geological knowledge constraint optimizer realizes the coordinated optimization among multi-objective training tasks, taking into account both accuracy and geological rationality. Finally, through the integration and reconstruction of the network structure and optimization strategy, the constructed geological knowledge constraint attention network has stronger knowledge fusion ability and geological interpretability, providing a stable and controllable model foundation for intelligent identification and fine evaluation in complex reservoir environments.

[0043] Especially importantly, the data synthesis and augmentation of small-sample reservoir accumulation for the training set through the generative adversarial network in step S5 includes: Conduct sample distribution statistics and clustering analysis for each reservoir type in the training, and identify small-sample paragraphs to obtain a small-sample paragraph label set; Generate feature-preserving data for the small-sample paragraph label set through the generative adversarial network to obtain synthetic reservoir small-sample data; Conduct distribution resampling for the training set and the synthetic reservoir small-sample data, and correct the weight ratio of each type of reservoir based on the distribution resampling result to obtain a balanced training data set.

[0044] In the embodiments of the present invention, first, the sample distributions of each reservoir type in the training set are statistically analyzed and clustered. The K-means clustering algorithm is used, with the K value set to 5 and the clustering accuracy to 95%. Small-sample paragraphs are identified to obtain a small-sample paragraph label set. The small-sample paragraph label set is subjected to feature-preserving data generation through a generative adversarial network (GAN). Both the generator and discriminator of the GAN are 5-layer neural networks, with the activation function being ReLU, the optimization algorithm being Adam, the learning rate being 0.002, the batch size being 32, and synthetic reservoir small-sample data is generated. The training set and the synthetic reservoir small-sample data are resampled in terms of distribution. The stratified sampling method is used, with a sampling rate of 1.5 times. Based on the distribution resampling results, the weight ratios of each type of reservoir are corrected, and the correction coefficient is 0.8 - 1.2, to obtain a balanced training data set.

[0045] Through the statistical analysis and clustering of reservoir type samples, the present invention identifies small-sample paragraphs, thereby accurately locating the data imbalance problem and effectively improving the learning ability of the model on minority class samples. The generative adversarial network ensures that the synthetic reservoir small-sample data not only retains key features but also enhances data diversity and representativeness through feature-preserving data generation, providing richer sample information for the training set. By distribution resampling and reservoir weight ratio correction, the problem of uneven distribution of various samples in the training data is solved, effectively avoiding the model's bias towards majority class samples and ensuring balanced improvement in the learning ability of different reservoir types. Overall, the synthetically augmented small-sample data not only improves the generalization ability of the model in small-sample scenarios but also enhances the diversity and representativeness of the training set, providing more stable and efficient data support for the training of subsequent deep learning models.

[0046] Preferably, in step S5, using geological knowledge to constrain the attention network and the balanced training data set to perform deep learning model training, and based on the validation set for hyperparameter optimization and performance monitoring includes: Performing forward propagation and backward propagation training on the balanced training data set based on the geological knowledge-constrained attention network to obtain an initial network weight parameter set; Statistically analyzing the loss function values, validation set prediction errors, and gradient change trends during the training process to generate a dynamic performance index data set; Performing automated hyperparameter search on the geological knowledge-constrained attention network based on the dynamic performance index data set to obtain an optimal hyperparameter combination set; Performing batch training iterations on the balanced training data set based on the optimal hyperparameter combination set to obtain a candidate model parameter set; Performing early stopping method and Dropout validation on the candidate model parameter set based on the validation set to obtain a candidate model.

[0047] In the embodiment of the present invention, first, a geological knowledge-constrained attention network is deployed on a distributed computing cluster (32 nodes, each node is configured with an 8-core CPU, 1 NVIDIA A100 GPU, and 64 GB of system memory). The network parameters are initialized by the Xavier method, the standard deviation of the weight matrix initialization is set to 0.01, and the initial value of the bias term is set to 0.001. The number of nodes in the input layer is the dimension of the complete feature set (including 10 logging curve parameters, 5 core physical property parameters, and 8 reservoir structure parameters). The first convolutional layer uses 32 convolutional kernels of size 3×1, and the activation function is LeakyReLU (negative slope is 0.01). The second convolutional layer uses 64 convolutional kernels of size 5×1, and the activation function is also LeakyReLU. The pooling layer uses a parallel structure of max pooling and average pooling, and the pooling kernel size is 2×1. The number of nodes in the fully connected layers is 512, 256, and 128 in sequence. The attention mechanism layer contains 8 attention heads, and the dimension of the attention heads is 32. The number of nodes in the output layer is the number of reservoir evaluation indicators (including pore connectivity index, reservoir quality factor, lithology control coefficient, and heterogeneity index); then, the balanced training dataset (including 15,000 samples, each sample contains an input feature vector and a target label value) is read, and the data is divided into mini-batches of size 32. During the data loading process, online data augmentation of random horizontal flipping (probability is 0.3) and Gaussian noise perturbation (mean is 0, standard deviation is 0.02) is implemented; then, the training parameters are set. Among them, the momentum parameter of the batch normalization layer is 0.99, the epsilon value is 1e-5, the dropout rate of the Dropout layer is 0.3, and the loss function uses a combined loss of weighted average, including mean squared error loss (weight is 0.5), Huber loss (weight is 0.3, delta value is 1.0), and geological rule consistency loss (weight is 0.2); the forward propagation training process is started. First, the feature data is input into the network through the input layer, local features are extracted through the convolutional layer, the dimension is reduced through the pooling layer and key information is retained, and the feature map is re-weighted through the geological constraint attention layer. The attention calculation formula is Attention(Q,K,V)=softmax(QK^T / √d)V, where Q, K, and V are the query matrix, key matrix, and value matrix respectively, and d is the dimension of the attention head. The output of the attention layer is further processed through the fully connected layer to extract high-order features, and finally, the reservoir evaluation indicator values are predicted through the output layer; after the forward propagation is completed, the loss value is calculated, and performance indicators such as the loss value, gradient norm, and prediction error of each mini-batch are recorded; then, the backpropagation process is executed, the gradient of the loss function with respect to each layer's parameters is calculated, and the Adam optimizer is used during the parameter update process. The initial learning rate is set to 0.001, β1 = 0.9, β2 = 0.999, ε = 1e-8, the weight decay coefficient is 3e-5, the learning rate scheduler adopts the CosineAnnealingLR strategy, the period is 10 epochs, and the minimum learning rate is 1 / 20 of the initial learning rate; after each round of training, the model performance is evaluated on the validation set (3000 samples), and four metrics are calculated: root mean square error (RMSE), mean absolute error (MAE), R² coefficient of determination, and geological rule consistency score (rule compliance rate). These metrics are recorded in the dynamic performance metric dataset; during the training process, an early stopping strategy is implemented, with a patience value of 15 epochs, that is, training stops if there is no improvement in the validation set performance for 15 consecutive epochs, and the maximum number of training epochs is 200; after the basic training is completed, the dynamic performance metric dataset is imported into the Bayesian hyperparameter optimizer, and the hyperparameter search space is defined, including the learning rate range [1e-4, 1e-2], Dropout rate range [0.1, 0.5], L2 regularization coefficient range [1e-6, 1e-3], convolutional kernel size combinations {(3, 5), (3, 7), (5, 7), (5, 9)}, number of attention heads range [4, 8, 12, 16], loss function weight ratio range {(0.4 - 0.6, 0.2 - 0.4, 0.1 - 0.3)}; Gaussian process regression model is used for hyperparameter optimization, the acquisition function is Expected Improvement, the maximum number of evaluations is 50 times, each evaluation includes 5 rounds of complete training, and the hyperparameter combination with the best average performance on the validation set is selected; after determining the optimal hyperparameter combination, the network parameters are reset, and the final training is performed on the balanced training dataset using the optimized hyperparameters. Techniques such as gradient clipping (threshold 5.0), weight normalization (performed every 5 rounds), and learning rate decay (starting from the 50th round, decaying by 10% every 10 rounds) are implemented during the training process to ensure training stability; after training is completed, a candidate model parameter set is generated, including the network parameters of 5 training rounds (the final round, the round with the best validation performance, the round with the best training performance, the round with the highest validation rule consistency, and the round with the best average performance); finally, the candidate model parameter set is evaluated on the validation set, and the integrated voting mechanism is used to select the final model parameters. The specific method is that the prediction results of each model are weighted according to the performance metrics (RMSE weight 0.4, MAE weight 0.2, R² weight 0.2, rule consistency weight 0.2). The model parameters with the highest weighted score are retained as candidate model parameters. At the same time, the Dropout validation technique (50 forward propagations, each using a different random seed) is applied to calculate the confidence interval of the predicted values. The width of the confidence interval does not exceed 15% of the standard deviation of the target value. The final retained candidate model must meet the conditions of the best average performance and a reasonable confidence interval.

[0048] Through the combination of geological knowledge-constrained attention network and balanced training datasets, the present invention ensures the integration of geological knowledge and the balance of data distribution during the training process of the deep learning model, thereby enhancing the model's ability to accurately capture reservoir characteristics. During the training process, through the statistical analysis of the loss function value, validation set error, and gradient change trend, the model performance can be monitored in real time and potential overfitting or underfitting problems can be quickly identified to ensure the stability and convergence during the training process. Automated hyperparameter search makes the adjustment of model hyperparameters more efficient, finds the optimal combination, and thus improves the model's performance and prediction accuracy. Batch training iteration further improves the model's ability to process large-scale data, while early stopping and Dropout validation ensure the generalization ability and stability of the model. Overall, by combining geological knowledge constraints and automated optimization processes, this method significantly improves the accuracy, efficiency, and reliability of the model, providing strong support for the accurate evaluation and development decision-making of complex reservoirs.

[0049] Preferably, step S6 includes the following steps: Step S61: Perform a prediction task on the candidate model using the test set to obtain a set of predicted reservoir parameter values; Step S62: Conduct an error comparison analysis of the true core labels based on the set of predicted reservoir parameter values to obtain a set of model error evaluation metrics; Step S63: Perform parameter fine-tuning and structure fine-tuning on the candidate model based on the set of model error evaluation metrics to obtain an initial reservoir evaluation model; Step S64: Perform integrated nesting on the initial reservoir evaluation model and the preset SZ3 framework to obtain the final reservoir evaluation model.

[0050] In the embodiment of the present invention, firstly, a three-layer cross-validation method is used to perform a reservoir parameter prediction task on a candidate model, and a test set is input into the candidate model. A predicted value set of reservoir porosity, permeability, water saturation and reservoir quality score is obtained by forward propagation calculation. CUDA is used to accelerate the calculation during the prediction process, the batch size is set to 32, and the output result is saved as a 64-bit floating point value. The root mean square error (RMSE), mean absolute error (MAE) and R² determination coefficient are used to quantitatively compare the predicted value with the real core label, and the error index is calculated for each reservoir segment to generate a model error evaluation index set containing the values ​​of each index of each segment. Based on the error evaluation results, parameter fine-tuning is performed in a targeted manner, and the Adam optimizer with a learning rate of 0.0001 is used to perform 1000 iterations of optimization for the relevant weights of the segment with RMSE>0.05, and the number of attention heads (value range 4-16) and the convolution kernel size in the model structure are optimized by the Bayesian optimization algorithm. The initial reservoir evaluation model was obtained by fine-tuning the number of neurons in the fully connected layer (value range: 3-11) and the number of neurons in the fully connected layer (value range: 128-512). The early stopping strategy was adopted in the fine-tuning process. The validation set loss function was stopped if there was no improvement for 10 consecutive rounds, and the initial reservoir evaluation model was obtained. The initial reservoir evaluation model was integrated and nested with the preset SZ3 framework (including lithofacies classification module, physical property prediction module and comprehensive evaluation module), specifically: the reservoir feature vector extracted by the initial model was used as the input of the lithofacies classification module of the SZ3 framework; the attention weight of the initial model was fused with the physical property prediction module of the SZ3 framework at the tensor level, and the fusion coefficient was 0.6:0.4; the prediction results of the initial model and the evaluation results of the SZ3 framework were merged into the final output through weighted averaging (weight of 0.7:0.3) to generate the final reservoir evaluation model, the output format of which is a standardized evaluation report including porosity (accuracy 0.01), permeability (accuracy 0.1mD) and reservoir quality (1-5 points).

[0051] The present invention predicts the candidate model through the test set, and compares the results with the real core labels for error analysis, effectively quantifying the prediction accuracy and error distribution of the model, thereby providing a reliable basis for further optimization. Based on the error evaluation index set, the model can perform parameter fine-tuning and structural fine-tuning to ensure higher accuracy and stronger robustness of the prediction results. The initial reservoir evaluation model obtained through preliminary tuning is integrated and nested with the preset SZ3 framework, which not only integrates the efficient prediction ability of the model, but also further strengthens the logical consistency of geological constraints and reservoir evaluation, and finally generates a more adaptable, accurate and interpretable reservoir evaluation model. This method effectively improves the intelligence level and prediction accuracy of reservoir evaluation, and provides efficient and reliable technical support for development decisions of complex reservoirs.

[0052] Preferably, the present invention further provides a construction system for a reservoir evaluation model based on deep learning, which is used to execute the above-mentioned method for constructing a reservoir evaluation model based on deep learning. The construction system for the reservoir evaluation model based on deep learning includes: A data preprocessing module, which is used to obtain well logging data and core analysis data of a target work area, and perform outlier processing and missing value filling respectively to obtain complete well logging data and complete core data; A pore parameter inversion module, which is used to calculate reservoir pore structure parameters based on the complete core data; and invert the macroscopic porosity of the reservoir based on the complete well logging data; A reservoir property analysis module, which is used to analyze the physical properties of reservoir lithofacies based on the reservoir pore structure parameters and the macroscopic porosity of the reservoir, determine the pore connectivity index, reservoir quality factor, lithology control coefficient and heterogeneity index, and generate a reservoir evaluation index dataset; A geological attention model construction module, which is used to divide the reservoir evaluation index dataset into a training set, a validation set and a test set according to layers; use the training set to construct the basic structure of a one-dimensional convolutional neural network model, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; A model training and enhancement module, which is used to guide the geological attention enhanced neural network with weak supervision signals to obtain a geological knowledge constrained attention network; expand the data synthesis of small sample reservoir accumulation for the training set through a generative adversarial network to obtain a balanced training dataset; use the geological knowledge constrained attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; A model evaluation and integration module, which is used to evaluate and optimize the performance of the candidate model using the test set to obtain an initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

[0053] Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is not limited by the above description. Therefore, it is intended to include all changes falling within the meaning and scope of the equivalent elements of the application documents within the present invention.

[0054] The above are only specific embodiments of the present invention, which enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A construction method of a reservoir evaluation model based on deep learning, characterized in that, It includes the following steps: Step S1: Obtain the logging data and core analysis data of the target work area, and perform outlier processing and missing value filling respectively to obtain complete logging data and complete core data; Step S2: Calculate the reservoir pore structure parameters based on the complete core data; Invert the macroscopic porosity of the reservoir based on the complete logging data; Step S3: Analyze the physical properties of the reservoir lithofacies based on the reservoir pore structure parameters and the macroscopic porosity of the reservoir, determine the pore connectivity index, reservoir quality factor, lithology control coefficient and heterogeneity index, and generate a reservoir evaluation index dataset; Step S4: Divide the reservoir evaluation index dataset into a training set, a validation set and a test set by stratification; use the training set to construct the basic structure of a one-dimensional convolutional neural network model, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; Step S5: Perform weak supervision signal guidance on the geological attention enhanced neural network to obtain a geological knowledge constrained attention network; Perform data synthesis and augmentation of small-sample reservoir accumulation on the training set through a generative adversarial network to obtain a balanced training dataset; use the geological knowledge constrained attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; Step S6: Use the test set to evaluate and optimize the performance of the candidate model to obtain an initial reservoir evaluation model; Integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

2. The construction method of the reservoir evaluation model based on deep learning according to claim 1, wherein Step S1 includes the following steps: Step S11: Collect natural gamma through a sodium iodide crystal detector, and collect acoustic travel time, density, neutron porosity and resistivity through a comprehensive logging tool to generate logging data; Step S12: Collect porosity, permeability and water saturation through a core porosity and permeability measuring instrument to generate core analysis data; Step S13: Perform depth matching on the logging data and the core analysis data, and perform outlier processing and missing value filling respectively to obtain complete logging data and complete core data.

3. The construction method of the reservoir evaluation model based on deep learning according to claim 1, characterized in that, The calculation of the reservoir pore structure parameters based on the complete core data in Step S2 includes: Perform CT scanning on the complete core data to generate a three-dimensional digital model of the core; Adaptive threshold segmentation of the reservoir pore space based on the three-dimensional digital model of the core to obtain three-dimensional distribution data of the pore space; Extract pore geometry-topological features from the three-dimensional distribution data of the pore space, and perform cluster analysis on the results of the pore geometry-topological feature extraction to obtain a pore type classification result; Calculate the lithology-dependent scale box dimension of the three-dimensional distribution data of the pore space to obtain pore fractal dimension feature data; Quantitatively integrate and calculate the volume of each type of pore based on the pore type classification result to obtain micro-porosity distribution data; Perform the calculation of the maximum inscribed sphere on the pore type classification result to obtain pore throat radius distribution data; Perform tensor field analysis on the pore geometry based on the pore type classification result to obtain reservoir pore geometry parameters.

4. The construction method of the reservoir evaluation model based on deep learning according to claim 1, wherein The inversion of the macroscopic porosity of the reservoir based on the complete logging data in Step S2 includes: Perform multi-sensor acquisition accuracy calibration and unified standardization processing on the complete logging data to obtain standardized logging curve data; Perform multivariate non - linear covariance analysis on the standardized log curve data to generate log response characteristic parameters; Analyze the reservoir fluid properties based on the log response characteristic parameter spectrum, and perform reservoir fluid phase separation based on the spectrum analysis results to obtain fluid type distribution data; Perform fluid - constrained neutron - density porosity joint inversion on the standardized log curve data and the fluid type distribution data to obtain initial effective porosity data; Perform X - ray diffraction on the clay mineral content in the reservoir based on the initial effective porosity data and correct it to obtain a clay correction coefficient; Perform numerical integration on the initial effective porosity data and the clay correction coefficient, and perform volume averaging on the numerical integration results to obtain reservoir macroscopic porosity data.

5. The construction method of the reservoir evaluation model based on deep learning according to claim 1, characterized in that Step S3 includes the following steps: Step S31: Evaluate the three - dimensional network connectivity of the reservoir pore geometric morphology parameters and the pore throat radius distribution data to obtain a pore network topology relationship diagram; Trace the seepage path based on the pore network topology relationship diagram to obtain effective fluid channel data; Perform numerical integration and statistical normalization processing on the effective fluid channel data to obtain a pore connectivity index; Step S32: Qualitatively analyze the microscopic pore structure based on the pore fractal dimension characteristic data scale to obtain an initial score of pore complexity; Weight - fuse the initial score of pore complexity and the microporosity distribution data to obtain an initial pore structure complexity index; Verify and calibrate the initial pore structure complexity index through a mercury intrusion porosimeter for porous media to obtain a corrected pore structure complexity index; Step S33: Construct a reservoir fluid storage capacity evaluation model based on the reservoir macroscopic porosity data and the pore connectivity index to generate a reservoir capacity factor; Analyze the correlation between the permeability data and the pore connectivity index in the complete core data to obtain a seepage capacity factor; Perform multi - objective optimization integration on the reservoir capacity factor and the seepage capacity factor to obtain a reservoir quality factor; Step S34: Analyze the rock - fluid interaction based on the fluid type distribution data and the reservoir macroscopic porosity data to obtain a wettability evaluation result; Perform multi - variable regression on the wettability evaluation result and the corrected pore structure complexity index to obtain a lithology control coefficient; Identify the spatial variability based on the reservoir pore geometric morphology parameters and the pore type classification results to generate a spatial variability coefficient; Perform sequence constraint processing on the spatial variability coefficient to obtain a heterogeneity index; Step S35: Perform analytic hierarchy process weight assignment on the reservoir quality factor, the lithology control coefficient, and the heterogeneity index to obtain an index weight matrix; Weight - fuse each evaluation index based on the index weight matrix to obtain a reservoir comprehensive evaluation score; Quantify and classify the reservoir comprehensive evaluation score to obtain a reservoir evaluation index data set.

6. The construction method of the reservoir evaluation model based on deep learning according to claim 1, characterized in that Step S4 includes the following steps: Step S41: Analyze the feature correlation of the complete data set and perform multicollinearity detection to generate a standardized feature data set; Step S42: Identify reservoir intervals based on the preset geological stratification information for the standardized feature dataset, and divide the reservoir intervals based on the identification results to obtain the labeled interval dataset; divide the labeled interval dataset according to a ratio, where 75% is the training set, 15% is the validation set, and 15% is the test set; Step S43: Use the training set to construct a one-dimensional convolutional neural network model and perform non-linear transformation configuration to obtain the basic network structure; Step S44: Perform Dropout and L2 weight regularization on the basic network structure to obtain an overfitting-preventing enhanced network structure; Step S45: Introduce a self-attention calculation unit into the overfitting-preventing enhanced network structure and perform weight initialization to obtain a geology-sensitive feature-enhanced component; Step S46: Perform weight allocation on the geology-sensitive feature-enhanced component to generate geology key features; perform multi-scale integration of the original convolutional feature application features based on the geology key features to obtain geology-enhanced features; Step S47: Construct skip connections based on the geology-enhanced features to obtain a gradient transfer optimization network structure; Step S48: Perform an attention-guided loss combination design on the gradient transfer optimization network structure to obtain a geology-aware comprehensive loss function; Step S49: Integrate the basic network structure, the geology-sensitive feature-enhanced component, the gradient transfer optimization network structure, and the geology-aware comprehensive loss function into a geology-attention-enhanced neural network through a distributed deep learning framework.

7. The construction method of the reservoir evaluation model based on deep learning according to claim 1, characterized in that The weak supervision signal guidance for the geology-attention-enhanced neural network in Step S5 includes: Obtain a geological constraint rule set; perform rule matching on the training set based on the geological constraint rule set to obtain weak supervision signal labels; Apply the weak supervision signal labels to the geology-attention-enhanced neural network and perform attention weight optimization to generate a weak supervision guidance architecture; Design a geological rule consistency loss term based on the weak supervision guidance architecture to generate a multi-objective loss function; Perform constraint optimization on the multi-objective loss function to obtain a geological knowledge constraint optimizer; Integrate and reconstruct the geology-attention-enhanced neural network and the geological knowledge constraint optimizer to obtain a geological knowledge constraint attention network.

8. The method for constructing a reservoir evaluation model based on deep learning according to claim 1, wherein In Step S5, use the geological knowledge constraint attention network and the balanced training dataset to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set, including: Perform forward propagation and backpropagation training on the balanced training dataset based on the geological knowledge constraint attention network to obtain an initial network weight parameter set; Statistically analyze the loss function values, validation set prediction errors, and gradient change trends during the training process to generate a dynamic performance index dataset; Perform automated hyperparameter search on the geological knowledge constraint attention network based on the dynamic performance index dataset to obtain an optimal hyperparameter combination set; Perform batch training iterations on the balanced training dataset based on the optimal hyperparameter combination set to obtain a candidate model parameter set; Perform early stopping method and Dropout validation on the candidate model parameter set based on the validation set to obtain a candidate model.

9. The method for constructing a reservoir evaluation model based on deep learning according to claim 8, wherein Step S6 includes the following steps: Step S61: Use the test set to perform a prediction task on the candidate model to obtain a set of reservoir parameter prediction values; Step S62: Conduct error comparison and analysis of true core labels based on the predicted reservoir parameter set to obtain a model error evaluation index set; Step S63: Perform parameter fine-tuning and structure fine-tuning on the candidate model based on the model error evaluation index set to obtain an initial reservoir evaluation model; Step S64: Perform integrated nesting on the initial reservoir evaluation model and the preset SZ3 framework to obtain a final reservoir evaluation model.

10. A construction system for a reservoir evaluation model based on deep learning, characterized in that, For implementing the method for constructing a reservoir evaluation model based on deep learning as described in claim 1, the system for constructing the reservoir evaluation model based on deep learning includes: A data preprocessing module, configured to obtain well logging data and core analysis data of a target work area, and perform outlier processing and missing value filling respectively to obtain complete well logging data and complete core data; A pore parameter inversion module, configured to calculate reservoir pore structure parameters based on the complete core data; and invert the reservoir macroscopic porosity based on the complete well logging data; A reservoir property analysis module, configured to analyze the physical properties of reservoir lithofacies based on the reservoir pore structure parameters and the reservoir macroscopic porosity, determine the pore connectivity index, reservoir quality factor, lithology control coefficient and heterogeneity index, and generate a reservoir evaluation index data set; A geological attention model construction module, configured to divide the reservoir evaluation index data set into a training set, a validation set and a test set by layer segment; use the training set to construct the basic structure of a one-dimensional convolutional neural network model, and introduce a geological attention mechanism to obtain a geological attention enhanced neural network; A model training and enhancement module, configured to guide the geological attention enhanced neural network with weak supervision signals to obtain a geological knowledge constrained attention network; expand the data synthesis of small sample reservoir accumulation in the training set through a generative adversarial network to obtain a balanced training data set; use the geological knowledge constrained attention network and the balanced training data set to perform deep learning model training, and perform hyperparameter optimization and performance monitoring based on the validation set to obtain a candidate model; A model evaluation and integration module, configured to perform performance evaluation and tuning on the candidate model using the test set to obtain an initial reservoir evaluation model; integrate the preset SZ3 framework with the initial reservoir evaluation model to generate a final reservoir evaluation model.

Images