Natural gas reserve estimation method and device based on deep learning

Abstract

The invention discloses a natural gas reserve estimation method and device based on deep learning, and relates to the technical field of natural gas resource exploration. The method comprises the steps of collecting multi-source monitoring data of a natural gas extraction work area, and performing data processing on the multi-source monitoring data to obtain a corresponding monitoring three-dimensional data volume model; according to the monitored three-dimensional data volume model, using a reservoir parameter prediction model constructed based on a deep learning algorithm to perform prediction to obtain a corresponding three-dimensional reservoir parameter volume; and according to the three-dimensional reservoir parameter body, estimating the natural gas reserves of the natural gas extraction work area to obtain a corresponding natural gas reserve estimation result. The problems that in the prior art, reservoir parameters are difficult to obtain and low in precision, multi-source data fusion is insufficient, subjectivity is high, and uncertainty quantification is difficult are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural gas resource exploration, and in particular to a natural gas reserve estimation method and device based on deep learning. BACKGROUND

[0002] Natural gas, as a clean and efficient fossil energy, plays a vital role in the global energy structure. Accurate and efficient estimation of natural gas reserves is a core link in the whole life cycle of oil and gas field exploration, development, production and management, which is directly related to investment decision-making, development plan formulation and final economic benefits.

[0003] The traditional method of natural gas reserve estimation mainly relies on the volumetric method. The basic principle of the volumetric method is to calculate the total volume of the reservoir rock, then multiply the key parameters of the reservoir, and correct the volume coefficient and recovery factor to finally obtain the geological reserves.

[0004] However, this method faces many challenges in practical application: 1) Difficulty in obtaining reservoir parameters with low precision: Reservoir key parameters are usually obtained through core analysis and well logging interpretation. Although core analysis data has high precision, it is expensive and only represents a very small range near the well point. Although well logging interpretation data covers the entire wellbore, the establishment of the interpretation model and the selection of parameters often depend on expert experience, and there are multiple solutions and uncertainties. How to accurately extrapolate the "one-sided view" of the well point to the entire three-dimensional space is a big problem for traditional methods.

[0005] 2) Insufficient multi-source data fusion: In natural gas exploration, data from different data sources are usually collected. Traditional methods have relatively simple means when fusing these multi-source, multi-scale and heterogeneous data, mostly simple co-Kriging interpolation or random simulation based on geostatistics, which is difficult to fully exploit the inherent complex nonlinear relationship between data, resulting in limited precision of the reservoir parameter model established.

[0006] 3) Strong subjectivity and difficulty in quantifying uncertainty: The traditional method highly depends on the experience and judgment of geologists in many aspects. This subjectivity not only may cause deviation of the estimation results, but also makes it extremely difficult to scientifically quantify the uncertainty of the estimation results. Decision makers can only get a single reserve value, but cannot understand the reliability and possible fluctuation range of the value, which brings great risk to decision-making. SUMMARY

[0007] The present application provides a natural gas reserve estimation method and device based on deep learning, which solves the problems of difficulty in obtaining reservoir parameters with low precision, insufficient multi-source data fusion and strong subjectivity and difficulty in quantifying uncertainty in the prior art.

[0008] In a first aspect, the embodiments of the present application provide a natural gas reserve estimation method based on deep learning, which comprises the following steps: Collecting multi-source monitoring data of a natural gas mining work area, and performing data processing on the multi-source monitoring data to obtain a corresponding monitoring three-dimensional data volume model; According to the monitoring three-dimensional data volume model, using a reservoir parameter prediction model constructed based on a deep learning algorithm, performing prediction to obtain a corresponding three-dimensional reservoir parameter volume; According to the three-dimensional reservoir parameter volume, estimating the natural gas reserves of the natural gas mining work area to obtain a corresponding natural gas reserve estimation result.

[0009] The technical scheme provided by the embodiments of the present application at least brings the following beneficial effects: By constructing a unified monitoring three-dimensional data volume model, the depth fusion of multi-source heterogeneous data such as seismic, logging, and core in the depth domain and unified grid is realized, which lays a high-quality data foundation for subsequent intelligent prediction; the reservoir parameter prediction model constructed based on the deep learning algorithm takes into account both local detail features and global geological rules, and is geologically constrained through a composite loss function, which significantly improves the prediction accuracy and spatial rationality of key reservoir parameters; the entire estimation process is automatically completed by the data-driven deep learning model, which minimizes the dependence on expert experience, improves the objectivity and repeatability of reserve estimation, and through model integration (multiple predictions) and Monte Carlo simulation, the final output is no longer a single reserve value, but a probability distribution of the reserve, which enables decision makers to clearly understand the possible range and confidence level of the reserve, thereby making more scientific risk assessment and management.

[0010] In an optional implementation manner, the multi-source monitoring data comprises a three-dimensional post-stack seismic data volume of the natural gas mining work area, logging curve data of a mining well, and core analysis data.

[0011] In an optional implementation manner, the multi-source monitoring data of the natural gas mining work area is collected, and the multi-source monitoring data is processed to obtain the monitoring three-dimensional data volume model, which comprises the following steps: The multi-source monitoring data of the natural gas mining work area is collected, and the multi-source monitoring data is standardized processed according to a preset data cataloging and metadata standard to obtain the multi-source monitoring data after the standardized processing; The data of different data sources in the multi-source monitoring data after the standardized processing is independently cleaned, and the generated adversarial network is used to repair the logging curve data after the data cleaning to obtain the multi-source monitoring data after the data cleaning; The multi-source monitoring data after the data cleaning is spatio-temporally aligned, and the multi-source monitoring data after the spatio-temporal alignment is grid divided to obtain the corresponding monitoring three-dimensional data volume model.

[0012] In one optional implementation, the cleaned multi-source monitoring data is spatiotemporally aligned, and the resulting spatiotemporally aligned multi-source monitoring data is meshed to obtain a corresponding 3D monitoring data volume model, including: Based on the cleaned 3D post-stack seismic data volume from the multi-source monitoring data, a structured mesh generation method is used to construct a 3D structural model mesh. Then, through resampling technology, the seismic data in the cleaned 3D post-stack seismic data volume is matched to the corresponding 3D structural model mesh to obtain the initial monitoring 3D data volume model. Based on a pre-constructed high-precision three-dimensional velocity field model, a time-depth conversion algorithm is used to convert the time-domain seismic data of each grid in the initial monitoring three-dimensional data volume model into depth-domain seismic data. The repaired well logging curve data and the cleaned core analysis data from the multi-source monitoring data are aligned and mapped with the seismic data in the depth domain of each grid in the initial monitoring three-dimensional data volume model to obtain the intermediate monitoring three-dimensional data volume model. The data vectors of each grid in the intermediate monitoring 3D data volume model are normalized to obtain the final monitoring 3D data volume model, which uses a four-dimensional tensor to represent the monitoring data vectors of the grid.

[0013] In one alternative implementation, the reservoir parameter prediction model is constructed based on the U-Net-Attention-Transformer-MLP algorithm. The reservoir parameter prediction model includes a backbone network constructed based on the U-Net algorithm, a channel-spatial dual attention module constructed based on the attention mechanism set at the skip connections of the backbone network, a global encoder constructed based on the Transformer algorithm connected to the end of the backbone network, a global-local fusion module constructed based on the attention mechanism set at the end of the backbone network and the end of the global encoder, and a reservoir parameter prediction module constructed based on the MLP algorithm connected to the end of the global encoder. The backbone network is equipped with a Dropout mechanism.

[0014] In one optional implementation, monitoring data vectors of several grids are extracted from the monitoring three-dimensional data volume model, and corresponding reservoir parameter preset labels are set to obtain a training sample set. The training sample set is then used to optimize and train the monitoring three-dimensional data volume model. The optimization goal is to obtain the optimal model parameters of the monitoring three-dimensional data volume model. The total loss function of the monitoring three-dimensional data volume model includes a prediction error loss function, a structural similarity loss function, and a geological constraint loss function.

[0015] In one optional implementation, based on the monitored 3D data volume model, a reservoir parameter prediction model constructed using a deep learning algorithm is used to predict and obtain the corresponding 3D reservoir parameter volume, including: The monitoring three-dimensional data volume model, which includes monitoring data vectors from several grids, is input into the reservoir parameter prediction model constructed based on deep learning algorithms. The encoder of the backbone network of the reservoir parameter prediction model is used to extract high-resolution features of the monitoring data vector, and the corresponding decoder is used to extract high semantic features of the monitoring data vector. Based on the channel-space dual attention weights, the channel-space dual attention module of the reservoir parameter prediction model is used to perform weighted fusion of high-resolution features and high semantic features to obtain the corresponding spatial context features. The global encoder of the reservoir parameter prediction model is used to extract global features from the monitoring data vector. Based on the dynamically generated weighted fusion attention weights, the global-local fusion module of the reservoir parameter prediction model is used to fuse the spatial context features and global features to obtain fused features. Based on the fusion characteristics, the reservoir parameter prediction module of the reservoir parameter prediction model is used to make predictions and obtain the reservoir parameter prediction values ​​for the corresponding grid. By traversing the monitoring data vectors of all grids in the monitoring three-dimensional data volume model, an initial three-dimensional reservoir parameter volume is obtained, which is composed of the reservoir parameter prediction values ​​of all grids. Repeat the above steps several times, and integrate the reservoir parameter prediction values ​​of each grid in the obtained initial three-dimensional reservoir parameter volumes to obtain the final three-dimensional reservoir parameter volume including the reservoir parameter prediction value sequence of each grid.

[0016] In one alternative implementation, the key reservoir parameters predicted for the reservoir parameters include porosity, permeability, and gas saturation.

[0017] In one optional implementation, the natural gas reserves of the natural gas extraction area are estimated based on the three-dimensional reservoir parameter volume, resulting in the corresponding natural gas reserve estimation results, including: In the final three-dimensional reservoir parameter volume, a reservoir parameter prediction value is randomly selected from the reservoir parameter prediction value sequence of each grid to form a three-dimensional parameter set covering all grids. Based on the three-dimensional parameter set, the volumetric method reserve calculation formula is used to calculate the natural gas reserves and obtain the estimated value of natural gas reserves. Repeat the above steps several times to obtain several natural gas reserve estimates, generate natural gas reserve probability distributions corresponding to several natural gas reserve estimates, and extract the natural gas reserve estimation results corresponding to the natural gas reserve probability distributions.

[0018] Secondly, embodiments of the present invention provide a natural gas reserve estimation device based on deep learning, used to implement a natural gas reserve estimation method, the device comprising: The multi-source monitoring data processing unit is used to collect multi-source monitoring data from the natural gas extraction area and process the multi-source monitoring data to obtain the corresponding three-dimensional monitoring data volume model. The reservoir parameter prediction unit is used to predict the corresponding three-dimensional reservoir parameter volume based on the monitoring three-dimensional data volume model and the reservoir parameter prediction model constructed based on the deep learning algorithm. The natural gas reserve estimation unit is used to estimate the natural gas reserves in a natural gas extraction area based on the three-dimensional reservoir parameter volume, and obtain the corresponding natural gas reserve estimation results.

[0019] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the steps of a deep learning-based natural gas reserve estimation method provided in an embodiment of the present invention. Figure 3 This is a functional unit diagram of a natural gas reserve estimation device based on deep learning provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0023] The present invention will be further described below with reference to the accompanying drawings.

[0024] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0025] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0026] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0027] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating device, a data storage module, a network communication module, a user interface module, and electronic programs.

[0028] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the deep learning-based natural gas reserve estimation device stored in the memory 1005 through the processor 1001 and executes the deep learning-based natural gas reserve estimation method provided in the embodiment of the present invention.

[0029] Reference Figure 2 The present invention provides a deep learning-based method for estimating natural gas reserves, the method comprising: S201: Collect multi-source monitoring data from the natural gas extraction area, process the multi-source monitoring data, and obtain the corresponding three-dimensional monitoring data volume model; S202: Based on the monitoring three-dimensional data volume model, use the reservoir parameter prediction model built based on deep learning algorithm to make predictions and obtain the corresponding three-dimensional reservoir parameter volume; S203: Based on the three-dimensional reservoir parameter volume, the natural gas reserves in the natural gas extraction area are estimated to obtain the corresponding natural gas reserve estimation results.

[0030] By constructing a unified 3D monitoring data volume model, deep fusion of multi-source heterogeneous data such as seismic, well logging, and core samples was achieved in the depth domain under a unified grid, laying a high-quality data foundation for subsequent intelligent prediction. The reservoir parameter prediction model built based on deep learning algorithms takes into account both local detailed features and global geological laws, and uses a composite loss function for geological constraints, significantly improving the prediction accuracy and spatial rationality of key reservoir parameters. The entire estimation process is automatically completed by a data-driven deep learning model, minimizing reliance on expert experience and improving the objectivity and repeatability of reserve estimation. Through model integration (multiple predictions) and Monte Carlo simulation, the final output is no longer a single reserve value, but a probability distribution of reserves. This allows decision-makers to clearly understand the possible range and confidence level of reserves, thereby enabling more scientific risk assessment and management.

[0031] In one alternative implementation, the multi-source monitoring data includes three-dimensional post-stack seismic data volumes of the natural gas extraction area, well logging data of the production wells, and core analysis data.

[0032] It is worth noting that, using commercial seismic interpretation software, 20 seismic attributes, including root mean square amplitude, instantaneous frequency, coherence volume, and curvature, were extracted from the original seismic data volume. Under a unified grid system, the 20 seismic attribute volumes, acoustic impedance inversion volumes, and lithology coding volumes were stacked along the depth direction to obtain a three-dimensional post-stack seismic data volume. Well logging curve data included gamma, sonic transit time, density, resistivity, etc. Core analysis data consisted of physical property analysis data from a 500-meter core, including porosity analysis data, permeability analysis data, and gas saturation analysis data.

[0033] In one optional implementation, multi-source monitoring data of the natural gas extraction area is collected, and the multi-source monitoring data is processed to obtain a three-dimensional monitoring data volume model, including: S2011: Collect multi-source monitoring data from natural gas extraction areas, and standardize the multi-source monitoring data according to the preset data cataloging and metadata standards to obtain standardized multi-source monitoring data; It is worth noting that the data cataloging and metadata standards define standard physical units, data formats, and metadata tags (such as collection time, processing parameters, and coordinate systems) for each data type, ensuring the standardization and traceability of subsequent processing. S2012: Data from different data sources in the standardized multi-source monitoring data are cleaned independently, and generative adversarial networks are used to repair the cleaned well logging curve data to obtain cleaned multi-source monitoring data. It is worth noting that data cleaning includes environmental correction and depth correction of well logging data, denoising (such as using fx domain denoising or deep learning denoising networks) and amplitude compensation of 3D post-stack seismic data to improve the signal-to-noise ratio, and normalization of core analysis data to eliminate systematic errors caused by different laboratory analyses. The introduction of intelligent repair of well logging data based on generative adversarial networks, compared with traditional interpolation, can learn the nonlinear variation law of well logging curves under complex geological backgrounds. The repair results are more consistent with the geological reality and provide higher quality hard data constraints for subsequent models. S2013: Perform spatiotemporal alignment on the multi-source monitoring data after data cleaning, and divide the spatiotemporally aligned multi-source monitoring data into grids to obtain the corresponding monitoring three-dimensional data volume model; It is worth noting that spatiotemporal alignment and mesh generation ensured the computational accuracy of critical regions while controlling the overall computational load, achieving a balance between accuracy and efficiency.

[0034] In one optional implementation, the cleaned multi-source monitoring data is spatiotemporally aligned, and the resulting spatiotemporally aligned multi-source monitoring data is meshed to obtain a corresponding 3D monitoring data volume model, including: S20131: Based on the cleaned three-dimensional post-stack seismic data volume from the multi-source monitoring data after data cleaning, a three-dimensional structural model mesh is constructed using a structured mesh generation method. Then, through resampling technology, the seismic data in the cleaned three-dimensional post-stack seismic data volume is matched to the corresponding three-dimensional structural model mesh to obtain the initial monitoring three-dimensional data volume model. S20132: Based on the pre-constructed high-precision three-dimensional velocity field model, the time-depth conversion algorithm is used to convert the time-domain seismic data of each grid in the initial monitoring three-dimensional data volume model into the depth-domain seismic data. S20133: Align and map the repaired well logging curve data and the cleaned core analysis data from the multi-source monitoring data with the seismic data in the depth domain of each grid in the initial monitoring three-dimensional data volume model to obtain an intermediate monitoring three-dimensional data volume model. S20134: Normalize the data vector of each grid in the intermediate monitoring 3D data volume model to obtain the final monitoring 3D data volume model in which the monitoring data vector of the grid is represented by a four-dimensional tensor. In this embodiment, the four-dimensional tensor is , Representing three-dimensional spatial coordinates, these three dimensions have clear spatial continuity and local correlation, meaning that the attributes of a grid point are highly correlated with the attributes of its neighboring grid points. F The tensor, which consists of seismic data representing the depth domain, repaired well logging data, and core analysis data after data cleaning, is a multi-dimensional data vector and serves as the input to the deep learning model.

[0035] In one alternative implementation, the reservoir parameter prediction model is constructed based on the U-Networks (U-Net)-Attention-Transformer-Multi-Layer Perceptron (MLP) algorithm. The reservoir parameter prediction model includes a backbone network constructed based on the U-Net algorithm, a channel-spatial dual attention module constructed based on the attention mechanism set at the skip connections of the backbone network, a global encoder constructed based on the Transformer algorithm connected to the end of the backbone network, a global-local fusion module constructed based on the attention mechanism set at the end of the backbone network and the end of the global encoder, and a reservoir parameter prediction module constructed based on the MLP algorithm connected to the end of the global encoder. The backbone network is equipped with a Dropout mechanism.

[0036] It is worth noting that the backbone network focuses on processing three-dimensional spatial data formed by channel dimension projection or unfolding. Its core task is to capture the local structure, morphology, and boundary information of reservoir parameters in three-dimensional space, such as identifying the morphology of channel sand bodies, the location of faults, and the contact surface between reservoirs and non-reservoirs. It linearly or nonlinearly combines the features of the monitoring data vectors, compressing or fusing them into one or more "pseudo-seismic attribute" three-dimensional volumes. This new feature map initially integrates multi-source information but retains the complete three-dimensional spatial structure. The encoder (downsampling path) of the backbone network consists of multiple "convolutional layers +..." The network consists of an activation function + max pooling block. By progressively increasing the receptive field, it extracts hierarchical spatial features from the input, ranging from low-level (semantic features) to high-level (geological body morphology, structure). Each downsampling step yields a higher-level but lower-spatial-resolution 3D feature. The decoder (upsampling path) gradually restores spatial resolution through an upsampling (e.g., transposed convolution) + skip connections + convolutional layers structure. The key skip connections concatenate high-resolution, low-level features from the encoder with low-resolution, high-level features from the decoder, enabling the network to accurately recover the spatial details and boundaries of the reservoir. The final output of the backbone network is rich, finely processed spatial context features. The global encoder receives features from the backbone network output and captures global dependencies through a self-attention mechanism to generate a global feature representation. A channel-spatial dual attention module is used for fusion encoding. The system incorporates features from both the encoder and decoder. Channel attention weights the importance of different attribute channels, highlighting key attributes (such as those strongly indicative of gas content), while spatial attention weights the importance of spatial location, highlighting key reservoir regions (such as high-porosity permeable zones). This dual attention mechanism enables adaptive selection of both attributes and space. A global-local fusion module fuses the spatial context features output from the backbone network and the global features output from the global encoder. A reservoir parameter prediction module receives the fused features, maps the high-dimensional features to the reservoir parameter space through fully connected layers, and outputs the predicted reservoir parameters for each grid. The Dropout mechanism enhances the model's generalization ability, preventing overfitting due to limited training data or excessive model complexity, thus ensuring high accuracy and stable prediction performance when processing new, unseen real-world data.

[0037] In one optional implementation, monitoring data vectors of several grids are extracted from the monitoring three-dimensional data volume model, and corresponding reservoir parameter preset labels are set to obtain a training sample set. The training sample set is then used to optimize and train the monitoring three-dimensional data volume model. The optimization goal is to obtain the optimal model parameters of the monitoring three-dimensional data volume model. The total loss function of the monitoring three-dimensional data volume model includes a prediction error loss function, a structural similarity loss function, and a geological constraint loss function. It is worth noting that geological constraints refer to the introduction of geological principles, geological parameters, and geological laws into the process of reserve estimation to limit or correct the model output results, ensuring that the estimation results conform to geological reality. Geological constraints include material balance constraints, volumetric constraints, spatial continuity constraints, and permeability-porosity relationship constraints. By incorporating the geological constraint formula as part of the loss function, the total loss function can significantly improve the spatial structure rationality and geological credibility of the prediction results while ensuring prediction accuracy, thereby training a high-performance reservoir parameter prediction model that is truly suitable for actual geological exploration and development.

[0038] In one optional implementation, based on the monitored 3D data volume model, a reservoir parameter prediction model constructed using a deep learning algorithm is used to predict and obtain the corresponding 3D reservoir parameter volume, including: S2021: Input the monitoring three-dimensional data volume model, which includes monitoring data vectors from several grids, into the reservoir parameter prediction model constructed based on deep learning algorithms; S2022: The encoder of the backbone network of the reservoir parameter prediction model is used to extract high-resolution features of the monitoring data vector, and the corresponding decoder is used to extract high semantic features of the monitoring data vector. S2023: Based on the channel-space dual attention weight, the channel-space dual attention module of the reservoir parameter prediction model is used to perform weighted fusion of high-resolution features and high semantic features to obtain the corresponding spatial context features. S2024: Using the global encoder of the reservoir parameter prediction model, extract global features from the monitoring data vector, and use the global-local fusion module of the reservoir parameter prediction model to fuse spatial context features and global features according to dynamically generated weighted fusion attention weights to obtain fused features; S2025: Based on the fusion characteristics, use the reservoir parameter prediction module of the reservoir parameter prediction model to make predictions and obtain the corresponding reservoir parameter prediction values ​​for the grid. S2026: Traverse the monitoring data vectors of all grids in the monitoring three-dimensional data volume model to obtain the initial three-dimensional reservoir parameter volume composed of the reservoir parameter prediction values ​​of all grids; Repeat the above steps several times, and integrate the reservoir parameter prediction values ​​of each grid in the obtained initial three-dimensional reservoir parameter volumes to obtain the final three-dimensional reservoir parameter volume including the reservoir parameter prediction value sequence of each grid.

[0039] In one alternative implementation, the key reservoir parameters predicted for the reservoir parameters include porosity, permeability, and gas saturation.

[0040] In one optional implementation, the natural gas reserves of the natural gas extraction area are estimated based on the three-dimensional reservoir parameter volume, resulting in the corresponding natural gas reserve estimation results, including: S2031: In the final three-dimensional reservoir parameter volume, a reservoir parameter prediction value is randomly selected from the reservoir parameter prediction value sequence of each grid to form a three-dimensional parameter set covering all grids. S2032: Based on the three-dimensional parameter set, the volumetric method of reserve calculation is used to calculate the natural gas reserves and obtain the estimated value of natural gas reserves; The formula is: In the formula, For the first j One estimated natural gas reserve; For the first i The predicted values ​​of volume, porosity, and gas saturation of each grid cell; For the first i The natural gas volume factor for each grid (usually a constant or varying with pressure); j This is an indicator of the number of times volumetric storage calculations are performed. i For grid indication; M Total number of grid cells; In the formula, For the first j One ultimate recoverable natural gas reserve; This is a function of the recovery rate; This is a predicted penetration rate. Permeability is not usually directly involved in calculations, but in practical applications, it is crucial for assessing the recoverability and economic viability of reserves. Permeability can help models better identify reservoir heterogeneity, improve the prediction accuracy of porosity and saturation, and be used to generate the corresponding probability distribution of natural gas reserves. S2033: Repeat the above steps several times to obtain several natural gas reserve estimates, generate a natural gas reserve probability distribution corresponding to the several natural gas reserve estimates, and extract the natural gas reserve estimation results corresponding to the natural gas reserve probability distribution. It is worth noting that the natural gas reserve estimation results include key indicators such as P10 (optimistic estimate), P50 (best estimate), and P90 (conservative estimate) of the probability distribution of natural gas reserves, realizing a complete and explicit transmission from parameter uncertainty to reserve uncertainty.

[0041] This invention also provides a deep learning-based natural gas reserve estimation device, referring to... Figure 3The diagram shows a functional unit diagram of a natural gas reserve estimation device 300 based on deep learning according to the present invention. The device may include the following units: The multi-source monitoring data processing unit 301 is used to collect multi-source monitoring data from the natural gas extraction area and process the multi-source monitoring data to obtain the corresponding three-dimensional monitoring data volume model. The reservoir parameter prediction unit 302 is used to predict the corresponding three-dimensional reservoir parameter volume based on the monitoring three-dimensional data volume model and the reservoir parameter prediction model constructed based on the deep learning algorithm. The natural gas reserve estimation unit 303 is used to estimate the natural gas reserves in the natural gas extraction area based on the three-dimensional reservoir parameter volume, and obtain the corresponding natural gas reserve estimation results.

[0042] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the deep learning-based natural gas reserve estimation method of the present invention.

[0043] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0044] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0045] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the deep learning-based natural gas reserve estimation method of the present invention.

[0046] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable vehicles (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0047] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0048] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxesFigure 1 The function specified in one or more boxes.

[0049] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0050] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "" and / or "" indicate that either one or both can be selected. Furthermore, the terms "includes," "contains," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the statement "includes a..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0051] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for estimating natural gas reserves based on deep learning, characterized in that, The method includes: Collect multi-source monitoring data from the natural gas extraction area, and process the multi-source monitoring data to obtain the corresponding three-dimensional monitoring data volume model; Based on the monitoring three-dimensional data volume model, a reservoir parameter prediction model constructed based on deep learning algorithm is used to make predictions and obtain the corresponding three-dimensional reservoir parameter volume. Based on the three-dimensional reservoir parameter volume, the natural gas reserves in the natural gas extraction area are estimated, and the corresponding natural gas reserve estimation results are obtained.

2. The deep learning-based natural gas reserve estimation method according to claim 1, characterized in that, The multi-source monitoring data includes three-dimensional post-stack seismic data volumes of the natural gas extraction area, well logging curve data of the extraction wells, and core analysis data.

3. The deep learning-based natural gas reserve estimation method according to claim 2, characterized in that, Multi-source monitoring data was collected from the natural gas extraction area, and the data was processed to obtain a three-dimensional monitoring data volume model, including: Collect multi-source monitoring data from the natural gas extraction area, and standardize the multi-source monitoring data according to the preset data cataloging and metadata standards to obtain standardized multi-source monitoring data; Data from different data sources in the standardized multi-source monitoring data were cleaned independently, and generative adversarial networks were used to repair the cleaned well logging curve data to obtain cleaned multi-source monitoring data. After data cleaning, the multi-source monitoring data is spatiotemporally aligned, and the resulting spatiotemporally aligned multi-source monitoring data is divided into grids to obtain the corresponding three-dimensional monitoring data volume model.

4. The deep learning-based natural gas reserve estimation method according to claim 3, characterized in that, After data cleaning, the multi-source monitoring data is spatiotemporally aligned, and the resulting spatiotemporally aligned multi-source monitoring data is then meshed to obtain the corresponding 3D monitoring data volume model, including: Based on the cleaned 3D post-stack seismic data volume from the multi-source monitoring data, a structured mesh generation method is used to construct a 3D structural model mesh. Then, through resampling technology, the seismic data in the cleaned 3D post-stack seismic data volume is matched to the corresponding 3D structural model mesh to obtain the initial monitoring 3D data volume model. Based on a pre-constructed high-precision three-dimensional velocity field model, a time-depth conversion algorithm is used to convert the time-domain seismic data of each grid in the initial monitoring three-dimensional data volume model into depth-domain seismic data. The repaired well logging curve data and the cleaned core analysis data from the multi-source monitoring data are aligned and mapped with the seismic data in the depth domain of each grid in the initial monitoring three-dimensional data volume model to obtain the intermediate monitoring three-dimensional data volume model. The data vectors of each grid in the intermediate monitoring 3D data volume model are normalized to obtain the final monitoring 3D data volume model, which uses a four-dimensional tensor to represent the monitoring data vectors of the grid.

5. The deep learning-based natural gas reserve estimation method according to claim 4, characterized in that, The reservoir parameter prediction model is constructed based on the U-Net-Attention-Transformer-MLP algorithm. The reservoir parameter prediction model includes a backbone network constructed based on the U-Net algorithm, a channel-spatial dual attention module constructed based on the attention mechanism set at the skip connections of the backbone network, a global encoder constructed based on the Transformer algorithm connected to the end of the backbone network, a global-local fusion module constructed based on the attention mechanism set at the end of the backbone network and the end of the global encoder, and a reservoir parameter prediction module constructed based on the MLP algorithm connected to the end of the global encoder. The backbone network is equipped with a Dropout mechanism.

6. The deep learning-based natural gas reserve estimation method according to claim 5, characterized in that, The monitoring data vectors of several grids are extracted from the monitoring three-dimensional data volume model, and corresponding reservoir parameter preset labels are set to obtain a training sample set. The training sample set is then used to optimize and train the monitoring three-dimensional data volume model. The optimization goal is to obtain the optimal model parameters of the monitoring three-dimensional data volume model. The total loss function of the monitoring three-dimensional data volume model includes the prediction error loss function, the structural similarity loss function, and the geological constraint loss function.

7. The deep learning-based natural gas reserve estimation method according to claim 6, characterized in that, Based on the monitored 3D data volume model, a reservoir parameter prediction model constructed using a deep learning algorithm is used to predict and obtain the corresponding 3D reservoir parameter volume, including: The monitoring three-dimensional data volume model, which includes monitoring data vectors from several grids, is input into the reservoir parameter prediction model constructed based on deep learning algorithms. The encoder of the backbone network of the reservoir parameter prediction model is used to extract high-resolution features of the monitoring data vector, and the corresponding decoder is used to extract high semantic features of the monitoring data vector. Based on the channel-space dual attention weights, the channel-space dual attention module of the reservoir parameter prediction model is used to perform weighted fusion of high-resolution features and high semantic features to obtain the corresponding spatial context features. The global encoder of the reservoir parameter prediction model is used to extract global features from the monitoring data vector. Based on the dynamically generated weighted fusion attention weights, the global-local fusion module of the reservoir parameter prediction model is used to fuse the spatial context features and global features to obtain fused features. Based on the fusion characteristics, the reservoir parameter prediction module of the reservoir parameter prediction model is used to make predictions and obtain the reservoir parameter prediction values ​​for the corresponding grid. By traversing the monitoring data vectors of all grids in the monitoring three-dimensional data volume model, an initial three-dimensional reservoir parameter volume is obtained, which is composed of the reservoir parameter prediction values ​​of all grids. Repeat the above steps several times, and integrate the reservoir parameter prediction values ​​of each grid in the obtained initial three-dimensional reservoir parameter volumes to obtain the final three-dimensional reservoir parameter volume including the reservoir parameter prediction value sequence of each grid.

8. The deep learning-based natural gas reserve estimation method according to claim 7, characterized in that, The key reservoir parameters predicted in the reservoir parameter prediction include porosity, permeability, and gas saturation.

9. The deep learning-based natural gas reserve estimation method according to claim 8, characterized in that, Based on the three-dimensional reservoir parameter volume, the natural gas reserves in the natural gas extraction area are estimated, and the corresponding natural gas reserve estimation results are obtained, including: In the final three-dimensional reservoir parameter volume, a reservoir parameter prediction value is randomly selected from the reservoir parameter prediction value sequence of each grid to form a three-dimensional parameter set covering all grids. Based on the three-dimensional parameter set, the volumetric method reserve calculation formula is used to calculate the natural gas reserves and obtain the estimated value of natural gas reserves. Repeat the above steps several times to obtain several natural gas reserve estimates, generate natural gas reserve probability distributions corresponding to several natural gas reserve estimates, and extract the natural gas reserve estimation results corresponding to the natural gas reserve probability distributions.

10. A deep learning-based natural gas reserve estimation device, used to implement the natural gas reserve estimation method as described in any one of claims 1-9, characterized in that, The device includes: The multi-source monitoring data processing unit is used to collect multi-source monitoring data from the natural gas extraction area and process the multi-source monitoring data to obtain the corresponding three-dimensional monitoring data volume model. The reservoir parameter prediction unit is used to predict the corresponding three-dimensional reservoir parameter volume based on the monitoring three-dimensional data volume model and the reservoir parameter prediction model constructed based on the deep learning algorithm. The natural gas reserve estimation unit is used to estimate the natural gas reserves in a natural gas extraction area based on the three-dimensional reservoir parameter volume, and obtain the corresponding natural gas reserve estimation results.

Images