Method for reconstructing single well U-TH radioactive element ratio through combination of three-dimensional modeling and neural network

Abstract

The invention discloses a method for reconstructing a single well U-TH radioactive element ratio through combination of three-dimensional modeling and a neural network. The method comprises the following steps: 1) obtaining U and TH values; 2) drawing U and TH curves; 3) establishing a U / TH ratio three-dimensional model; 4) reconstructing a U / TH ratio prediction curve flow based on a single model; and 5) reconstructing a U / TH ratio prediction curve flow based on multiple models. The invention provides a method for reconstructing the single well U-Th radioactive element ratio through combination of three-dimensional modeling and a neural network, through combination of a model method and a neural network method, the defects of the model method and the neural network method are overcome, the advantages of the model method and the neural network method are brought into full play, effectiveness and reliability are achieved through verification of a sparse well, and the trend of oxidation-reduction condition change can be reflected through a U / TH ratio curve; generally, the larger the ratio is, the stronger the reducing environment is, the deeper the paleo-water depth is, and vice versa, and the method can be used for assisting geological fine layering, sedimentary environment evolution analysis and lithofacies paleo-geography recovery, and assisting shale fine stratigraphic framework establishment and reservoir dessert evaluation.

Description

Technical Field

[0001] The invention relates to the technical field of shale reservoir research, and in particular to a method for jointly reconstructing a single well U-TH radioactive element ratio by three-dimensional modeling and neural network. Background Art

[0002] As the ballast of unconventional natural gas, shale gas is of great significance to the security of my country's energy supply. Since 2010, my country's shale gas exploration and development has made great progress and development. Since the lithological characteristics of shale determine the instability of the wellbore and the problems of engineering factors such as cementing, the logging quality of shale gas wells is often not guaranteed, and even logging is often impossible. Therefore, it also promotes the development of logging curve reconstruction technology. At present, the more mainstream method is to calculate the missing curve value of the target well through the correlation analysis of the target parameter curve of the known well and other parameter curves using multiple regression, neural network or other mathematical algorithms to achieve the purpose of curve reconstruction.

[0003] Three-dimensional stochastic modeling is a technology that uses computer graphics technology to generate three-dimensional quantitative stochastic models to accurately depict the three-dimensional distribution of underground geological attributes based on comprehensive analysis of geological, logging, geophysical data and various interpretation results. It is an interdisciplinary comprehensive technology involving geology, data / information analysis, and computational science. The same set of data can generate many similar but different models, which are stochastic models. Underground reservoirs, especially unconventional reservoirs, have heterogeneity and continuity at multiple scales, but it is impossible to directly measure all these details for various reasons. Geostatistics, also known as spatial statistics, aims to describe the distribution of a certain attribute in space, and can better describe the distribution law and geometric form of underground geological attributes in three-dimensional space. The U / TH ratio reflects the underground ancient sedimentary environment and is related to the lithofacies attributes. Therefore, using geostatistical technology to describe lithofacies attributes and predict the U / TH ratio is highly reliable and is an effective means of predicting reservoir attributes.

[0004] In general, there are the following problems in related research at home and abroad: 1) At present, the logging curve reconstruction method based on conventional linear assumptions can no longer meet the requirements of fine description of reservoir characteristics. Although neural networks can identify the nonlinear mapping relationship between logging data, the trend of data changes with reservoir depth, and the correlation between historical data, they are still in the exploratory stage, and the requirements for data samples are very high, and the scope of application and application effect are relatively limited; 2) The current reconstruction methods are mostly aimed at conventional logging curves such as density and acoustic time difference, while the prediction methods for radioactive elements that reflect changes in sedimentary environment are still very lacking. In particular, the U / TH ratio plays an important role in the sedimentary evolution and sweet spot prediction of shale; 3) There are few foreign literatures on well logging curve reconstruction. The publicly published curve reconstruction methods mostly focus on the application of self-organizing map neural networks and genetic algorithms. The effect is good, but the process is cumbersome and has high requirements for data samples.

[0005] In summary, the research and theoretical development of fine-grained rock sedimentology is relatively weak compared to conventional rock sedimentology, and the prediction accuracy of traditional modeling methods is limited. In addition, due to the lack of sample curves with good correlation between U / TH, the single neural network prediction method has low accuracy.

[0006] Therefore, it is an urgent problem for technicians in this field to provide a method for effectively and reliably reconstructing the U-TH radioactive element ratio of a single well by combining the model method and the neural network method to overcome their respective shortcomings and give full play to the advantages of both. Summary of the invention

[0007] The present invention proposes a method for jointly reconstructing the U-Th radioactive element ratio of a single well by three-dimensional modeling and neural network. The method uses natural gamma ray spectrum curves to interpret the U / TH ratios of known wells, combines the constraints of the lithofacies paleogeographic model of the study area, and uses three-dimensional random modeling technology to establish a random model of the U / TH ratio in the study area. Sparse wells are selected, and a reconstruction curve 1 of the sparse wells is generated according to the established U / TH ratio model. Then, the result curve and the conventional curve are combined in a supervised mode to perform a neural network to calculate a U / TH ratio prediction curve 2, which is compared with the actual curve. After multiple iterations until the deviation of the two curve values ​​is less than 5%, the optimal model is finally determined. The results can be applied to fine stratification of single well geology, analysis of sedimentary environment evolution, and restoration of lithofacies paleogeography, and help establish a fine stratigraphic framework for shale and evaluate reservoir sweet spots.

[0008] In order to achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows: The method of jointly reconstructing the U-TH radioactive element ratio of a single well by three-dimensional modeling and neural network includes the following steps: 1) Obtaining U and TH values: Obtaining the values ​​of U and TH by processing and interpreting the natural gamma ray spectrum logging curves of multiple known wells; 2) Draw U and TH curves: According to the values ​​obtained in step 1), calculate the U / TH ratio of each well and draw the U / TH ratio curve; 3) Establish a three-dimensional U / TH ratio model: Based on the U / TH ratio of known wells, multiple three-dimensional U / TH ratio models of the study area are established under the guidance and constraints of the lithofacies paleogeography model by using geological statistics theory and three-dimensional random modeling technology. Three-dimensional random modeling is a technology that uses computer graphics technology to generate three-dimensional quantitative random models based on comprehensive analysis of geological, logging, geophysical data and various interpretation results to accurately depict the three-dimensional distribution of underground geological attributes. It aims to describe the distribution of a certain attribute of the reservoir in space, and can better depict the distribution law and geometric form of underground geological attributes in three-dimensional space. It is an effective means of predicting reservoir attributes. Sedimentary facies is the sum of sedimentary environment and rock characteristics in its environment. The lithofacies paleogeography model is a three-dimensional sedimentary facies model established by using core, logging and other data, aiming to depict the three-dimensional distribution law of sedimentary facies. In underground reservoirs, sedimentary facies controls the spatial distribution of multiple reservoir properties, including U / TH, porosity and other properties. The method is to first establish a lithofacies paleogeographic model, that is, define the spatial distribution of various rock types. In the second step, constrained interpolation of the spatial distribution of U / TH using the lithofacies paleogeographic model can produce a more reasonable U / TH model. 4) Reconstruction of the U / TH ratio prediction curve process based on a single model: Select the target sparse well, and generate a U / TH ratio result curve based on the model at the sparse well according to the model established in step 3), and then use the result curve and the GR curve in a joint supervision mode to calculate the U / TH ratio prediction curve through a neural network; a sparse well refers to a well that does not participate in the above model establishment, but uses the data of other wells to establish a model, and then compares the model data with the actual well data at the well location to verify the reliability of the model. After the model is established, the curve can be generated by extracting values ​​along the well trajectory at the sparse well location according to the grid attribute values ​​​​traversed by the actual well trajectory; 5) Reconstruction process of U / TH ratio prediction curve based on multiple models: select 3-4 high-quality models, apply them to the target wells to generate model-based reconstruction curves, and then use the result curve and GR curve to jointly supervise the model to calculate the prediction curve through neural network. Finally, calculate the average value of each prediction curve. The result is a comprehensive prediction curve. The selection of 3-4 high-quality models has basic probability and statistical significance, covering the possible good, medium and poor realizations; it has high computational timeliness. Usually, geological models are at the level of tens of millions of grids, and it usually takes 10 minutes to run a simulation. Then, it usually takes 30-40 minutes to run 3-4 models. If there are more than 4 models, the timeliness is too low.

[0009] Furthermore, the specific process of the neural network calculation in step 4) is to input the GR and U / TH achievement curves of the known well, establish the mapping relationship between the GR and U / TH curves through computational learning, and then apply it to the target well to predict and generate the U / TH ratio achievement curve.

[0010] Furthermore, when the average relative deviation between the achievement curve and the prediction curve in step 4) is less than 5%, proceed to the next step; when the average relative deviation between the achievement curve and the prediction curve in step 4) is ≥5%, return to step 3), readjust the model parameters, and update the U / TH ratio model.

[0011] Furthermore, the specific process of adjusting the model parameters is to return to the model data analysis module and refit the horizontal range and vertical range of the variogram of the multi-well U / TH data to optimize the range values.

[0012] The specific process of updating the U / TH ratio model is to apply the re-optimized parameters - the range, the values ​​of the major axis and the minor axis - to the static three-dimensional simulation to generate a new three-dimensional attribute model.

[0013] Furthermore, in step 3), multiple U / TH ratio three-dimensional models of the study area are established, and the specific process is: a. Curve gridding: grid the U / TH curve of the input known well. The grid density is related to the formation thickness and the level of refinement of the reservoir research. The vertical density of the grid is the number of grids / reservoir thickness. The more refined the reservoir research requirements, the greater the grid density needs to be, and the greater the total number of grids in the model. b. Data analysis: after gridding, assign a value to each grid to generate a data point. According to the geological characteristics of the data, the data is processed for normal distribution and variogram analysis to generate the range of variation and nugget value applicable to each small layer; c. Parameter simulation: perform normal transformation on the gridded U / TH values ​​to make them obey normal distribution; calculate the experimental variogram; select random paths through the grid cells to be simulated; perform simple Kriging estimation on the grids that have not been simulated; establish a normal distribution based on the estimation and variance, randomly extract values ​​on the distribution as simulation values, and obtain multiple three-dimensional models of U / TH values ​​in the study area.

[0014] Furthermore, the horizontal and vertical grid designs of the grid are 0.5~1m / grid, and the plane grid is 40x40m.

[0015] Furthermore, the variogram analysis includes first analyzing the sedimentary phase belts and layers separately to obtain the variogram model and characteristic parameters of the geological variables in the vertical and horizontal directions, and then summarizing the changing trends of reservoir facies and physical properties in various directions. After selecting the direction of data analysis using the modeling software, the experimental variogram can be automatically generated. Usually, better variogram parameters can be obtained by fitting the model along the first 2-4 points of the well direction variogram, thereby obtaining the range of variation and nugget value applicable to each small layer.

[0016] Furthermore, in step b, the trend of the data is analyzed. For the gridded U / TH ratio data of the known wells, the data distribution trends along the three directions of X, Y, and Z are analyzed using modeling software, and the data are fitted and the correlation coefficient is calculated. Usually, the correlation coefficient along the main direction of lithofacies development is higher.

[0017] Furthermore, the specific process of obtaining the values ​​of U and TH in step 1) is as follows: after the downhole instrument detects the formation gamma rays through the crystal, an energy spectrum is formed using a multi-channel analyzer, and the values ​​of U and TH can be obtained according to different energy spectrum peaks and counting rates.

[0018] Beneficial effects of the present invention: 1. The present invention proposes a method for jointly reconstructing the U-Th radioactive element ratio of a single well by three-dimensional modeling and neural network. By combining the model method with the neural network method, the shortcomings of each method are overcome and the advantages of both methods are fully utilized. It has been verified to be effective and reliable by sampling sparse wells.

[0019] 2. In the present invention, due to the difference in their chemical properties, the contents of Th (thorium) and U (uranium) change accordingly when the oxidation-reduction conditions change. Therefore, the U / TH ratio curve can reflect the trend of the oxidation-reduction conditions, and then invert the change of paleo-water depth. Generally, the larger the ratio, the stronger the reduction environment, and the deeper the corresponding paleo-water depth, and vice versa. It can be used to assist in fine geological stratification, sedimentary environment evolution analysis and lithofacies paleogeographic restoration, and help establish a fine stratigraphic framework for shale and evaluate reservoir sweet spots. Generally, when the U / TH ratio is >1.25, it is interpreted as a deep-water shelf, and when the U / TH ratio is <1.25, it is interpreted as a shallow-water shelf. Therefore, through the final comprehensive prediction curve calculated, 1) the interpretation of different stratigraphic sedimentary facies types and the change of sedimentary environment over time can be carried out; 2) according to the above changes, combined with the characteristics of sedimentary cycles, a basis for geological stratification is provided; 3) the potential strata of reservoir sweet spots are identified according to the distribution strata of favorable sedimentary facies.

[0020] 3. In the present invention, three-dimensional random modeling is a technology that uses computer graphics technology to generate three-dimensional quantitative random models based on comprehensive analysis of geological, logging, geophysical data and various interpretation results to accurately depict the three-dimensional distribution of underground geological attributes. It aims to describe the distribution of a certain attribute of the reservoir in space, and can better depict the distribution law and geometric form of underground geological attributes in three-dimensional space. It is an effective means of predicting reservoir attributes.

[0021] 4. In the present invention, the BP (back propagation) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm, and has an input layer, a hidden layer and an output layer; the GR and U / TH curves of the known wells are input into the neural network, and the mapping relationship between the GR and U / TH curves is established through computational learning, and then applied to the target well to predict and generate the U / TH curve.

[0022] 5. In the present invention, the thinned-out well refers to a well that does not participate in the above-mentioned model establishment, but uses the data of other wells to establish the model, and then compares the model data with the actual well data at the well location to verify the reliability of the model. After the model is established, the numerical value can be extracted along the well trajectory at the thinned-out well location according to the grid attribute values ​​​​crossed by the actual well trajectory to generate a curve.

[0023] 6. In the present invention, screening 3-4 high-quality models has basic probabilistic statistical significance, covering possible good, medium and poor different realizations; it has high computational timeliness. Usually, geological models are at the level of tens of millions of grids, and the time for running a simulation is usually 10 minutes. Then, it usually takes 30 minutes to 40 minutes to run 3-4 models once. If there are more than 4 models, the timeliness is too low. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The present invention is a flow chart of the method for jointly reconstructing the U-Th radioactive element ratio of a single well by three-dimensional modeling and neural network.

[0025] Figure 2 This is a U / TH curve diagram generated for a known well in Example 2 of the present invention.

[0026] Figure 3 This is the random modeling process and quality control diagram of the U / TH ratio of Example 2 of the present invention.

[0027] Figure 4 This is one of the random simulation implementations of the U / TH ratio of Example 2 of the present invention.

[0028] Figure 5 This is the second random simulation implementation of the U / TH ratio of Example 2 of the present invention.

[0029] Figure 6 This is the third random simulation implementation of the U / TH ratio of Example 2 of the present invention.

[0030] Figure 7 This is the fourth random simulation implementation of the U / TH ratio of Example 2 of the present invention.

[0031] Figure 8 This is a U / TH reconstruction curve diagram generated for the target thinning well in Example 2 of the present invention.

[0032] Fig. 9 This is the comprehensive prediction curve of Example 2 of the present invention. DETAILED DESCRIPTION

[0033] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention will be further described in detail with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their description are used to explain the present invention, but are not intended to limit the present invention.

[0034] Embodiment 1: like Figure 1 As shown, this embodiment provides a method for jointly reconstructing the U-Th radioactive element ratio of a single well by three-dimensional modeling and neural network, including the following steps: 1) Obtaining U and TH values: The values ​​of U and TH are obtained by processing and interpreting the natural gamma ray spectrum logging curves of three known wells. Specifically, after the downhole instrument detects the formation gamma rays through crystals, it uses a multi-channel analyzer to form an energy spectrum. According to different energy spectrum peaks and counting rates, the values ​​of U and TH can be obtained; 2) Draw U and TH curves: According to the values ​​obtained in step 1), calculate the U / TH ratio of each well and draw three U / TH ratio curves; 3) Establish a three-dimensional U / TH ratio model: Based on the U / TH ratio curve of the known wells, using geostatistical theory and three-dimensional stochastic modeling technology, multiple three-dimensional U / TH ratio models of the study area are established under the guidance and constraints of the lithofacies paleogeographic model. The specific process is: a. Gridding: Grid the U / TH curve of the known well. The grid density is related to the formation thickness and the level of refinement of the reservoir research. The horizontal and vertical grids in this example are designed to be 1m / grid, and the plane grid is 40x40m. The vertical density of the grid is the number of grids / reservoir thickness. The more refined the reservoir research requirements, the greater the grid density needs to be, and the greater the total number of grids in the model. b. Data analysis, such as Figure 3 As shown in the figure, after gridding, each grid is assigned a value to generate a data point. According to the geological characteristics of the data, the data is processed for normal distribution and variogram analysis. In the variogram analysis, the sedimentary phase belts and layers are first analyzed separately, with the aim of obtaining the variogram model and characteristic parameters of the geological variables in the vertical and horizontal directions, and then summarizing the variation trends of the reservoir facies and physical properties in various directions. After selecting the direction of data analysis using the modeling software, the experimental variogram can be automatically generated, thereby obtaining the range of variation and nugget value applicable to each small layer; c. Parameter simulation: make a normal transformation on the gridded U / TH value to make it obey the normal distribution; calculate the experimental variogram; select a random path through the grid cells to be simulated; perform simple Kriging estimation on the grids that have not been simulated; establish a normal distribution based on the estimation and variance, randomly extract values ​​from the distribution as simulation values, and obtain three three-dimensional models of the U / TH value of the study area, which represent various possible realizations of the underground U / TH ratio. The quality control of the simulation results includes the comparison between the actual wellbore data and the simulation results; the comparison between the simulation results of the whole area and the overall distribution histogram of the well data; and the control of the model quality through the matching degree of the comparison; 4) Reconstruct the U / TH ratio prediction curve process based on a single model: Select the target sparsely-drawn well, and generate a model-based U / TH ratio result curve on the sparsely-drawn well according to the model established in step 3), and then use this result curve and the GR curve in a joint supervision mode to calculate the U / TH ratio prediction curve through a neural network; Calculate the average relative deviation between the achievement curve and the prediction curve. If the average relative deviation between the achievement curve and the prediction curve is less than 5%, proceed to the next step. If the average relative deviation between the achievement curve and the prediction curve in step 4) is ≥ 5%, return to step 3), readjust the model parameters, and update the U / TH ratio model. 5) Reconstruct the U / TH ratio prediction curve process based on multiple models: Figure 6 As shown in the figure, according to the average relative deviation of the reconstructed curve and the actual curve <5%, three high-quality models were screened out and applied to the target wells to generate model-based reconstructed curves. Then, the achievement curve and the GR curve were jointly supervised in a neural network to calculate the prediction curves (UTh1, UTh2, UTh3). Finally, the comprehensive prediction curve was calculated: UTh0=(UTh1+UTh2+UTh3) / 3.

[0035] Embodiment 2: This example takes the Longmaxi Formation shale gas well in the Weiyuan shale block in the Sichuan Basin as an example: This embodiment provides Figure 1 The method of jointly reconstructing the U-Th radioactive element ratio of a single well by three-dimensional modeling and neural network shown in the figure includes the following steps: 1) obtaining the U and TH values: obtaining the values ​​of U and TH by processing and interpreting the natural gamma ray spectrum logging curves of four known wells. Specifically, after the downhole instrument detects the formation gamma rays through the crystal, it uses a multi-channel analyzer to form an energy spectrum. According to different energy spectrum peaks and counting rates, the values ​​of U and TH can be obtained; 2) Draw U and TH curves: According to the values ​​obtained in step 1), calculate the U / TH ratio of each well and draw the following curves: Figure 2 The 4 U / TH ratio curves shown; 3) Establish a three-dimensional U / TH ratio model: Based on the U / TH ratio curve of the known wells, using geostatistical theory and three-dimensional stochastic modeling technology, multiple three-dimensional U / TH ratio models of the study area are established under the guidance and constraints of the lithofacies paleogeographic model. The specific process is: a. Gridding: Grid the U / TH curve of the known wells. The grid density is related to the formation thickness and the level of refinement of the reservoir research. The horizontal and vertical grids in this example are designed to be 0.5m / grid, and the plane grid is 40x40m. The vertical density of the grid is the number of grids / reservoir thickness. The more refined the reservoir research requirements, the greater the grid density needs to be, and the greater the total number of grids in the model. b. Data analysis, such as Figure 3 As shown in the figure, after gridding, each grid is assigned a value to generate a data point. The number of grids in the entire work area reaches 30,121,416 grids, and the data set also reaches 30 million. According to the geological characteristics of the data, the data is processed for normal distribution and variogram analysis. In the variogram analysis, the sedimentary phase belts and layers are first analyzed separately, with the aim of obtaining the variogram model and characteristic parameters of the geological variables in the vertical and horizontal directions, and then summarizing the variation trends of the reservoir facies and physical properties in various directions. After selecting the direction of data analysis using the modeling software, the experimental variogram can be automatically generated. The model fitting can be performed along the first 2-4 points of the variogram in the well direction to obtain better variogram parameters, thereby obtaining the range of variation and nugget value applicable to each small layer; The trend of the data is analyzed. For the U / TH ratio data of known wells after gridding, the data distribution trends along the three directions of X, Y, and Z are analyzed using modeling software. Data fitting is performed and correlation coefficients are calculated. Usually, the correlation coefficient along the main direction of lithofacies development is higher. The reservoir contains various sedimentary phases and high permeability layers caused by microcracks. Since the variogram is calculated by the sum of squared differences, the singular values ​​in the data have a prominent influence on the variogram.

[0036] c. Parameter simulation: perform normal transformation on the gridded U / TH values ​​to make them obey normal distribution; calculate the experimental variogram; select random paths through the grid cells to be simulated; perform simple Kriging estimation on the grids that have not been simulated; establish a normal distribution based on the estimation and variance, randomly extract values ​​from the distribution as simulation values, and obtain multiple three-dimensional models of U / TH values ​​in the study area, such as Figure 4-7 The random simulation results of the U / TH ratio shown represent various possible realizations of the underground U / TH ratio. The quality control of the simulation results includes the comparison between the actual wellbore data and the simulation results; the comparison between the simulation results of the whole area and the overall distribution histogram of the well data; and the control of the model quality through the matching degree of the comparison. 4) Reconstruct the U / TH ratio prediction curve process based on a single model: Figure 8As shown, a target sparsely pumped well is selected, and according to the model established in step 3), a model-based U / TH ratio achievement curve is generated on the sparsely pumped well, and then the U / TH ratio prediction curve is calculated by a neural network using the achievement curve and the GR curve in a joint supervision mode; The specific process of neural network calculation is to input the GR and U / TH achievement curves of known wells, establish the mapping relationship between GR and U / TH curves through computational learning, and then apply it to the target well to predict and generate the U / TH ratio achievement curve; Calculate the average relative deviation between the achievement curve and the prediction curve. If the average relative deviation between the achievement curve and the prediction curve is <5%, proceed to the next step. When the average relative deviation between the achievement curve and the prediction curve in step 4) is ≥5%, return to step 3), readjust the model parameters, and update the U / TH ratio model. The specific process of adjusting the model parameters is to return to the model data analysis module, refit the horizontal range and vertical range of the variation function of the multi-well U / TH data to optimize the range value. The specific process of updating the U / TH ratio model is to apply the re-optimized parameter-range, primary axis and secondary axis values ​​to the static three-dimensional simulation to generate a new three-dimensional attribute model. 5) Reconstruct the U / TH ratio prediction curve process based on multiple models: Figure 6 As shown in Figure 2, based on the average relative deviation of the reconstructed curve and the actual curve of <5%, four high-quality models were selected and applied to the target wells to generate model-based reconstructed curves. Then, the neural network was used to calculate the prediction curves (UTh1, UTh2, UTh3, and UTh4) using the joint supervision mode of the achievement curve and the GR curve. Finally, the following was calculated: Fig. 9 The comprehensive forecast curve shown is: UTh0=(UTh1+UTh2+UTh3+UTh4) / 4.

[0037] Usually, when the U / TH ratio is >1.25, it is interpreted as a deep-water shelf, and when the U / TH ratio is <1.25, it is interpreted as a shallow-water shelf. Therefore, the final comprehensive prediction curve calculated can: 0) provide an alternative prediction curve for development wells that cannot be logged in areas with complex well conditions; 1) interpret the sedimentary phase types of different formations and the changes in the sedimentary environment over time; 2) provide a basis for geological stratification based on the above changes and combined with the characteristics of sedimentary cycles; 3) identify the potential sweet spots of the reservoir according to the distribution of favorable sedimentary phases. Fig. 9As shown in the figure, the U / TH ratio of layer1 at the bottom of the Longmaxi Formation is greater than 1.25, representing a strong reducing environment, belonging to a deep-water shelf, and rich in organic matter; layer2 U / TH is less than 1.25, the water body becomes shallower, representing a weak reducing environment, and the organic matter content decreases; layer3 U / TH rises to greater than 1.25, the water body becomes deeper during this period, representing a strong reducing environment, and the organic matter content also increases compared with layer2; layer4 U / TH is less than 1.25, this period is the sea regression stage, the water body becomes shallower as a whole, representing a shallow-water shelf, and the organic matter content decreases as a whole.

[0038] In summary, the bottom of the Longmaxi Formation can be divided into four small layers according to the changes in the U / TH ratio and the law of cyclic changes. Among them, Layer 1 has the deepest water body and a strong reducing environment that is more conducive to the enrichment of organic matter. It is a sweet spot for shale gas development.

[0039] Although the specific implementation of the present invention is described in detail in conjunction with the drawings, it should not be understood as limiting the scope of protection of this patent. Within the scope described in the claims, various modifications and variations that can be made by those skilled in the art without creative work still fall within the scope of protection of this patent.

[0040] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for jointly reconstructing the U-TH radioactive element ratio of a single well by three-dimensional modeling and neural network, characterized in that: The steps include: 1) Obtaining U and TH values: Obtaining the values ​​of U and TH by processing and interpreting the natural gamma ray spectrum logging curves of multiple known wells; 2) Draw U and TH curves: According to the values ​​obtained in step 1), calculate the U / TH ratio of each well and draw the U / TH ratio curve; 3) Establish a three-dimensional U / TH ratio model: Based on the U / TH ratios of known wells, multiple three-dimensional U / TH ratio models of the study area were established under the guidance and constraints of the lithofacies paleogeographic model by using geostatistical theory and three-dimensional stochastic modeling technology; 4) Reconstruct the U / TH ratio prediction curve process based on a single model: Select the target sparsely-drawn well, and generate a model-based U / TH ratio result curve on the sparsely-drawn well according to the model established in step 3), and then use this result curve and the GR curve in a joint supervision mode to calculate the U / TH ratio prediction curve through a neural network; 5) Reconstruction process of U / TH ratio prediction curve based on multiple models: 3-4 high-quality models are selected and applied to the target wells to generate model-based reconstruction curves. Then, the neural network is used to calculate the prediction curve using the joint supervision mode of the achievement curve and the GR curve. Finally, the average value of each prediction curve is calculated, and the result is a comprehensive prediction curve.

2. The method according to claim 1, characterized in that: In step 3), multiple U / TH ratio three-dimensional models of the study area are established. The specific process is as follows: a. Curve gridding: grid the U / TH curve of the input known well. The grid density is related to the formation thickness and the level of refinement of the reservoir research. The vertical density of the grid is the number of grids / reservoir thickness. The more refined the reservoir research requirements, the greater the grid density needs to be, and the greater the total number of grids in the model. b. Data analysis: after gridding, assign a value to each grid to generate a data point. According to the geological characteristics of the data, the data is processed for normal distribution and variogram analysis to generate the range of variation and nugget value applicable to each small layer; c. Parameter simulation: make a normal transformation on the gridded U / TH value to make it obey the normal distribution; calculate the experimental variogram; Select a random path through the grid cells to be simulated; perform simple Kriging estimation on the grids that have not been simulated; establish a normal distribution based on the estimation and variance, randomly extract values ​​from the distribution as simulation values, and obtain multiple three-dimensional models of U / TH values ​​in the study area.

3. The method according to claim 1, characterized in that: The specific process of the neural network calculation in step 4) is to input the GR and U / TH achievement curves of the known well, establish the mapping relationship between the GR and U / TH curves through computational learning, and then apply it to the target well to predict and generate the U / TH ratio achievement curve.

4. The method according to claim 1, characterized in that: When the average relative deviation between the achievement curve and the prediction curve in step 4) is less than 5%, proceed to the next step; when the average relative deviation between the achievement curve and the prediction curve in step 4) is ≥5%, return to step 3), readjust the model parameters, and update the U / TH ratio model.

5. The method according to claim 4, characterized in that: The specific process of adjusting the model parameters is to return to the model data analysis module and refit the horizontal range and vertical range of the variogram of the multi-well U / TH data to optimize the range values.

6. The method according to claim 4, characterized in that: The specific process of updating the U / TH ratio model is to apply the re-optimized parameters - the range, the values ​​of the major axis and the minor axis - to the static three-dimensional simulation to generate a new three-dimensional attribute model.

7. The method according to claim 2, characterized in that: The horizontal and vertical grid designs of the gridding are 0.5-1 m / grid, and the plane grid is 40×40 m.

8. The method according to claim 2, characterized in that: The variogram analysis includes firstly analyzing the sedimentary phase belts and layers separately, obtaining the variogram model and characteristic parameters of the geological variables in the vertical and horizontal directions, and then summarizing the changing trends of the reservoir facies and physical properties in various directions. After selecting the direction of data analysis using the modeling software, the experimental variogram can be automatically generated. The model fitting is performed along the first 2-4 points of the well direction variogram to obtain better variogram parameters, thereby obtaining the range of variation and nugget value applicable to each small layer.

9. The method according to claim 2, characterized in that: The step b also includes analyzing the trend of the data. For the gridded U / TH ratio data of the known wells, the data distribution trends along the three directions of X, Y, and Z are analyzed using modeling software, and the data are fitted and the correlation coefficient is calculated.

10. The method according to claim 1, characterized in that: The specific process of obtaining the values ​​of U and TH in step 1) is as follows: after the downhole instrument detects the formation gamma rays through the crystal, an energy spectrum is formed using a multi-channel analyzer, and the values ​​of U and TH can be obtained according to different energy spectrum peaks and counting rates.

Images