Method and system for detecting key indexes of deep coal bed gas reservoir

Abstract

The invention provides a method for detecting key indexes of a deep coal bed gas reservoir, which comprises the following steps of: performing differential screening and preprocessing on sampling well sections, and acquiring reservoir data through a multi-field coupling test system to obtain in-situ multi-field original data; the method comprises the following steps: acquiring a reservoir coal sample through pressure-maintaining sampling, and carrying out nano-scale and micron-scale pore characterization to obtain original data of a dual-scale structure; performing data preprocessing on the in-situ multi-field original data and the dual-scale structure original data to obtain a standardized data set; constructing a multi-branch coupling model to perform differentiation operation on the standardized data set to obtain different branch intermediate results; and carrying out hierarchical combination calculation on the branch intermediate result to obtain a reservoir key index. According to the method, through differential screening pretreatment, multi-field coupling data acquisition, dual-scale pore characterization and multi-branch coupling model operation, environment, structure and seepage parameters are accurately coupled, and the detection precision of key indexes of the deep coal bed gas reservoir is improved.

Description

Technical Field

[0001] This invention relates to the field of reservoir evaluation technology, and in particular to a method and system for detecting key indicators of deep coalbed methane reservoirs. Background Technology

[0002] Deep coalbed methane (usually referring to deposits deeper than 3000m) is a clean energy source with abundant reserves. Its efficient exploration and development are of great strategic significance for ensuring energy security and reducing carbon emissions. Accurate detection of key indicators such as total gas content, effective permeability, and stress sensitivity coefficient is a core prerequisite for identifying sweet spots in reservoirs, optimizing extraction schemes, and predicting production capacity, directly determining the economics and feasibility of deep coalbed methane development. As coalbed methane exploration and development extends to deeper layers, the complexity of the reservoir environment increases significantly. The coupling effects of high temperature (100-200℃), high pressure (30-100MPa), high ground stress, and strong heterogeneity place stringent demands on the adaptability and accuracy of detection technologies. Existing detection technologies have formed multiple pathways, such as pressure-holding coring tests, geophysical logging, and laboratory rock sample characterization. For example, gas content data can be obtained through pressure-holding coring, pore structure can be analyzed using nuclear magnetic resonance (NMR) or CT scans, and permeability can be determined by well testing. However, these technologies are mostly developed based on shallow and medium-depth reservoir conditions, and their application in deep, special environments faces many adaptability challenges.

[0003] Currently, the core shortcomings of existing technologies are mainly reflected in three aspects: First, in-situ environmental simulation is distorted. Traditional pressure-holding coring methods lack sealing performance and have weak adaptability to high temperature and high pressure. Laboratory tests cannot accurately reproduce the deep geostress and fluid occurrence state, resulting in significant deviations between data and actual reservoirs. Second, the ability to perform fine characterization is lacking. There are blind spots in the identification and quantification of the main gas storage space (<10nm nanopores) in deep coalbed methane reservoirs, making it impossible to accurately capture the connectivity characteristics of the pore network. Third, multi-parameter coupling analysis is lacking. Existing methods often collect environmental, structural, and seepage data in isolation, ignoring the dynamic correlation between temperature, pressure, stress, and pore structure. This leads to the calculation of key indicators ignoring coupling effects, resulting in large errors and making it difficult to meet the actual needs of accurate evaluation of deep coalbed methane reservoirs. Therefore, it is essential to design a method and system for detecting key indicators of deep coalbed methane reservoirs. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for detecting key indicators of deep coalbed methane reservoirs. Through differentiated screening preprocessing, multi-field coupled data acquisition, dual-scale pore characterization, and multi-branch coupled model calculation, the method accurately couples environmental, structural, and seepage parameters, thereby improving the detection accuracy of key indicators of deep coalbed methane reservoirs.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for detecting key indicators of deep coalbed methane reservoirs includes the following steps: Differential screening and preprocessing were performed on the sampling well sections of the target deep coalbed methane reservoir, and reservoir environmental parameters, physical property parameters and seepage data were collected through a multi-field coupling test system to obtain in-situ multi-field raw data; Reservoir coal samples were obtained by pressure-holding sampling, and the pore size of the reservoir coal samples was characterized at both the nanoscale and microscale to obtain the original data of the dual-scale structure. Outlier removal, normalization, and spatiotemporal alignment were performed on the in-situ multi-field raw data and dual-scale structure raw data to obtain a standardized dataset. Construct a multi-branch coupling model, and perform differentiated computational processing on the standardized dataset based on the multi-branch coupling model to obtain intermediate results of different branches; By performing hierarchical combination calculations on the intermediate results of the branches, key reservoir indicators are obtained.

[0006] Optionally, the sampling well sections of the target deep coalbed methane reservoir are subjected to differentiated screening and preprocessing, and reservoir environmental parameters, physical property parameters, and seepage data are collected through a multi-field coupled testing system to obtain in-situ multi-field raw data, including: The sampling well sections were screened using previous natural gamma logging and density logging data to obtain the target well sections; Casing windows were opened in the target well section, and coal dust was removed from the wellbore wall using high-pressure water jets. Inspect the surface flatness of the target well section after cleaning to ensure that the fluctuation range of the surface flatness does not exceed 2 mm.

[0007] Optionally, the multi-field coupling test system includes: a sensing unit, an environmental control unit, and a sealing unit; the sensing unit incorporates a fiber optic stress sensor, a miniature pressure sensor, a platinum resistance temperature sensor, a high-frequency electromagnetic sensor, an ultrasonic probe, and a fluid flow sensor; the environmental control unit consists of a miniature heating rod and a pressure compensation pump; and the sealing unit is a sealing ring composed of hydrogenated nitrile rubber and a metal skeleton.

[0008] Optionally, reservoir coal samples are obtained through pressure-holding sampling, and the samples are characterized at both nanoscale and microscale pore sizes to obtain raw data on the dual-scale structure, including: Coal samples from the reservoir were obtained by pressure sampling using a diamond drill bit. Cross-sectional processing and scanning of reservoir coal samples yielded nanoscale images; Resonance tests and X-ray scanning were performed on reservoir coal samples to obtain micron-sized images; A three-dimensional model is generated based on nanoscale and microscale images, and the original two-scale structural data containing specific surface area and pore size distribution are calculated using the BET and BJH methods.

[0009] Optionally, outlier removal, normalization, and spatiotemporal alignment are performed on the in-situ multi-field raw data and the dual-scale structured raw data to obtain a standardized dataset, including: The density-based DBSCAN algorithm was used to remove outliers from the in-situ multi-field raw data and the dual-scale structure raw data, and then the data was integrated into outlier-free data. Max-min standardization is performed on the static features in the outlier-removed data, and median standardization is performed on the dynamic data in the outlier-removed data to obtain normalized data. A standardized dataset is obtained by spatiotemporally aligning time-series and discrete data in normalized data using cubic spline interpolation.

[0010] Optionally, a multi-branch coupling model is constructed, and differential computation processing is performed on the standardized dataset based on the multi-branch coupling model to obtain intermediate results for different branches, including: An environmental dynamic response branch is constructed based on the GRU model, and attention weights are adjusted on the time series data in the standardized dataset to obtain environmental correction coefficients. The environmental correction coefficients include stress attenuation coefficients and temperature correction coefficients. A static structural representation branch is constructed based on a convolutional neural network, and features are extracted from the structural parameter data in the standardized dataset to obtain the basic structural parameters, including: basic gas content and basic permeability. Based on the gradient boosting tree model, a seepage coupling correlation branch is constructed, and the linkage data in the standardized dataset are nonlinearly correlated to obtain the seepage efficiency coefficient.

[0011] Optionally, the environmental dynamic response branch includes: an input layer, two GRU hidden layers, an attention layer, and an output layer; the attention weight of the attention layer is the ratio of the absolute gradient value of the current time step data to the sum of the absolute gradient values ​​of the full time series data, and the absolute gradient value is calculated by the first-order central difference method; The structural static representation branch includes: an input layer, three CNN layers, two fully connected layers, and an output layer; the number of convolutional kernels in the CNN layers are 32, 64, and 128, respectively, and the activation function of the output layer is a linear function. The seepage coupling correlation branch is nonlinearly correlated through the pressure-temperature interaction term and the stress-permeability interaction term.

[0012] Optionally, the intermediate results of the branches are combined in a hierarchical manner to obtain key reservoir indicators, including: The total gas content is calculated based on the structural foundation parameters and environmental correction factors; the formula for calculating the total gas content is: ;in, Based on gas content, This is the temperature correction factor. The seepage efficiency coefficient is... For nanopore volume, For micron pore volume, In-situ pressure, Absolute temperature Based on the coefficient, The density of coal; The effective permeability is calculated based on the structural foundation parameters, environmental correction factor, and seepage efficiency factor; the formula for calculating the effective permeability is: ;in, This is the stress attenuation coefficient; The stress sensitivity coefficient is calculated based on the effective permeability; the formula for calculating the stress sensitivity coefficient is: ;in, This is in-situ stress; The confidence weights are calculated based on the environmental correction factor, structural foundation parameters, and seepage efficiency factor. Based on the confidence weight, the total gas content, effective permeability, and stress sensitivity coefficient are corrected to obtain the key reservoir indicators.

[0013] A key indicator detection system for deep coalbed methane reservoirs includes: The first data acquisition module is used to perform differentiated screening and preprocessing of the sampling well sections of the target deep coalbed methane reservoir, and to acquire reservoir environmental parameters, physical property parameters and seepage data through a multi-field coupling test system to obtain in-situ multi-field raw data. The second data acquisition module is used to obtain reservoir coal samples by pressure sampling and to characterize the reservoir coal samples at both nanoscale and microscale pore sizes to obtain raw data of the dual-scale structure. The data processing module is used to perform outlier removal, normalization, and spatiotemporal alignment on in-situ multi-field raw data and dual-scale structure raw data to obtain a standardized dataset. The parameter identification module is used to construct a multi-branch coupling model and perform differentiated operations on the standardized dataset based on the multi-branch coupling model to obtain intermediate results of different branches. The indicator detection module is used to perform hierarchical combination calculations on the intermediate results of the branches to obtain key reservoir indicators.

[0014] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The method for detecting key indicators of deep coalbed methane reservoirs provided by the present invention includes: differential screening and preprocessing of sampling well sections of the target deep coalbed methane reservoir; acquiring reservoir environmental parameters, physical property parameters, and seepage data through a multi-field coupling test system to obtain in-situ multi-field raw data; obtaining reservoir coal samples through pressure-maintaining sampling; characterizing the reservoir coal samples at both nanoscale and microscale porosity to obtain dual-scale structural raw data; performing outlier removal, normalization, and spatiotemporal alignment on the in-situ multi-field raw data and dual-scale structural raw data to obtain a standardized dataset; constructing a multi-branch coupling model; and performing differential calculations on the standardized dataset according to the multi-branch coupling model to obtain different branch intermediate results; and performing hierarchical combination calculations on the branch intermediate results to obtain key reservoir indicators. This method, through differential screening and preprocessing, multi-field coupling data acquisition, dual-scale porosity characterization, and multi-branch coupling model calculations, accurately couples environmental, structural, and seepage parameters, and improves the detection accuracy of key indicators of deep coalbed methane reservoirs. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the method for detecting key indicators of deep coalbed methane reservoirs according to the present invention; Figure 2 This is a schematic diagram of the key indicator detection system for deep coalbed methane reservoirs according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] 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.

[0019] like Figure 1 As shown, this invention provides a method for detecting key indicators of deep coalbed methane reservoirs, comprising the following steps: Step 100: Differentiated screening and preprocessing of the sampling well section of the target deep coalbed methane reservoir, and collection of reservoir environmental parameters, physical property parameters and seepage data through a multi-field coupling test system to obtain in-situ multi-field raw data; Step 200: Obtain reservoir coal samples by pressure-holding sampling, and characterize the reservoir coal samples at both nanoscale and microscale pore sizes to obtain raw data of the dual-scale structure; Step 300: Perform outlier removal, normalization, and spatiotemporal alignment on the in-situ multi-field raw data and dual-scale structure raw data to obtain a standardized dataset; Step 400: Construct a multi-branch coupling model, and perform differentiated computation on the standardized dataset based on the multi-branch coupling model to obtain intermediate results for different branches; Step 500: Perform hierarchical combination calculations on the intermediate results of the branches to obtain key reservoir indicators.

[0020] Preferably, the sampling well sections of the target deep coalbed methane reservoir are subjected to differentiated screening and preprocessing, and reservoir environmental parameters, physical property parameters, and seepage data are collected through a multi-field coupled testing system to obtain in-situ multi-field raw data, including: The sampling well sections were screened using previous natural gamma logging and density logging data to obtain the target well sections; Casing windows were opened in the target well section, and coal dust was removed from the wellbore wall using high-pressure water jets. Inspect the surface flatness of the target well section after cleaning to ensure that the fluctuation range of the surface flatness does not exceed 2 mm.

[0021] Preferably, the multi-field coupling test system includes: a sensing unit, an environmental control unit, and a sealing unit; the sensing unit incorporates a fiber optic stress sensor, a miniature pressure sensor, a platinum resistance temperature sensor, a high-frequency electromagnetic sensor, an ultrasonic probe, and a fluid flow sensor; the environmental control unit consists of a miniature heating rod and a pressure compensation pump; and the sealing unit is a sealing ring composed of hydrogenated nitrile rubber and a metal skeleton.

[0022] In some embodiments, the natural gamma logging curve of the target deep well is first retrieved. Because the content of radioactive elements in the coal seam is significantly lower than that in the surrounding rock, the natural gamma value is usually in the lower range of 50-100 API, and the curve shape is relatively stable. Based on this, well sections with gamma values ​​matching the characteristics are initially identified and marked as potential coal seam sections. Subsequently, density logging data is introduced for secondary verification. The density of coal seams is generally 1.3-1.8 g / cm³. 3 Between these values, the density is significantly lower than that of dense sandstone and mudstone. Density logging values ​​were checked one by one for potential coal seam intervals, and those with densities falling between 1.3 and 1.8 g / cm³ were selected. 3The well sections with no abnormal peaks in the density curve and those that simultaneously meet the natural gamma and density characteristics are identified as target well sections. This ensures that subsequent operations are only targeted at the real deep coalbed methane reservoirs and avoids interference with the detection results from non-target formations. Then, using casing coupling positioning logging, the precise depth of the casing and the location of the coupling within the target well section are obtained, and the center point for opening the window is determined by avoiding the coupling. Subsequently, a hydraulically driven casing window opening tool is lowered. The tool is equipped with a carbide end mill with a diameter of 80-100mm. When the tool reaches the center point of the window, the end mill angle is adjusted to 35° by the downhole guide mechanism. The hydraulic system is started to drive the end mill at a constant speed of 250r / min to cut the casing wall. At the same time, the drilling fluid circulation system is started to deliver drilling fluid with a viscosity of 30mPa·s at a flow rate of 15L / min. This is used to cool the end mill and to carry the casing debris generated during cutting out of the wellbore. Cutting continues until the end mill completely penetrates the casing wall, forming a circular window with a diameter 8mm larger than the subsequent test probe. After confirming with a downhole camera that there are no burrs or casing cracks at the edge of the window, the end mill is retracted, and the casing window opening is completed. Next, the high-pressure cleaning device equipped with a fan-shaped nozzle is lowered to the window position, and the distance between the nozzle and the wall is adjusted to 12cm and the spray angle to 18°. The ground high-pressure pump is started to pressurize the clean water to 20MPa and spray high-pressure water flow onto the wall at a pulse frequency of 12Hz. At the same time, the wellbore return system is turned on to return the wastewater carrying coal dust to the ground at a flow rate of 20L / min. The coal dust content in the return water is monitored in real time by an online particle size analyzer. When the coal dust content is below 0.1% for 10 minutes, it is determined that the coal dust on the wall has been cleaned. The high-pressure pump is then turned off, and the cleaning is completed. Finally, a wellbore wall measuring instrument with eight built-in miniature displacement sensors was used. The instrument was lowered into the window area, ensuring the sensor probes were in close contact with the wall. The instrument was controlled to move along the wellbore axis at a uniform speed of 8 cm / min. Sensor data acquisition was triggered every 1 mm of movement, recording the distance between the sensor probes and the wall at different axial positions. After covering the window area and a 0.5 m section above and below, the collected distance data was transmitted to the ground terminal, and the difference between the maximum and minimum distances within that section was calculated. If the difference was ≤2 mm, the wall flatness met the requirements. If the difference was >2 mm, low-pressure water jet trimming was required for the protruding parts, and the measurement was repeated until the undulation amplitude met the requirements.

[0023] Furthermore, before acquiring reservoir environmental parameters, physical property parameters, and seepage data through the multi-field coupling testing system, each component of the sensing unit is calibrated. The fiber optic stress sensor is calibrated using a 0-100MPa standard stress source to ensure a stress measurement error ≤0.5MPa; the miniature pressure sensor is calibrated using a 0-150MPa standard pressure gauge to ensure an error ≤0.1MPa; the platinum resistance temperature sensor is calibrated using a 0-250℃ constant temperature bath to ensure an error ≤0.2℃; the high-frequency electromagnetic sensor is calibrated using a 1-100MHz standard electromagnetic signal source to ensure a sensitivity error ≤5%; the ultrasonic probe is calibrated using a standard steel test block with a known sound velocity to ensure a propagation time measurement error ≤1μs; and the fluid flow sensor is calibrated using a 0-100mL / min standard flow device to ensure an error ≤2%. In the environmental control unit, the miniature heating rod has a preset heating program, and the pressure compensation pump is connected to a nitrogen source; the sealing unit uses a composite sealing ring made of hydrogenated nitrile rubber and a metal skeleton, installed at the gap between the test probe and the sleeve window, and utilizes the support of the metal skeleton and the elasticity of the rubber to achieve a seal and prevent fluid leakage. The test probe was then lowered into the windowed area. After confirming good contact between the sensors and the well wall using a downhole camera, the environmental control unit was activated. A miniature heating rod heated the reservoir to its in-situ temperature at a rate of 3℃ / min, and a pressure compensation pump pressurized the reservoir to its in-situ pressure at a rate of 1.5MPa / min. Once the temperature and pressure stabilized within ±0.5℃ and ±0.1MPa of their in-situ values, respectively, for 30 minutes, the sensing unit was activated to collect data. A fiber optic stress sensor collected reservoir stress data at a frequency of 1Hz and recorded dynamic stress changes. A miniature pressure sensor collected pressure data at a frequency of 2Hz and monitored pressure stability. A platinum resistance temperature sensor collected temperature data at a frequency of 1Hz to ensure a constant ambient temperature. A high-frequency electromagnetic sensor emitted electromagnetic signals at a frequency of 0.5Hz and received the electromagnetic response of the reservoir coal and rock. An ultrasonic probe emitted ultrasonic waves at a frequency of 1Hz and calculated the density of the coal and rock by the propagation time of the reflected waves. Simultaneously, a micro-pump was activated to inject simulated formation water into the reservoir at a flow rate controlled at 3mL / min, and a fluid flow sensor collected seepage flow data at a frequency of 0.5Hz. All data is transmitted in real time to the ground storage system via cable, continuously collected for 24 hours, and finally formed in-situ multi-field raw data containing environmental parameters, physical property parameters and seepage data.

[0024] Preferably, reservoir coal samples are obtained through pressure-holding sampling, and the reservoir coal samples are characterized at both nanoscale and microscale pore sizes to obtain raw data on the dual-scale structure, including: Coal samples from the reservoir were obtained by pressure sampling using a diamond drill bit. Cross-sectional processing and scanning of reservoir coal samples yielded nanoscale images; Resonance tests and X-ray scanning were performed on reservoir coal samples to obtain micron-sized images; A three-dimensional model is generated based on nanoscale and microscale images, and the original two-scale structural data containing specific surface area and pore size distribution are calculated using the BET and BJH methods.

[0025] In some embodiments, a diamond core drill bit pressure-holding sampling system is used, with a drill bit diameter of 50 mm, a helical diamond cutting edge arrangement, and an 80-mesh cutting edge grit size. The main body of the pressure-holding chamber is made of high-strength titanium alloy. The initial pressure of the pressure-holding chamber is preset to the reservoir in-situ pressure, and pressure changes are monitored in real time by an in-chamber pressure sensor. When the collected coal core enters the pressure-holding chamber, the chamber door is automatically closed. At the moment of closure, nitrogen gas is injected into the chamber through an in-chamber pressure compensation pump to maintain the pressure stable within the range of ±0.2 MPa of the in-situ pressure. Drilling is then stopped, and the chamber is slowly pulled out to transport it to the surface. The coal core is then cut into three coal sample blocks with dimensions of 50 mm × 20 mm × 20 mm. One spare coal sample block is then selected, and a cross-section is prepared using a focused ion beam processing system. After processing, the cross-section is blown with inert gas to remove residual gallium ions and coal dust particles. The fresh cross-section is then scanned using a transmission electron microscope to obtain a nanoscale pore image covering the entire processed area. Next, a second coal sample was selected, and a resonance test was conducted using a dynamic mechanical analyzer. Based on the intensity and phase difference of the resonance peaks detected by the frequency sweep, the concentrated area of ​​micron-sized pores was marked as the high-porosity response region. Then, a micron-scale X-ray computed tomography system was used to scan the sample to obtain a three-dimensional image of the micron-scale pores.

[0026] Furthermore, a feature-point matching-based stitching algorithm was used to stitch and denoise the nanoscale images, forming a 20μm×20μm×5μm nanopore 3D sub-model. The micrometer-scale reconstructed model was then cropped, retaining a 10mm×10mm×10mm micrometer-scale pore 3D sub-model corresponding to the high-porosity response region. Using the marked points on the coal sample surface as a reference, the nano-scale sub-model was embedded into the corresponding position of the micrometer-scale sub-model through coordinate calibration, forming a dual-scale pore 3D fusion model. Then, the specific surface area of ​​the coal sample was calculated using the liquid nitrogen adsorption-desorption method (BET method), with the following formula: ; ; in, This is the equilibrium partial pressure of nitrogen (kPa). This is the saturated vapor pressure of nitrogen at liquid nitrogen temperature (kPa). Adsorption capacity (cm) 3 / g, standard state), Monolayer saturation adsorption capacity (cm) 3 / g, standard state), It is a constant related to the heat of adsorption. Let Avogadro's constant be 1. The cross-sectional area of ​​a nitrogen molecule. For sample mass. Based on the nitrogen desorption curve obtained from the BET test, the pore size distribution at the nanometer and micrometer scales was calculated using the BJH (Barrett-Joyner-Halenda) method. The calculation formula is as follows: ; in, The surface tension of nitrogen gas, Where is the pore radius (nm). It is the gas constant (8.314 J / (mol·K)). The absolute temperature is 77K. Finally, the nanopore volume extracted from the three-dimensional fusion model was... Micron pore volume, specific surface area calculated by the BET method The full-aperture distribution data, which is obtained by combining the BJH method with CT analysis, is integrated to form the original data of the dual-scale structure.

[0027] Preferably, outlier removal, normalization, and spatiotemporal alignment are performed on the in-situ multi-field raw data and the dual-scale structured raw data to obtain a standardized dataset, including: The density-based DBSCAN algorithm was used to remove outliers from the in-situ multi-field raw data and the dual-scale structure raw data, and then the data was integrated into outlier-free data. Max-min standardization is performed on the static features in the outlier-removed data, and median standardization is performed on the dynamic data in the outlier-removed data to obtain normalized data. A standardized dataset is obtained by spatiotemporally aligning time-series and discrete data in normalized data using cubic spline interpolation.

[0028] In some specific embodiments, the outlier removal process in step 300 is implemented as follows: First, the input data is divided into time-series dynamic data and static structured data. Time-series dynamic data consists of parameters that change over time in in-situ multi-field data, characterized by strong temporal continuity and high correlation of local fluctuations. Static structured data consists of static parameters in dual-scale structured data and in-situ multi-field data, characterized by spatial discrete distribution and high numerical stability. Statistical characteristics are calculated for both types of data. For time-series dynamic data, the sliding window local density and time-series gradient standard deviation are calculated; for static structured data, the Pearson correlation coefficient and interquartile range (IQR) of specific surface area and nanopore volume are calculated. Then, the key parameters of DBSCAN are adaptively determined. For time-series dynamic data, the neighborhood radius ε is calculated using the local density back-calculation method. That is, each segment of time-series data is first divided into several subsequences according to a sliding window, and the mean Euclidean distance between all points in each subsequence is calculated. 1.2 times the average distance of all subsequences is taken as ε for the time-series parameter. The minimum number of points is determined to be the stress data sampled at sampling frequency × 2 + 1Hz, so as to ensure that the number of local points is sufficient to judge clustering while avoiding the failure to detect anomalies due to the excessive minimum number of points caused by the continuity of time-series data. For static structural data, the neighborhood radius ε is calculated using the correlation constraint method. That is, 2-3 specific surface areas with a correlation > 0.6 with the target parameter are selected to construct a multidimensional feature space associated with nanopore volume and pore size distribution, and the average Euclidean distance of all data points in this space is calculated. 0.8 times the distance is taken as ε. The minimum number of points is determined to be 9 to adapt to the multidimensional discrete characteristics of static data.

[0029] After applying the adaptive DBSCAN algorithm to both types of data, outliers are identified and marked in the time-series dynamic data if they do not belong to any cluster; and in the static structural data, noise points far from the core points in the multidimensional feature space are marked as outliers. If an outlier in the time-series pressure data corresponds to a pressure jump > 5 MPa, or a static specific surface area outlier < 0.1 m², then an outlier is considered an outlier. 2 If the outlier is / g, it is confirmed to be removed; if the outlier meets the criteria for extreme geological conditions, it is retained and the special geological points are marked. Finally, the normal points and the marked special geological points of the two types of data will be merged into the de-outliered data.

[0030] Furthermore, during spatiotemporal alignment, the spatiotemporal attributes of the data were first analyzed to clarify the characteristics of static structural data and in-situ multi-field static data, which correspond only to discrete depth points, and dynamic time-series data, which have different sampling frequencies. A spatiotemporal reference grid was established with the target well depth as the vertical axis and the unified time point of the dynamic data as the horizontal axis. Subsequently, spatial dimension alignment was performed. For the static data, based on regional geological data, the permeability threshold was determined to be 5% / m and the specific surface area threshold to be 8% / m. A cubic spline function was constructed by selecting the values ​​of three adjacent discrete depth points, and a first-order derivative constraint was added during the solution to meet the change rate requirement. If the interpolation result exceeded the constraint, the weights of adjacent points were adjusted and recalculated. The static parameters were interpolated to depth points at 0.1m intervals to form spatially aligned static data. Next, time-dimensional alignment was performed. The time fluctuation coefficient within each 10-minute interval was calculated for the dynamic data and categorized into three levels: fluctuation coefficient < 0.1, time step 10 minutes; 0.1 ≤ fluctuation coefficient < 0.3, time step 5 minutes; fluctuation coefficient ≥ 0.3, time step 2 minutes. Time periods were divided according to the categorization results. Cubic spline functions were constructed to interpolate the original data for each time period to the corresponding time point. If the difference between adjacent interpolation points exceeded twice the average parameter fluctuation value, the step size was reduced by 50% and re-interpolated to form time-aligned dynamic data. Finally, spatiotemporal consistency verification was performed to determine physical correlation rules (in some embodiments, this manifests as a slight decrease in effective permeability when pressure increases). The correlation of parameters for each spatiotemporal grid node was checked. Nodes that did not conform to the rules were adjusted and recalculated until the correlation compliance rate was ≥ 95%. Finally, the spatially aligned static data and the time-aligned dynamic data were integrated according to the spatiotemporal grid nodes to generate a standardized dataset.

[0031] Preferably, a multi-branch coupling model is constructed, and differentiated computational processing is performed on the standardized dataset based on the multi-branch coupling model to obtain intermediate results for different branches, including: An environmental dynamic response branch is constructed based on the GRU model, and attention weights are adjusted on the time series data in the standardized dataset to obtain environmental correction coefficients. The environmental correction coefficients include stress attenuation coefficients and temperature correction coefficients. A static structural representation branch is constructed based on a convolutional neural network, and features are extracted from the structural parameter data in the standardized dataset to obtain the basic structural parameters, including: basic gas content and basic permeability. Based on the gradient boosting tree model, a seepage coupling correlation branch is constructed, and the linkage data in the standardized dataset are nonlinearly correlated to obtain the seepage efficiency coefficient.

[0032] In some embodiments, three types of core data are selected from the standardized dataset. Environmental time-series data are directed into the environmental dynamic response branch, including in-situ temperature time series and in-situ stress time series at 24-hour 10-minute intervals. Structural static data are directed into the structural static characterization branch, including nanopore volume, micropore volume, specific surface area, pore size distribution, and coal and rock density. Seepage linkage data are directed into the seepage coupling correlation branch, including in-situ pressure, temperature, and stress linkage data with seepage flow rate.

[0033] Specifically, the structure of the environmental dynamic response branch, from top to bottom, is as follows: input layer, gradient calculation layer, attention layer, two GRU hidden layers, and fully connected output layer. The input layer has an input dimension of 2 (two temporal parameters, temperature and stress) × 144 (time points), i.e., a 288-dimensional vector, used to map timestamps to the [0,1] interval. The gradient calculation layer quantifies the severity of temporal fluctuations by calculating the absolute value of the first-order central difference gradient of the data at each time step. The absolute value of the gradient at time step t... The calculation formula is: ;in, For the first The parameter values ​​for the time step, The time interval is 10 minutes, normalized to 10 minutes. The first and last time steps employ forward differencing and backward differencing, respectively. The attention layer assigns attention weights based on the absolute value of the gradient, thus giving higher weights to time steps with significant fluctuations. The formula for calculating the attention weight at time step t is: The GRU hidden layer has 64 hidden units in its first layer and 32 hidden units in its second layer, with a 0.2 Dropout applied to each layer. The activation function is tanh. The hidden state of the last time step of the second GRU layer is input into the fully connected output layer. The fully connected output layer has 16 neurons in its first layer and 2 neurons in its second layer, directly outputting the stress attenuation coefficient and temperature correction coefficient. The output stress attenuation coefficient ranges from 0.6 to 1.0, and the temperature correction coefficient ranges from 0.8 to 1.2. If the output coefficients exceed the constraints, the average of the nearest time steps is used for replacement.

[0034] The structure of the static structural representation branch, from top to bottom, is as follows: a multi-channel input layer, three CNN feature extraction layers, a cross-layer feature fusion layer, two fully connected layers, and a linear output layer. The multi-channel input layer has five channels, taking the volume distribution of nanopores, the volume distribution of micropores, the specific surface area distribution, the pore size distribution of 2-10nm, and the pore size distribution of 100nm-1μm as input channels. The first CNN layer uses 32 3×3 convolutional kernels with a stride of 1, padding of 1, and ReLU as the activation function, followed by a 2×2 max pooling operation, outputting a 32-channel 5×5 feature map. The second CNN layer uses 64 5×5 convolutional kernels with a stride of 1, padding of 2, and ReLU as the activation function, followed by a 2×2 max pooling operation, outputting a 64-channel 3×3 feature map. The third CNN layer uses 128 3×3 convolutional kernels with a stride of 1, padding of 1, and ReLU as the activation function, without pooling, retaining the 128-channel 3×3 feature map. The cross-layer feature fusion layer first compresses the pooled features (32×5×5) from the first layer to 32×3×3 (the same size as the second and third layers) using a 1×1 convolution. Then, it concatenates the three layers' features (32×3×3 + 64×3×3 + 128×3×3) into a 224×3×3 feature map. Weights are then assigned and weighted using a channel attention mechanism to obtain the fused feature map, avoiding the dominance of a single layer's features. The fused feature map is flattened into a 224×3×3 = 2016-dimensional vector and input into a fully connected layer. The first layer of the fully connected layer has 256 neurons, and the second layer has 64 neurons, with a 0.3 Dropout factor. Finally, the linear output layer has two neurons, directly outputting the basic gas content and basic permeability. The loss function of the linear output layer is a combination of the MSE loss function and a range constraint penalty term, expressed as: ;in, Based on gas content, Based on penetration rate, It is 0.6. It is 0.9. For predictive data, This is real data.

[0035] The seepage coupling correlation branch selects in-situ pressure from the dataset. In-situ temperature In-situ stress and basic penetration rate Four core original features were identified, and three types of physical interaction terms were constructed, including first-order interaction terms. and The second-order interaction term is and The normalized interaction terms are mapped to [0,1] using max-min normalization of the first and second order interaction terms. Micro-fracture testing is performed on the target well section to measure the actual seepage flow rate under different pressures and temperatures. The actual seepage efficiency is then derived using Darcy's law, resulting in labeled data. This labeled data is used as the training set to train the XGBoost model. The three types of physical interaction terms are input into the trained XGBoost model, and the original predicted values ​​are then mapped using physical constraints to obtain the final seepage efficiency coefficients. The mapping rule uses the sigmoid function to compress the original predicted values ​​to [0.5,1.0], expressed as: ;in, These are the original predicted values ​​from the model.

[0036] It should be noted that step 400 constructs a gradient-aware attention GRU network. This network calculates the absolute value of the gradient in the time-series data using first-order central difference and assigns attention weights, ensuring that time points with significant impacts on reservoir indicators, such as sudden stress increases and temperature abrupt changes, receive a higher contribution. This improves the matching degree between the model output and the actual dynamic response of the reservoir. Based on a multi-scale feature fusion CNN network, structural data is reconstructed into multi-channel 2D feature maps. Through the fusion of 32 / 64 / 128 differentiated convolutional kernels with cross-layer features, the network overcomes the representation blind spots of single-scale CNNs, significantly reducing computational errors and effectively identifying the dense nanopore regions unique to deep coal seams. Simultaneously, an interaction-enhanced gradient boosting tree model is employed, addressing the shortcomings of conventional XGBoost, which relies solely on original features and struggles to learn deep coupling patterns. This avoids logical contradictions caused by isolated calculations of parameters, providing reliable support for the accurate calculation of subsequent key reservoir indicators and significantly improving the model's adaptability to deep, high-temperature, high-pressure, and highly heterogeneous environments.

[0037] In some embodiments, step 500 includes the following specific steps: The total gas content is calculated based on the structural foundation parameters and environmental correction factors. The calculation formula is as follows: ; in, Based on gas content, This is the temperature correction factor. The seepage efficiency coefficient is... For nanopore volume, For micron pore volume, In-situ pressure, Absolute temperature Based on the coefficient, This refers to the density of coal.

[0038] The effective permeability is calculated based on the structural foundation parameters, environmental correction factor, and seepage efficiency factor. The calculation formula is as follows: ; in, This is the stress attenuation coefficient.

[0039] The stress sensitivity coefficient is calculated based on the effective permeability, and the calculation formula is as follows: ; in, This refers to in-situ stress.

[0040] The R-value trained in step 400 is used to train the model branch. 2 Values ​​and three branches R 2 The ratio of the sums of the values ​​is used as the confidence weight for the corresponding branch result. Based on the confidence weight, the total gas content, effective permeability, and stress sensitivity coefficient are corrected to obtain the key reservoir indicators. The corrected formula is: ; in, , and These are the confidence weights for total gas content, effective permeability, and stress sensitivity coefficient, respectively.

[0041] like Figure 2 As shown, the present invention also provides a key indicator detection system for deep coalbed methane reservoirs, comprising: The first data acquisition module is used to perform differentiated screening and preprocessing of the sampling well sections of the target deep coalbed methane reservoir, and to acquire reservoir environmental parameters, physical property parameters and seepage data through a multi-field coupling test system to obtain in-situ multi-field raw data. The second data acquisition module is used to obtain reservoir coal samples by pressure sampling and to characterize the reservoir coal samples at both nanoscale and microscale pore sizes to obtain raw data of the dual-scale structure. The data processing module is used to perform outlier removal, normalization, and spatiotemporal alignment on in-situ multi-field raw data and dual-scale structure raw data to obtain a standardized dataset. The parameter identification module is used to construct a multi-branch coupling model and perform differentiated operations on the standardized dataset based on the multi-branch coupling model to obtain intermediate results of different branches. The indicator detection module is used to perform hierarchical combination calculations on the intermediate results of the branches to obtain key reservoir indicators.

[0042] The beneficial effects of this invention are as follows: 1) Based on the multi-field coupling test system, the high temperature and high pressure environment of deep reservoirs was accurately simulated. Environmental parameters, physical property parameters and seepage data were collected simultaneously. This solved the problems of insufficient sealing of traditional pressure-holding coring and distortion of in-situ environment simulation. It ensured that the original data was close to the actual state of the reservoir and improved the accuracy of data acquisition. 2) The use of dual-scale pore characterization technology at the nano and micro scales fills the gap in the existing technology for the identification and quantification of nanopores, and realizes the fine characterization of pore structure. 3) A multi-branch coupling model of "environmental dynamic response - structural static characterization - seepage coupling correlation" was constructed. The time-series dynamic correlation was captured by the GRU model, the dual-scale structural features were extracted by the CNN model, and the multi-parameter nonlinear correlation was established by the gradient boosting tree model. This achieved accurate coupling of the dynamic interaction between temperature, pressure, stress and pore structure and seepage capacity, and enhanced the analytical capability of multi-parameter coupling. 4) By combining the calculation logic of layered combination, the basic parameters, environmental correction coefficient and seepage efficiency coefficient are integrated, and the total gas content, effective permeability and stress sensitivity coefficient are corrected by combining confidence weight. This significantly reduces the calculation error of key indicators. While meeting the accurate evaluation requirements of deep reservoir sweet spot identification, mining scheme optimization and production capacity prediction, it also improves the detection accuracy of key indicators.

[0043] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0044] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for detecting key indicators of deep coalbed methane reservoirs, characterized in that, Includes the following steps: Differential screening and preprocessing were carried out on the sampling well sections of the target deep coalbed methane reservoir, and reservoir environmental parameters, physical property parameters and seepage data were collected through a multi-field coupling test system to obtain in-situ multi-field raw data; Coal samples from the reservoir were obtained by pressure-holding sampling, and the pore size of the coal samples was characterized at both the nanoscale and microscale to obtain the original data of the dual-scale structure. Outlier removal, normalization, and spatiotemporal alignment are performed on the in-situ multi-field raw data and the dual-scale structure raw data to obtain a standardized dataset. A multi-branch coupling model is constructed, and differential operation processing is performed on the standardized dataset according to the multi-branch coupling model to obtain different branch intermediate results; The intermediate results of the branches are combined and calculated in a hierarchical manner to obtain the key reservoir indicators.

2. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 1, characterized in that, Differential screening and preprocessing were performed on the sampling well sections of the target deep coalbed methane reservoir. Reservoir environmental parameters, physical property parameters, and seepage data were collected using a multi-field coupled testing system to obtain in-situ multi-field raw data, including: The sampled well sections were screened using previous natural gamma logging and density logging data to obtain the target well sections; Casing windows were opened in the target well section, and coal dust was removed from the wellbore wall using high-pressure water jets. The smoothness of the target well section wall is checked after cleaning to ensure that the fluctuation range of the wall smoothness does not exceed 2 mm.

3. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 1, characterized in that, The multi-field coupling test system includes a sensing unit, an environmental control unit, and a sealing unit. The sensing unit incorporates a fiber optic stress sensor, a miniature pressure sensor, a platinum resistance temperature sensor, a high-frequency electromagnetic sensor, an ultrasonic probe, and a fluid flow sensor. The environmental control unit consists of a miniature heating rod and a pressure compensation pump. The sealing unit is a sealing ring composed of hydrogenated nitrile rubber and a metal skeleton.

4. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 1, characterized in that, Coal samples from the reservoir were obtained through pressure-holding sampling, and the samples were characterized for nanoscale and microscale pore size to obtain raw data on the dual-scale structure, including: The coal sample from the reservoir was obtained by pressure sampling using a diamond drill bit; The reservoir coal sample was cross-sectionally processed and scanned to obtain nanoscale images; Resonance tests and X-ray scanning were performed on the coal samples from the reservoir to obtain micron-sized images; A three-dimensional model is generated based on the nanoscale image and the microscale image, and the original data of the dual-scale structure, including specific surface area and pore size distribution, are calculated using the BET method and the BJH method.

5. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 1, characterized in that, The in-situ multi-field raw data and the dual-scale structured raw data are subjected to outlier removal, normalization, and spatiotemporal alignment to obtain a standardized dataset, including: The density-based DBSCAN algorithm is used to remove outliers from the in-situ multi-field raw data and the dual-scale structure raw data, and then integrates them into outlier-free data. The static features in the anomaly-removed data are subjected to min-max standardization, and the dynamic data in the anomaly-removed data are subjected to median standardization to obtain normalized data. The normalized dataset is obtained by spatiotemporally aligning the time-series data and discrete data in the normalized data using cubic spline interpolation.

6. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 1, characterized in that, A multi-branch coupling model is constructed, and differential computation processing is performed on the standardized dataset based on the multi-branch coupling model to obtain intermediate results of different branches, including: An environmental dynamic response branch is constructed based on the GRU model, and attention weights are adjusted on the time-series data in the standardized dataset to obtain environmental correction coefficients; the environmental correction coefficients include: stress attenuation coefficients and temperature correction coefficients; A static structural representation branch is constructed based on a convolutional neural network, and features are extracted from the structural parameter data in the standardized dataset to obtain the basic structural parameters; the basic structural parameters include: basic gas content and basic permeability; Based on the gradient boosting tree model, a seepage coupling correlation branch is constructed, and the linkage data in the standardized dataset are nonlinearly correlated to obtain the seepage efficiency coefficient.

7. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 6, characterized in that, The environmental dynamic response branch includes: an input layer, two GRU hidden layers, an attention layer, and an output layer; the attention weight of the attention layer is the ratio of the absolute value of the gradient of the current time step data to the sum of the absolute values ​​of the gradients of the full time series data, and the absolute value of the gradient is calculated by the first-order central difference method; The static structural representation branch includes: an input layer, three CNN layers, two fully connected layers, and an output layer; the number of convolutional kernels in the CNN layers are 32, 64, and 128, respectively, and the activation function of the output layer is a linear function; The seepage coupling correlation branch is nonlinearly correlated through the pressure-temperature interaction term and the stress-permeability interaction term.

8. The method for detecting key indicators of deep coalbed methane reservoirs according to claim 6, characterized in that, The intermediate results of the branches are combined in a hierarchical manner to obtain key reservoir indicators, including: The total gas content is calculated based on the structural basic parameters and the environmental correction coefficient; the formula for calculating the total gas content is: ;in, Based on gas content, This is the temperature correction factor. The seepage efficiency coefficient is... For nanopore volume, For micron pore volume, In-situ pressure, Absolute temperature Based on the coefficient, The density of coal; The effective permeability is calculated based on the structural foundation parameters, the environmental correction coefficient, and the seepage efficiency coefficient; the formula for calculating the effective permeability is: ;in, This is the stress attenuation coefficient; The stress sensitivity coefficient is calculated based on the effective permeability; the formula for calculating the stress sensitivity coefficient is: ;in, This is in-situ stress; The confidence weight is calculated based on the environmental correction coefficient, the structural foundation parameters, and the seepage efficiency coefficient. The total gas content, effective permeability, and stress sensitivity coefficient are corrected according to the confidence weight to obtain the key reservoir indicators.

9. A key indicator detection system for deep coalbed methane reservoirs, characterized in that, include: The first data acquisition module is used to perform differentiated screening and preprocessing of the sampling well sections of the target deep coalbed methane reservoir, and to acquire reservoir environmental parameters, physical property parameters and seepage data through a multi-field coupling test system to obtain in-situ multi-field raw data. The second data acquisition module is used to obtain reservoir coal samples by pressure-holding sampling, and to characterize the reservoir coal samples at both nanoscale and microscale pore sizes to obtain raw data of the dual-scale structure. The data processing module is used to perform outlier removal, normalization, and spatiotemporal alignment on the in-situ multi-field raw data and the dual-scale structure raw data to obtain a standardized dataset. The parameter identification module is used to construct a multi-branch coupling model and perform differential operation processing on the standardized dataset according to the multi-branch coupling model to obtain different branch intermediate results; The indicator detection module is used to perform hierarchical combination calculations on the intermediate results of the branches to obtain key reservoir indicators; The key reservoir indicators include: total gas content, effective permeability, and stress sensitivity coefficient.

Images