Method, processor, and storage medium for intelligent prediction of reservoir sandstone thickness

Abstract

The embodiment of the application provides a method, a processor and a storage medium for intelligent prediction of reservoir sandstone thickness, and belongs to the technical field of intelligent prediction of sandstone thickness. The method for intelligent prediction of reservoir sandstone thickness comprises the following steps: obtaining a first sandstone thickness prediction result of a to-be-predicted point according to seismic attribute characteristics based on a predetermined first sandstone thickness prediction model; obtaining a second sandstone thickness prediction result of the to-be-predicted point according to a three-dimensional seismic data body based on a predetermined second sandstone thickness prediction model; obtaining a riverway facies distribution map corresponding to the to-be-predicted point according to well point phase types, first trend constraint surface data and second trend constraint surface data; and obtaining a predicted sandstone thickness of the to-be-predicted point according to the first sandstone thickness prediction result, the second sandstone thickness prediction result, the riverway facies distribution map and an actual sandstone thickness. The application can realize intelligent prediction of reservoir sandstone thickness.

Description

Technical Field

[0001] This invention relates to the field of intelligent prediction technology for sandstone thickness, and more specifically to a method, processor, and storage medium for intelligent prediction of reservoir sandstone thickness. Background Technology

[0002] Sandstone reservoirs, as an important type of concealed oil and gas reservoir, are mostly developed in the gentle slope zones of continental rift basins and lacustrine basins. They are known for their large oil-bearing area and moderate reserve abundance, and are increasingly attracting attention in exploration and development. Thin interbedded reservoirs are composed of different terrigenous sediments, with fine sandstone as the main lithology and local sandy mudstone. They are mainly structural-stratigraphic traps and are light oil reservoirs. Therefore, the prediction of sandstone thickness becomes more important. Existing technologies exist for exploring sandstone thickness at well locations within the reservoir, but there are no existing technologies for exploring sandstone thickness at non-well location locations within the reservoir.

[0003] Therefore, how to predict the sandstone thickness at non-wellpoint locations in the reservoir has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this application is to provide a method, processor, and storage medium for intelligent prediction of sandstone thickness in reservoirs, so as to solve the problem of how to intelligently predict the thickness of sandstone at non-wellpoint locations in reservoirs in the prior art.

[0005] To achieve the above objectives, the first aspect of this application provides a method for intelligent prediction of reservoir sandstone thickness, the method comprising:

[0006] The actual sandstone thickness at each depth of the target well in the target reservoir, the three-dimensional seismic data volume of the point to be predicted in the reservoir, the seismic attribute characteristics of the point to be predicted, and the well point facies type corresponding to the target well are obtained. The point to be predicted is the point in the target reservoir other than the target well.

[0007] Based on the predetermined first sandstone thickness prediction model, the predicted first sandstone thickness at the point to be predicted is obtained according to the seismic attribute characteristics.

[0008] Based on the predetermined second sandstone thickness prediction model, the second sandstone thickness prediction result at the point to be predicted is obtained according to the three-dimensional seismic data volume.

[0009] Trend constraint surface transformation is performed on the first sandstone thickness prediction results and the second sandstone thickness prediction results respectively to obtain the first trend constraint surface data corresponding to the first sandstone thickness prediction results and the second trend constraint surface data corresponding to the second sandstone thickness prediction results.

[0010] Based on the well point facies type, the first trend constraint surface data, and the second trend constraint surface data, the channel facies distribution map corresponding to the point to be predicted is obtained;

[0011] Based on the predicted thickness of the first sandstone, the predicted thickness of the second sandstone, the channel facies distribution map, and the actual sandstone thickness, the predicted sandstone thickness at the point to be predicted is obtained.

[0012] In this embodiment, the predicted sandstone thickness of a point to be predicted is obtained based on the predicted thickness of the first sandstone, the predicted thickness of the second sandstone, the channel facies distribution map, and the actual sandstone thickness. This includes: determining the product of the actual sandstone thickness at the depth location of the point to be predicted (where the distance to the point to be predicted is less than a preset distance threshold) and a preset sandstone thickness weight, and summing the product values ​​to obtain the actual sandstone thickness influence value; determining the product of the predicted thickness of the first sandstone at each point to be predicted (where the distance to the point to be predicted is less than a preset distance threshold) and a preset first sandstone thickness weight, and summing the product values ​​to obtain the first sandstone thickness. The influence value of the thickness prediction result is calculated as follows: The product of the second sandstone thickness prediction result and the preset second sandstone thickness weight for each prediction point whose distance from the prediction point is less than a preset distance threshold is determined, and the product values ​​are summed to obtain the influence value of the second sandstone thickness prediction result; The product of the channel facies and the preset channel facies weight for each prediction point whose distance from the prediction point is less than a preset distance threshold in the channel facies distribution map is determined, and the product values ​​are summed to obtain the channel facies influence value; The influence values ​​of the actual sandstone thickness, the first sandstone thickness prediction result, the second sandstone thickness prediction result, and the channel facies influence value are added together to obtain the predicted sandstone thickness.

[0013] In this embodiment, determining the first sandstone thickness prediction model includes: obtaining a first initial sandstone thickness prediction model, wherein the first initial sandstone thickness prediction model includes multiple first initial sandstone thickness prediction processing layers, each first initial sandstone thickness prediction processing layer includes multiple first basic thickness prediction models, and the input of the current processing layer is the average value of all basic thickness prediction models in the previous processing layer; obtaining the planar seismic attribute characteristics and actual sandstone thickness at each depth location of the target well; based on the first initial sandstone thickness prediction model, and according to the planar seismic attribute characteristics, determining the first sandstone thickness prediction value at each depth location of the target well output by each first basic thickness prediction model in each first initial sandstone thickness prediction processing layer; determining the average value of all first sandstone thickness prediction values ​​corresponding to each first initial sandstone thickness prediction processing layer to obtain the average value of the first sandstone thickness prediction value corresponding to each first initial sandstone thickness prediction processing layer; obtaining the first dispersion value corresponding to each first initial sandstone thickness prediction model processing layer based on the first sandstone thickness prediction value and the average value of the first sandstone thickness prediction value; and obtaining the first dispersion value corresponding to each first initial sandstone thickness prediction processing layer. Based on all the predicted and actual sandstone thicknesses, the first deviation coefficient value corresponding to each first initial sandstone thickness prediction processing layer is determined. The first evaluation value corresponding to each first initial sandstone thickness prediction processing layer is determined by multiplying the first dispersion value and the first deviation coefficient value. Based on the first evaluation value, each first basic thickness prediction model in all first initial sandstone thickness prediction processing layers is sequentially screened until all first basic thickness prediction models in all first initial sandstone thickness prediction processing layers have been screened. If a first model is found, it is retained; if a second model is found, it is discarded to obtain the first sandstone thickness prediction model. The first model is the first basic thickness prediction model whose first evaluation value after removing the first basic thickness prediction model is less than the first evaluation value after retaining the first basic thickness prediction model. The second model is the first basic thickness prediction model whose first evaluation value after removing the first basic thickness prediction model is greater than or equal to the first evaluation value after retaining the first basic thickness prediction model.

[0014] In this embodiment of the application, the method further includes: according to the accuracy index corresponding to the first initial sandstone thickness prediction processing layer, sequentially screening multiple first initial sandstone thickness prediction processing layers in the first sandstone thickness prediction model; retaining the first processing layer when the first processing layer is screened; removing the second processing layer and stopping the screening when the second processing layer is screened, so as to obtain the first sandstone thickness prediction model. The first processing layer is the processing layer with an accuracy index greater than or equal to the previous processing layer, and the second processing layer is the processing layer with an accuracy index less than the previous processing layer. The accuracy index is the ratio of the number of first basic thickness prediction models in the first initial sandstone thickness prediction processing layer that meet the preset prediction standard to the total number of all first basic thickness prediction models in the first initial sandstone thickness prediction processing layer. The preset prediction standard is that the difference between the predicted value of the first sandstone thickness and the actual sandstone thickness is less than a preset deviation threshold.

[0015] In this embodiment of the application, determining the first deviation coefficient value corresponding to each first initial sandstone thickness prediction treatment layer based on all predicted values ​​of first sandstone thickness and actual sandstone thickness for each first initial sandstone thickness prediction treatment layer includes: determining the sum of deviations between all predicted values ​​of first sandstone thickness and actual sandstone thickness for each treatment layer; adding a preset constant to the sum of deviations to obtain a corrected sum of deviations; and taking the reciprocal of the corrected sum of deviations to obtain the first deviation coefficient value corresponding to each treatment layer.

[0016] In this embodiment, determining the second sandstone thickness prediction model includes: obtaining a second initial sandstone thickness prediction model, wherein the second initial sandstone thickness prediction model includes multiple second initial sandstone thickness prediction processing layers, each second initial sandstone thickness prediction processing layer includes multiple second basic thickness prediction models, and the input of the current processing layer is the average value of all basic thickness prediction models in the previous processing layer; obtaining the target well seismic data characteristics and actual sandstone thickness at each depth location of the target well; based on the second initial sandstone thickness prediction model, and according to the target well seismic data characteristics, determining the second sandstone thickness prediction value output by each second basic thickness prediction model in each second initial sandstone thickness prediction processing layer; determining the average value of all sandstone thickness prediction values ​​corresponding to each second initial sandstone thickness prediction processing layer, obtaining the average value of the second sandstone thickness prediction value corresponding to each second initial sandstone thickness prediction processing layer; obtaining the second dispersion value corresponding to each second initial sandstone thickness prediction processing layer based on the second sandstone thickness prediction value and the average value of the second sandstone thickness prediction value; and further processing the second initial sandstone thickness prediction model. Based on the predicted and actual sandstone thicknesses of all layers, the second deviation coefficient value corresponding to each second initial sandstone thickness prediction treatment layer is determined. Based on the second dispersion value and the second deviation coefficient value, the second evaluation value corresponding to each second initial sandstone thickness prediction treatment layer is determined. Based on the second evaluation value, each second basic thickness prediction model in all second initial sandstone thickness prediction treatment layers is sequentially screened until all second basic thickness prediction models in all second initial sandstone thickness prediction treatment layers have been screened. If a third model is found, it is retained; if a fourth model is found, it is discarded to obtain the second sandstone thickness prediction model. The third model is the second basic thickness prediction model whose second evaluation value after removing all second basic thickness prediction models is less than the second evaluation value after retaining the second basic thickness prediction model. The fourth model is the second basic thickness prediction model whose second evaluation value after removing all second basic thickness prediction models is greater than or equal to the second evaluation value after retaining the second basic thickness prediction model.

[0017] In this embodiment of the application, the method further includes: according to the accuracy index corresponding to the second initial sandstone thickness prediction processing layer, sequentially screening multiple second initial sandstone thickness prediction processing layers in the second sandstone thickness prediction model; retaining the third processing layer when the third processing layer is screened; removing the fourth processing layer and stopping the screening when the fourth processing layer is screened, so as to obtain the first sandstone thickness prediction model. The third processing layer is the second initial sandstone thickness prediction processing layer with an accuracy index greater than or equal to that of the previous processing layer, and the fourth processing layer is the second initial sandstone thickness prediction processing layer with an accuracy index less than that of the previous processing layer. The accuracy index is the ratio of the number of second basic thickness prediction models in the second initial sandstone thickness prediction processing layer that meet the preset prediction standard to the total number of all second basic thickness prediction models in the second initial sandstone thickness prediction processing layer. The preset prediction standard is that the difference between the predicted value of the second sandstone thickness and the actual sandstone thickness is less than a preset deviation threshold.

[0018] In this embodiment of the application, the method further includes: obtaining the cumulative porosity, permeability, and lithology sensitivity curve of the target reservoir; and classifying the target reservoir into oil production grades based on a clustering algorithm, according to the cumulative porosity, permeability, lithology sensitivity curve, and predicted sandstone thickness.

[0019] A second aspect of this application provides an apparatus for intelligent prediction of unconventional reservoir productivity. The apparatus includes: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the method described above for intelligent prediction of reservoir sandstone thickness.

[0020] A third aspect of this application provides a machine-readable storage medium, characterized in that the machine-readable storage medium stores instructions for causing a machine to execute the above-described method for intelligent prediction of reservoir sandstone thickness.

[0021] The above technical solution, based on a predetermined first sandstone thickness prediction model, obtains the first sandstone thickness prediction result of the point to be predicted according to the seismic attribute characteristics of the point to be predicted. Based on a predetermined second sandstone thickness prediction model, it obtains the second sandstone thickness prediction result of the point to be predicted according to the three-dimensional seismic data volume of the point to be predicted. This application performs trend constraint surface conversion on the first sandstone thickness prediction result and the second sandstone thickness prediction result obtained by the above method to obtain the first trend constraint surface data corresponding to the first sandstone thickness prediction result and the second trend constraint surface data corresponding to the second sandstone thickness prediction result. Then, based on the well point facies type, the first trend constraint surface data and the second trend constraint surface data, the channel facies distribution map corresponding to the point to be predicted is predicted. Finally, based on the first sandstone thickness prediction result, the second sandstone thickness prediction result, the channel facies distribution map and the actual sandstone thickness, the predicted sandstone thickness of the point to be predicted is obtained. This application integrates the predicted results of the first sandstone thickness, the predicted results of the second sandstone thickness, the channel facies distribution map, and the actual sandstone thickness to predict the sandstone thickness. Therefore, this application realizes the prediction of sandstone thickness at the point to be predicted in the reservoir and improves the prediction accuracy of sandstone thickness at the point to be predicted in the reservoir. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0023] Figure 1 The schematic diagram illustrates a process for predicting reservoir sandstone thickness in one embodiment of this application;

[0024] Figure 2 A schematic diagram illustrating reservoir sand body thickness prediction in one embodiment of this application is shown.

[0025] Figure 3 The illustration shows a schematic diagram of reservoir oil production (sweet spot) prediction in one embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0028] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0029] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0030] Figure 1 The illustration shows a flowchart of intelligent prediction of reservoir sandstone thickness in one embodiment of this application. Figure 1 As shown in the figure, this application provides a method for intelligent prediction of reservoir sandstone thickness. Taking the application of this method to a processor as an example, the method may include the following steps:

[0031] Step S101: Obtain the actual sandstone thickness at each depth of the target well in the target reservoir, the three-dimensional seismic data volume of the point to be predicted in the reservoir, the seismic attribute characteristics of the point to be predicted, and the well point facies type corresponding to the target well. The point to be predicted is the point in the target reservoir other than the target well.

[0032] Step S102: Based on the predetermined first sandstone thickness prediction model, the first sandstone thickness prediction result of the point to be predicted is obtained according to the seismic attribute characteristics.

[0033] Step S103: Based on the predetermined second sandstone thickness prediction model, the second sandstone thickness prediction result of the point to be predicted is obtained according to the three-dimensional seismic data volume.

[0034] Step S104: Perform trend constraint surface transformation on the first sandstone thickness prediction result and the second sandstone thickness prediction result respectively to obtain the first trend constraint surface data corresponding to the first sandstone thickness prediction result and the second trend constraint surface data corresponding to the second sandstone thickness prediction result.

[0035] Step S105: Based on the well point facies type, the first trend constraint surface data, and the second trend constraint surface data, obtain the channel facies distribution map corresponding to the point to be predicted.

[0036] Step S106: Based on the predicted thickness of the first sandstone, the predicted thickness of the second sandstone, the channel facies distribution map, and the actual sandstone thickness, the predicted sandstone thickness of the point to be predicted is obtained.

[0037] It can be understood that a 3D seismic data volume is a data volume that records the seismic reflection information of the subsurface medium in a three-dimensional spatial form. The 3D seismic data volume can be extracted from the seismic volume using Fast Fourier Transform or Wavelet Transform to obtain 3D attribute volumes of different frequencies. The seismic volume is exploration data obtained from oilfield seismic exploration. Seismic attribute features are one-dimensional attribute features of earthquakes. These features can be obtained by decomposing the 3D seismic data volume, including maximum amplitude, average instantaneous phase, root mean square amplitude, coherence, curvature, etc. Alternatively, seismic trace data from adjacent trace numbers around the point to be predicted can be extracted. The similarity between the target layer seismic waveform of the point to be predicted and the seismic waveforms of adjacent traces can be calculated. The similarity index between different traces is calculated using the Dynamic Time Warping (DTW) method. The threshold T is automatically calculated using the Otsu method. Based on the threshold, seismic traces with similar target layer seismic waveforms to the seismic traces of the point to be predicted are determined. The seismic attributes of the seismic traces with similar target layer seismic waveforms to the seismic traces of the point to be predicted are calculated, and the average value is taken as the seismic attribute feature of the seismic trace where the point to be predicted is located. The reservoir is divided into wellpoint and non-wellpoint facies types. Wellpoint facies types include channel facies type (assigned a value of 1) or non-channel facies type (assigned a value of 0). The point to be predicted is a non-wellpoint. The first sandstone thickness prediction model is a model that predicts sandstone thickness based on seismic attribute characteristics. The second sandstone thickness prediction model is a model that predicts sandstone thickness based on 3D seismic data. The first sandstone thickness prediction result is the sandstone thickness of the point to be measured predicted according to the first sandstone thickness prediction model. The second sandstone thickness prediction result is the sandstone thickness of the point to be measured predicted according to the second sandstone thickness prediction model. The first trend constraint surface data is the trend constraint surface data corresponding to the first sandstone thickness prediction result. The second trend constraint surface data is the trend constraint surface data corresponding to the second sandstone thickness prediction result. The channel facies distribution map is a map used to display the distribution of different channel facies types and their related sedimentary characteristics in a specific area. The processor obtains the channel facies distribution map of the reservoir based on the wellpoint facies type, the first trend constraint surface data, and the second trend constraint surface data. The actual sandstone thickness is the sandstone thickness at the wellpoint location extracted from the single-well stratification data. The predicted sandstone thickness is the predicted sandstone thickness at the point to be predicted.

[0038] Specifically, the processor first acquires the 3D seismic data volume and seismic attribute characteristics of the point to be predicted, so as to obtain the corresponding sandstone thickness prediction results based on the first sandstone thickness prediction model and the second sandstone thickness prediction model. The actual sandstone thickness obtained is the actual sandstone thickness of well points other than the point to be predicted. The facies type of the target well is obtained to determine the channel facies distribution map. Finally, the processor can comprehensively obtain the predicted sandstone thickness of the point to be predicted in the reservoir based on the first sandstone thickness prediction result, the second sandstone thickness prediction result, the actual sandstone thickness, and the channel facies distribution map, thus achieving the effect of predicting the sandstone thickness of the point to be predicted in the reservoir. At the same time, since the predicted sandstone thickness is not based on only one model, but integrates the first sandstone thickness prediction result, the second sandstone thickness prediction result, the actual sandstone thickness, and the channel facies distribution map, the technical effect of improving the prediction accuracy of the sandstone thickness of the point to be predicted in the reservoir is achieved.

[0039] The above technical solution, based on a predetermined first sandstone thickness prediction model, obtains the first sandstone thickness prediction result of the point to be predicted according to the seismic attribute characteristics of the point to be predicted. Based on a predetermined second sandstone thickness prediction model, it obtains the second sandstone thickness prediction result of the point to be predicted according to the three-dimensional seismic data volume of the point to be predicted. This application performs trend constraint surface conversion on the first sandstone thickness prediction result and the second sandstone thickness prediction result obtained by the above method to obtain the first trend constraint surface data corresponding to the first sandstone thickness prediction result and the second trend constraint surface data corresponding to the second sandstone thickness prediction result. Then, based on the well point facies type, the first trend constraint surface data and the second trend constraint surface data, the channel facies distribution map corresponding to the point to be predicted is predicted. Finally, based on the first sandstone thickness prediction result, the second sandstone thickness prediction result, the channel facies distribution map and the actual sandstone thickness, the predicted sandstone thickness of the point to be predicted is obtained. This application integrates the predicted results of the first sandstone thickness, the predicted results of the second sandstone thickness, the channel facies distribution map, and the actual sandstone thickness to predict the sandstone thickness. Therefore, this application realizes the prediction of sandstone thickness at the point to be predicted in the reservoir and improves the prediction accuracy of sandstone thickness at the point to be predicted in the reservoir.

[0040] In this embodiment, the predicted sandstone thickness of a point to be predicted is obtained based on the predicted thickness of the first sandstone, the predicted thickness of the second sandstone, the channel facies distribution map, and the actual sandstone thickness. This includes: determining the product of the actual sandstone thickness at the depth location of the point to be predicted (where the distance to the point to be predicted is less than a preset distance threshold) and a preset sandstone thickness weight, and summing the product values ​​to obtain the actual sandstone thickness influence value; determining the product of the predicted thickness of the first sandstone at each point to be predicted (where the distance to the point to be predicted is less than a preset distance threshold) and a preset first sandstone thickness weight, and summing the product values ​​to obtain the first sandstone thickness. The influence value of the thickness prediction result is calculated as follows: The product of the second sandstone thickness prediction result and the preset second sandstone thickness weight for each prediction point whose distance from the prediction point is less than a preset distance threshold is determined, and the product values ​​are summed to obtain the influence value of the second sandstone thickness prediction result; The product of the channel facies and the preset channel facies weight for each prediction point whose distance from the prediction point is less than a preset distance threshold in the channel facies distribution map is determined, and the product values ​​are summed to obtain the channel facies influence value; The influence values ​​of the actual sandstone thickness, the first sandstone thickness prediction result, the second sandstone thickness prediction result, and the channel facies influence value are added together to obtain the predicted sandstone thickness.

[0041] It can be understood that the preset distance threshold is a distance threshold pre-set by the processor to obtain the predicted results of the first sandstone thickness, the second sandstone thickness, the channel facies, and the actual sandstone thickness of well points within a certain range for the points to be predicted. The preset sandstone thickness weight can be pre-calculated based on the co-kriging equation and represents the weight value of the actual sandstone thickness. The preset first sandstone thickness weight is pre-calculated based on the co-kriging equation and represents the weight value of the first sandstone thickness prediction result. The preset second sandstone thickness weight is pre-calculated based on the co-kriging equation and represents the weight value of the second sandstone thickness prediction result. The preset channel facies weight can be pre-calculated based on the co-kriging equation and represents the weight value of the channel facies. The actual sandstone thickness influence value is the influence value of the actual sandstone thickness on the predicted sandstone thickness. The first sandstone thickness prediction result influence value is the influence value of the first sandstone thickness prediction result on the predicted sandstone thickness. The second sandstone thickness prediction result influence value is the influence value of the second sandstone thickness prediction result on the predicted sandstone thickness. The channel facies influence value is the influence value of the channel facies data within the preset distance threshold range for the points to be predicted. The first sandstone thickness prediction result is the sandstone thickness at the measured point predicted based on the first sandstone thickness prediction model. The second sandstone thickness prediction result is the sandstone thickness at the measured point predicted based on the second sandstone thickness prediction model. The channel facies distribution map is a distribution map of channel facies obtained based on well point facies type, first trend constraint surface data, and second trend constraint surface data.

[0042] Specifically, the processor first determines the actual sandstone thickness at the depth locations of the prediction points whose distance to the prediction point is less than a preset distance threshold. It then calculates the influence value of each actual sandstone thickness based on the product of the preset sandstone thickness weights corresponding to each actual sandstone thickness and the actual sandstone thickness itself. These values ​​are then summed to obtain the actual sandstone thickness influence value. Next, the processor determines the first sandstone thickness prediction result for each prediction point whose distance to the prediction point is less than the preset distance threshold. It then calculates the influence value of each first sandstone thickness prediction result based on the product of the preset first sandstone thickness weights corresponding to each first sandstone thickness prediction result and the summed to obtain the first sandstone thickness prediction result influence value. Finally, the processor determines the influence value of each prediction point whose distance to the prediction point is less than the preset distance threshold. The processor calculates the influence value of each second sandstone thickness prediction result based on the product of the preset second sandstone thickness weights corresponding to the second sandstone thickness prediction results for each prediction point, and adds them together to obtain the influence value of the second sandstone thickness prediction result. The processor then identifies channel facies in the channel facies distribution map that are less than a preset distance threshold from the prediction point, calculates the influence value of each channel facies based on the product of the preset channel facies weights corresponding to each channel, and adds them together to obtain the channel facies influence value. Finally, the processor adds the previously calculated influence values ​​of the actual sandstone thickness, the first sandstone thickness prediction result, the second sandstone thickness prediction result, and the channel facies influence value to obtain the predicted sandstone thickness for each prediction point. Furthermore, a channel model index is established. ,in, for Predicted phase value of well point This represents the predicted average phase value at well points. This represents 20% of the well point facies data. Simulation goodness index. ,in express The error between the predicted and actual phase values ​​at well points is calculated, with the input well point facies data used as the actual phase value. Using a collaborative geostatistical simulation algorithm, 80% of the well point sand thickness Z is used as hard data. The predicted thicknesses of the first and second sandstone layers, along with the channel facies distribution map, are used as trend data to obtain the predicted sandstone thickness. The reserved 20% of hard data is used for comparison with channel model indices. and simulation goodness index The product and the lower bound of error ,if The hyperparameters of the collaborative geostatistical simulation algorithm are iteratively optimized; otherwise, the optimal comprehensive sandstone thickness for the corresponding target layer is obtained.

[0043] In one embodiment, determining the first sandstone thickness prediction model includes: acquiring a first initial sandstone thickness prediction model, wherein the first initial sandstone thickness prediction model includes multiple first initial sandstone thickness prediction processing layers, each first initial sandstone thickness prediction processing layer includes multiple first basic thickness prediction models, and the input of the current processing layer is the average value of all basic thickness prediction models in the previous processing layer; acquiring the planar seismic attribute characteristics and actual sandstone thickness at each depth location of the target well; based on the first initial sandstone thickness prediction model, and according to the planar seismic attribute characteristics, determining the first sandstone thickness prediction value at each depth location of the target well output by each first basic thickness prediction model in each first initial sandstone thickness prediction processing layer; determining the average value of all first sandstone thickness prediction values ​​corresponding to each first initial sandstone thickness prediction processing layer to obtain the average value of the first sandstone thickness prediction value corresponding to each first initial sandstone thickness prediction processing layer; obtaining the first dispersion value corresponding to each first initial sandstone thickness prediction model processing layer based on the first sandstone thickness prediction value and the average value of the first sandstone thickness prediction value; and obtaining the first dispersion value corresponding to each first initial sandstone thickness prediction processing layer. Based on all the predicted and actual sandstone thicknesses, the first deviation coefficient value corresponding to each first initial sandstone thickness prediction processing layer is determined. The first evaluation value corresponding to each first initial sandstone thickness prediction processing layer is determined by multiplying the first dispersion value and the first deviation coefficient value. Based on the first evaluation value, each first basic thickness prediction model in all first initial sandstone thickness prediction processing layers is sequentially screened until all first basic thickness prediction models in all first initial sandstone thickness prediction processing layers have been screened. If a first model is found, it is retained; if a second model is found, it is discarded to obtain the first sandstone thickness prediction model. The first model is the first basic thickness prediction model whose first evaluation value after removing the first basic thickness prediction model is less than the first evaluation value after retaining the first basic thickness prediction model. The second model is the first basic thickness prediction model whose first evaluation value after removing the first basic thickness prediction model is greater than or equal to the first evaluation value after retaining the first basic thickness prediction model.

[0044] It can be understood that the first initial sandstone thickness prediction model is the model before updating the first basic thickness prediction model in the first initial sandstone thickness prediction processing layer. The first initial sandstone thickness prediction processing layer is the processing layer in the first initial sandstone thickness prediction model. The first basic thickness prediction model is the basic model in the first initial sandstone thickness prediction processing layer; the basic model can be Transformer, KAN, Mammba, RF, etc. Planar seismic attribute characteristics are the seismic attribute characteristics in a certain plane of the reservoir. The method for obtaining planar seismic attribute characteristics is to extract seismic trace data from adjacent trace numbers around the well point, calculate the similarity between the seismic waveform of the target layer at the well point and the seismic waveform of adjacent traces, calculate the similarity index between different traces using the Dynamic Time Warping (DTW) method, automatically calculate the threshold T using the Otsu method, determine the seismic traces similar to the target layer seismic waveform of the well point seismic trace based on the threshold, calculate the seismic attributes of the seismic traces similar to the target layer seismic waveform of the well point seismic trace, and calculate the average value. The average value is used as the planar seismic attribute characteristic of the seismic trace where the well point is located. The first sandstone thickness prediction value is the sandstone thickness prediction value output by each first basic thickness prediction model. The mean of the first sandstone thickness prediction value is the average of the predicted first sandstone thickness values. The first dispersion value is the dispersion value corresponding to each first initial sandstone thickness prediction treatment layer in the first initial sandstone thickness prediction model after removing each first basic thickness prediction model; variance can be used to express the first dispersion value. The first deviation coefficient value is the deviation coefficient value corresponding to each first initial sandstone thickness prediction treatment layer in the first initial sandstone thickness prediction model after removing each first basic thickness prediction model. The first evaluation value is the product of the first dispersion value and the first deviation coefficient value corresponding to each first initial sandstone thickness prediction treatment layer in the first initial sandstone thickness prediction model after removing each first basic thickness prediction model. The first model is the first basic thickness prediction model where the first evaluation value after removing the first basic thickness prediction model is less than the first evaluation value when retaining the first basic thickness prediction model; the second model is the first basic thickness prediction model where the first evaluation value after removing the first basic thickness prediction model is greater than or equal to the first evaluation value when retaining the first basic thickness prediction model. The actual sandstone thickness is the sandstone thickness at the well point location extracted from single-well stratified data.

[0045] Specifically, the processor sequentially eliminates each first basic thickness prediction model in the first initial sandstone thickness prediction processing layer of the first initial sandstone thickness prediction model. It calculates the first dispersion value and the first deviation coefficient value after eliminating the first basic thickness prediction model, and then multiplies them to obtain the first evaluation value. The processor determines whether to retain a first basic thickness prediction model based on the difference between the first evaluation value before and after elimination. This process continues until the last first basic thickness prediction model completes the screening step. For example, if the first initial sandstone thickness prediction model has 8 first initial sandstone thickness prediction processing layers, and each first initial sandstone thickness prediction processing layer has 8 first basic thickness prediction models, the processor eliminates the first first basic thickness prediction model in the first first initial sandstone thickness prediction processing layer. It calculates the evaluation value of the first first initial sandstone thickness prediction processing layer after elimination. If the evaluation value after elimination is greater than or equal to the evaluation value before elimination, the first first basic thickness prediction model is eliminated; if the evaluation value after elimination is less than the evaluation value before elimination, the first first basic thickness prediction model is retained. Next, following the method described above, the second first basic thickness prediction model in the first initial sandstone thickness prediction processing layer is eliminated... until the eighth first basic thickness prediction model in the eighth initial sandstone thickness prediction processing layer is eliminated. Therefore, the first initial sandstone thickness prediction processing layer may retain 6 first basic thickness prediction models, and the second initial sandstone thickness prediction processing layer may retain 5 first basic thickness prediction models. The first initial sandstone thickness prediction processing layer is averaged, and this average is used as the input for each first basic thickness prediction model in the second initial sandstone thickness prediction processing layer. The average is then applied to each first basic thickness prediction model in the eighth initial sandstone thickness prediction processing layer to obtain the first sandstone thickness prediction result. Thus, the processor determines the first sandstone thickness prediction model in the above manner to predict the first sandstone thickness prediction result.

[0046] In one embodiment, the method further includes: according to the accuracy index corresponding to the first initial sandstone thickness prediction processing layer, sequentially screening multiple first initial sandstone thickness prediction processing layers in the first sandstone thickness prediction model; retaining the first processing layer if the first processing layer is selected; removing the second processing layer and stopping the screening if the second processing layer is selected, so as to obtain the first sandstone thickness prediction model. The first processing layer is the processing layer with an accuracy index greater than or equal to the previous processing layer, and the second processing layer is the processing layer with an accuracy index less than the previous processing layer. The accuracy index is the ratio of the number of first basic thickness prediction models in the first initial sandstone thickness prediction processing layer that meet the preset prediction standard to the total number of all first basic thickness prediction models in the first initial sandstone thickness prediction processing layer. The preset prediction standard is that the difference between the predicted value of the first sandstone thickness and the actual sandstone thickness is less than a preset deviation threshold.

[0047] It can be understood that the first processing layer is the processing layer with an accuracy index greater than or equal to the previous processing layer, and the second processing layer is the processing layer with an accuracy index less than the previous processing layer.

[0048] Specifically, based on the accuracy index corresponding to the first initial sandstone thickness prediction processing layer, multiple first initial sandstone thickness prediction processing layers in the first sandstone thickness prediction model are sequentially screened. For example: the accuracy index of the first first initial sandstone thickness prediction processing layer in the first sandstone thickness prediction model is calculated. When screening the second first initial sandstone thickness prediction processing layer in the first sandstone thickness prediction model, the accuracy index corresponding to the second first initial sandstone thickness prediction processing layer is calculated and compared with the accuracy index of the first first initial sandstone thickness prediction processing layer. If the accuracy index of the second first initial sandstone thickness prediction processing layer is greater than or equal to the accuracy index of the first first initial sandstone thickness prediction processing layer, the second first initial sandstone thickness prediction processing layer is retained. The remaining first initial sandstone thickness prediction processing layers are then screened sequentially. If the accuracy index of the third first initial sandstone thickness prediction processing layer is less than the accuracy index of the first first initial sandstone thickness prediction processing layer, the remaining six first initial sandstone thickness prediction processing layers, including the third first initial sandstone thickness prediction processing layer, are removed. The number of layers in the first sandstone thickness prediction model is set to two. Therefore, the processor can filter the number of layers in the first sandstone thickness prediction model in the above manner, which can improve the prediction accuracy of the first sandstone thickness prediction model.

[0049] In one embodiment, determining the first deviation coefficient value corresponding to each first initial sandstone thickness prediction treatment layer based on all predicted values ​​of first sandstone thickness and actual sandstone thickness for each first initial sandstone thickness prediction treatment layer includes: determining the sum of deviations between all predicted values ​​of first sandstone thickness and actual sandstone thickness for each treatment layer; adding a preset constant to the sum of deviations to obtain a corrected sum of deviations; and taking the reciprocal of the corrected sum of deviations to obtain the first deviation coefficient value corresponding to each treatment layer.

[0050] It is understandable that the preset constant can be 0.1 to correct for deviations and values.

[0051] Specifically, the first deviation coefficient value is calculated so that the first evaluation value can be calculated in the subsequent calculation. Based on the first evaluation value, the first basic thickness prediction models in the first initial sandstone thickness prediction processing layer can be screened to improve the prediction accuracy of the first sandstone thickness prediction result.

[0052] In one embodiment, determining the second sandstone thickness prediction model includes: acquiring a second initial sandstone thickness prediction model, wherein the second initial sandstone thickness prediction model includes multiple second initial sandstone thickness prediction processing layers, each second initial sandstone thickness prediction processing layer includes multiple second basic thickness prediction models, and the input of the current processing layer is the average value of all basic thickness prediction models in the previous processing layer; acquiring the target well seismic data characteristics and actual sandstone thickness at each depth location of the target well; based on the second initial sandstone thickness prediction model, and according to the target well seismic data characteristics, determining the second sandstone thickness prediction value output by each second basic thickness prediction model in each second initial sandstone thickness prediction processing layer; determining the average value of all sandstone thickness prediction values ​​corresponding to each second initial sandstone thickness prediction processing layer, obtaining the average value of the second sandstone thickness prediction value corresponding to each second initial sandstone thickness prediction processing layer; obtaining the second dispersion value corresponding to each second initial sandstone thickness prediction processing layer based on the second sandstone thickness prediction value and the average value of the second sandstone thickness prediction value; and further processing the second initial sandstone thickness prediction model based on the average value of the second initial sandstone thickness prediction processing layer. Based on the predicted and actual sandstone thicknesses of all layers, the second deviation coefficient value corresponding to each second initial sandstone thickness prediction treatment layer is determined. Based on the second dispersion value and the second deviation coefficient value, the second evaluation value corresponding to each second initial sandstone thickness prediction treatment layer is determined. Based on the second evaluation value, each second basic thickness prediction model in all second initial sandstone thickness prediction treatment layers is sequentially screened until all second basic thickness prediction models in all second initial sandstone thickness prediction treatment layers have been screened. If a third model is found, it is retained; if a fourth model is found, it is discarded to obtain the second sandstone thickness prediction model. The third model is the second basic thickness prediction model whose second evaluation value after removing all second basic thickness prediction models is less than the second evaluation value after retaining the second basic thickness prediction model. The fourth model is the second basic thickness prediction model whose second evaluation value after removing all second basic thickness prediction models is greater than or equal to the second evaluation value after retaining the second basic thickness prediction model.

[0053] It can be understood that the second initial sandstone thickness prediction model is the model before updating the second basic thickness prediction model in the second initial sandstone thickness prediction processing layer. The second initial sandstone thickness prediction processing layer is the processing layer in the second initial sandstone thickness prediction model. The second basic thickness prediction model is the basic model in the second initial sandstone thickness prediction processing layer; the basic model can be a deep time-series learning model such as Mamba, KAN, Transformer, and LSTM. The target well seismic data characteristics are the seismic attribute characteristics of the target well in the reservoir. The method for obtaining the target well seismic attribute characteristics is to extract the seismic trace data of adjacent trace numbers around the well point, calculate the similarity between the seismic waveform of the target layer at the well point and the seismic waveform of adjacent traces, calculate the similarity index between different traces using the Dynamic Time Warping (DTW) method, automatically calculate the threshold T using the Otsu method, determine the seismic traces similar to the target layer seismic waveform of the well point seismic traces based on the threshold, directly extract the seismic attributes of the seismic traces similar to the target layer seismic waveform of the well point seismic traces, and use the seismic attributes of similar seismic traces as the target well seismic attribute characteristics. The second sandstone thickness prediction value is the sandstone thickness prediction value output by each second basic thickness prediction model. The mean of the second sandstone thickness prediction value is the average of the predicted second sandstone thickness values. The second dispersion value is the dispersion value corresponding to each second initial sandstone thickness prediction treatment layer in the second initial sandstone thickness prediction model after removing each second basic thickness prediction model; variance can be used to express the first dispersion value. The second deviation coefficient value is the deviation coefficient value corresponding to each second initial sandstone thickness prediction treatment layer in the second initial sandstone thickness prediction model after removing each second basic thickness prediction model. The second evaluation value is the product of the second dispersion value and the second deviation coefficient value corresponding to each second initial sandstone thickness prediction treatment layer in the second initial sandstone thickness prediction model after removing each second basic thickness prediction model. The third model is a second basic thickness prediction model where the second evaluation value after removing the second basic thickness prediction model is less than the second evaluation value when retaining the second basic thickness prediction model. The fourth model is a second basic thickness prediction model where the second evaluation value after removing the second basic thickness prediction model is greater than or equal to the second evaluation value when retaining the second basic thickness prediction model. The actual sandstone thickness is the sandstone thickness at the well point location extracted from the single-well stratified data.

[0054] Specifically, the processor sequentially eliminates each second basic thickness prediction model in the second initial sandstone thickness prediction processing layer of the second initial sandstone thickness prediction model. It calculates the second dispersion value and the second deviation coefficient value after eliminating the second basic thickness prediction model, and then multiplies them to obtain the second evaluation value. The processor determines whether to retain a second basic thickness prediction model based on the difference between the second evaluation value before and after elimination. This process continues until the last second basic thickness prediction model completes the screening step. For example, if there are 8 second initial sandstone thickness prediction processing layers in the second initial sandstone thickness prediction model, and each second initial sandstone thickness prediction processing layer has 8 second basic thickness prediction models, the processor eliminates the first second basic thickness prediction model in the first second initial sandstone thickness prediction processing layer. It calculates the evaluation value of the first second initial sandstone thickness prediction processing layer after elimination. If the evaluation value after elimination is greater than or equal to the evaluation value before elimination, the first second basic thickness prediction model is eliminated; if the evaluation value after elimination is less than the evaluation value before elimination, the first second basic thickness prediction model is retained. Next, following the method described above, the second basic thickness prediction model in the first second initial sandstone thickness prediction processing layer is eliminated... until the eighth basic thickness prediction model in the eighth second initial sandstone thickness prediction processing layer is eliminated. Therefore, the first second initial sandstone thickness prediction processing layer may retain 6 basic thickness prediction models, and the second second initial sandstone thickness prediction processing layer may retain 5 basic thickness prediction models. The first second initial sandstone thickness prediction processing layer is averaged, and this average is used as the input for each basic thickness prediction model in the second second initial sandstone thickness prediction processing layer. The average is then applied to each basic thickness prediction model in the eighth second initial sandstone thickness prediction processing layer to obtain the second sandstone thickness prediction result. Thus, the processor determines the second sandstone thickness prediction model in the above manner to predict the second sandstone thickness prediction result.

[0055] In one embodiment, the method further includes: according to the accuracy index corresponding to the second initial sandstone thickness prediction processing layer, sequentially screening multiple second initial sandstone thickness prediction processing layers in the second sandstone thickness prediction model; retaining the third processing layer if the screening reaches the third processing layer; removing the fourth processing layer and stopping the screening if the screening reaches the fourth processing layer, so as to obtain the first sandstone thickness prediction model. The third processing layer is a second initial sandstone thickness prediction processing layer with an accuracy index greater than or equal to the previous processing layer, and the fourth processing layer is a second initial sandstone thickness prediction processing layer with an accuracy index less than the previous processing layer. The accuracy index is the ratio of the number of second basic thickness prediction models in the second initial sandstone thickness prediction processing layer that meet the preset prediction standard to the total number of all second basic thickness prediction models in the second initial sandstone thickness prediction processing layer. The preset prediction standard is that the difference between the predicted second sandstone thickness and the actual sandstone thickness is less than a preset deviation threshold.

[0056] It can be understood that the second processing layer is the processing layer with an accuracy index greater than or equal to the previous processing layer, and the third processing layer is the processing layer with an accuracy index less than the previous processing layer.

[0057] Specifically, based on the accuracy index corresponding to the second initial sandstone thickness prediction treatment layer, multiple second initial sandstone thickness prediction treatment layers in the second sandstone thickness prediction model are sequentially screened. For example: the accuracy index of the first second initial sandstone thickness prediction treatment layer in the second sandstone thickness prediction model is calculated. When screening the second second initial sandstone thickness prediction treatment layer in the first sandstone thickness prediction model, the accuracy index corresponding to the second second initial sandstone thickness prediction treatment layer is calculated and compared with the accuracy index of the first second initial sandstone thickness prediction treatment layer. If the accuracy index of the second second initial sandstone thickness prediction treatment layer is greater than or equal to the accuracy index of the first second initial sandstone thickness prediction treatment layer, the second second initial sandstone thickness prediction treatment layer is retained. The remaining second initial sandstone thickness prediction treatment layers are then screened sequentially. If the accuracy index of the third second initial sandstone thickness prediction treatment layer is less than the accuracy index of the first second initial sandstone thickness prediction treatment layer, the remaining six first initial sandstone thickness prediction treatment layers, including the third second initial sandstone thickness prediction treatment layer, are removed. The number of layers in the second sandstone thickness prediction model is set to two. Therefore, the processor can filter the number of layers in the second sandstone thickness prediction model in the above manner, which can improve the prediction accuracy of the first sandstone thickness prediction model.

[0058] In one embodiment, the method further includes: obtaining the cumulative porosity, permeability, and lithology sensitivity curve of the target reservoir; and classifying the target reservoir into oil production grades based on a clustering algorithm, according to the cumulative porosity, permeability, lithology sensitivity curve, and predicted sandstone thickness.

[0059] Cumulative porosity can be understood as the ratio of the total pore volume to the total volume of the rock or formation within a given rock or formation volume. Permeability is the ability of a rock to allow fluid to pass through under a given pressure difference. A lithology-sensitive curve is a logging curve in geophysical well logging that exhibits significant differences in response to different lithologies, thus effectively distinguishing different rock types.

[0060] Specifically, the processor, based on a clustering algorithm, can classify the oil production of the target reservoir into high, medium, and low levels by inputting cumulative porosity, permeability, lithology sensitivity curves, and predicted sandstone thickness.

[0061] The specific steps can be as follows:

[0062] Step 1: Intelligent prediction of sandstone thickness based on the first sandstone thickness prediction model.

[0063] (1) Establishment of a well-seismic calibration database based on sliding window and attribute overlay technology.

[0064] A sliding multi-scale window is used to determine the extended layer positions. The target layer's seismic interpretation top and bottom layers are shifted upwards and downwards by X ms to determine the new top and bottom layers, where X ∈ {1,2,…,NX} represents the shift step size. Simultaneously, a new top and bottom layer position is determined by simultaneously expanding from the top and bottom of the layer towards the top and bottom by Y ms, where Y ∈ {1,2,…,NY}. Specifically, the range of NX (3ms-6ms) and NY (3ms-6ms) can be determined based on the actual geological conditions.

[0065] Seismic attribute extraction is based on waveform-constrained seismic attribute overlay. When extracting seismic attributes at each well point, instead of directly extracting the seismic attributes, the seismic trace data of adjacent trace numbers around that point are extracted. The similarity between the target layer seismic waveform at point i and the waveform of the adjacent trace j is calculated. The similarity index Sij between different traces is calculated using the Dynamic Time Warping (DTW) method. A threshold T is automatically calculated using the Otsu method. Based on the threshold, seismic traces with similar target layer seismic waveforms to the trace at that point are identified, and their seismic attributes are calculated for each. The average value is used as the seismic attribute of that trace. If Sij > T, then trace j can be used for the seismic attribute calculation at point i. Commonly used seismic attributes include maximum amplitude, average instantaneous phase, root mean square amplitude, coherence, and curvature. Appropriate seismic attributes can be selected based on prior experience in different regions, or optimal seismic attributes can be selected using methods such as correlation coefficient heatmaps, contribution analysis, and SHAP.

[0066] Well-seismic matching identifies a labeled sample set. Sandstone thickness at each well location is extracted from the single-well stratified data. Four seismic traces surround each well location; based on the principle of most similarity, the sandstone thickness at the single well is assigned as the attribute of the trace most similar to the target point. The data is then divided into k equal parts for subsequent training of an automated machine learning model.

[0067] (2) Sandstone thickness prediction based on the first sandstone thickness prediction model.

[0068] Data cleaning and normalization. Invalid values ​​are removed, and the data is normalized using a max-min method.

[0069] The selection of base models at each layer in automated machine learning involves several steps. The automated machine learning model is a multi-layered cascaded machine learning combination, with each layer composed of multiple base machine learning models. Each layer predicts sandstone thickness based on the outputs of different classifiers, and the average value is taken as the prediction result for that layer. The output of each model in that layer serves as the input for the next layer. The first dispersion value and the first deviation coefficient value are calculated. From the base machine learning model set M = {Transformer, KAN, Mammba, RF, ...}, one classifier is removed sequentially. If the first dispersion value and the first deviation coefficient value of the final output of the automated machine learning model increase, the classifier is removed; otherwise, it is retained.

[0070] The number of layers in the first sandstone thickness prediction model is automatically determined. The accuracy index Pi for the i-th layer is calculated by dividing the number of samples meeting the prediction criteria by the total number of samples. Generally, the difference between the samples meeting the prediction criteria and the actual sandstone thickness is within ±20%. In other words, the accuracy index is the number of wellpoints with sandstone thickness meeting the prediction criteria divided by the total number of samples. If the accuracy index Pi+1 is lower than that of the (i+1)-th layer, then the automated machine learning model should use layer i, where i = 1, 2, ..., N.

[0071] The established first sandstone thickness prediction model is used to predict sandstone thickness between wells. The seismic attributes extracted from each point between wells are input into the established automated machine learning model, which outputs the corresponding predicted sandstone thickness, thereby obtaining a sandstone thickness planar map.

[0072] Step 2: A second sandstone thickness prediction model based on integrated deep learning.

[0073] (1) Construction of input attribute three-dimensional data volume based on frequency division and similarity superposition enhancement technology.

[0074] Three-dimensional data volumes of different frequencies, especially seismic attribute volumes of 10-90Hz, are extracted using fast Fourier transform or wavelet transform. Attribute volumes such as root mean square amplitude are extracted from different seismic data volumes, and these attribute volumes will be used as inputs for three-dimensional seismic inversion.

[0075] (2) Improved resolution well-seismic matching and establishment of labeled sample library based on seismic data.

[0076] Multiple seismic traces from wells with similar lithological interpretations were extracted. Spline interpolation was performed on the seismic traces to match the logging resolution. The recalculated seismic traces and logging lithological interpretation curves were compared. The similarity index between the logging curves and surrounding seismic traces was calculated using the dynamic time warping method. The threshold T was automatically calculated using the Otsu method. Based on the threshold, seismic traces similar to the single well were identified. These seismic traces were superimposed to enhance the lithological response characteristics. A sample library was established based on the seismic data and label information at the corresponding depths for subsequent lithological prediction model development.

[0077] (3) Establishment of a second sandstone thickness prediction model integrating deep learning.

[0078] Using deep temporal learning models such as Mamba, KAN, Transformer, and LSTM as base learners, a second discreteness value and a second deviation coefficient value are established. From the set of base machine learning models M = {Transformer, KAN, Mamba, LSTM, ...}, one classifier is sequentially removed. If the product of the second discreteness value and the second deviation coefficient value of the final output of the deep learning model is improved, then that base classifier is removed; otherwise, it is retained.

[0079] (4) Lithology prediction and sandstone thickness calculation based on the second sandstone thickness prediction model based on integrated deep learning.

[0080] By inputting multiple three-dimensional data volumes from step (1) into the prediction model in step (3), the corresponding three-dimensional lithology prediction results can be obtained. Based on the three-dimensional seismic horizons, the corresponding three-dimensional lithology prediction results are extracted, and the cumulative lithology thickness of each trace is calculated to obtain the sandstone thickness of the corresponding target layer.

[0081] Step 3: Simulation of sandstone thickness in well seismic wells under the control phase mode.

[0082] (1) Channel facies classification based on well point sandstone thickness.

[0083] Based on the distribution of sandstone thickness data at well points, the region corresponding to the upper 5% quantile is defined as channel facies (assigned a value of 1), and the region corresponding to the lower 5% quantile is defined as non-channel facies (assigned a value of 0), which serves as the well point facies data.

[0084] (2) Construction of training images and trend constraint surface data of two-phase river channels.

[0085] A 64×64 grid scanning template was used to divide the training images of two-phase channels in the study area into datasets. Based on the proportion of channel phases in the grid cells, the channel phase pattern library was divided into four categories: high channel proportion, medium channel proportion, low channel proportion, and no channel proportion. In particular, the images in the medium channel proportion category were further subdivided into two types: "with bifurcation" and "without bifurcation".

[0086] By performing convolution operations on training images from the channel facies model library, and applying Gaussian and bilateral filtering functions to weighted smooth the data, two trend surfaces corresponding to the channel facies training data were constructed. For the target layer, the sandstone thickness prediction results and lithology 3D inversion results were converted into trend constraint surface data with a 64×64 grid size using pixel sampling technology.

[0087] (3) Intelligent simulation of river facies based on H-DDPM method.

[0088] Channel facies simulation was conducted using the Hybrid-DDPM (H-DDPM) method. Hybrid-DDPM is a comprehensive facies model generation algorithm that integrates residual neural networks, Transformer models, and DDPM networks. A pre-trained H-DDPM model was trained based on known two-phase channel training images and trend constraint surface data. 80% of the wellpoint facies data from step 3 (1) and the trend constraint surface data of the target layer from step 3 (2) were input into the pre-trained model to generate a channel facies distribution map. A channel model index was established. ,in, For H-DDPM in Predicted phase value of well point This represents the predicted average phase value at well points. This represents 20% of the well point facies data. Simulation goodness index. ,in express The error between the predicted and actual phase values ​​of well points is calculated, with the input well point phase data used as the actual phase value. This is achieved through comparison. Sum of error lower bound ,if Iteratively optimize the hyperparameters of the pre-trained model; otherwise, input all wellpoint facies data into the pre-trained model to obtain the optimal channel facies distribution map.

[0089] Step 4: Comprehensive prediction of sandstone thickness.

[0090] Using a collaborative geostatistical simulation algorithm, with 80% of the wellpoint sand thickness Z as hard data, and the predicted thicknesses of the first and second sandstone layers, along with the channel facies distribution map, as trend data, the predicted sandstone thickness is obtained. The reserved 20% of hard data is used for comparison with channel model indices. and simulation goodness index The product and the lower bound of error ,if The hyperparameters of the collaborative geostatistical simulation algorithm are iteratively optimized; otherwise, the optimal comprehensive sandstone thickness for the corresponding target layer is obtained.

[0091] Step 5: Predicting reservoir sweet spots.

[0092] Figure 2 This illustration schematically shows a diagram of reservoir sand body thickness prediction in one embodiment of this application, as shown below. Figure 2 As shown, the reservoir sand body thickness prediction results in one embodiment of this application can be divided into 0.5m-4.5m. Figure 2 The magnitude of the reservoir sand body thickness prediction results is mapped using gradient chromatography, with red representing high value areas (such as high sandstone thickness) and blue representing low value areas (such as low sandstone thickness).

[0093] Figure 3 This illustration schematically shows a diagram of reservoir oil production (sweet spot) prediction in one embodiment of this application, as shown below. Figure 3 As shown, the reservoir oil production (sweet spot) prediction results in one embodiment of this application can classify sweet spots into three categories: Category I, Category II, and Category III. Figure 3 In this study, category chromatography is used to distinguish reservoir oil production (sweet spots). Red (Class I sweet spot) is the optimal region, green (Class II sweet spot) is the suboptimal region, and blue (Class III sweet spot) is the low potential region.

[0094] A three-dimensional physical property model of porosity, permeability, and lithological sensitivity curves for the target layer is established. The cumulative physical property parameters for each pass are calculated, thus obtaining a planar distribution map of the cumulative porosity, permeability, and lithological sensitivity curves for the corresponding target layer. This is combined with the overall sandstone thickness (i.e.,...) Figure 2 ), using K-Means clustering algorithm or establishing cross-plot evaluation criteria for sandstone thickness and yield (i.e. Figure 3 These characteristics are analyzed comprehensively to classify the target reservoir into three categories: sweet spots (Class I, Class II, and Class III).

[0095] A second aspect of this application provides an apparatus for intelligent prediction of reservoir sandstone thickness. The apparatus includes: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the method for intelligent prediction of reservoir sandstone thickness as described above.

[0096] A third aspect of this application provides a machine-readable storage medium, characterized in that the machine-readable storage medium stores instructions for causing a machine to execute the above-described method for intelligent prediction of reservoir sandstone thickness.

[0097] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0098] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligent prediction of reservoir sandstone thickness, characterized in that, The method comprises: acquiring actual sandstone thicknesses of respective depth positions of a target well in a target reservoir, a three-dimensional seismic data body of a to-be-predicted point in the reservoir, seismic attribute features of the to-be-predicted point, and a well point facies type corresponding to the target well, wherein the to-be-predicted point is a point in the target reservoir other than the target well; obtaining a first sandstone thickness prediction result of the to-be-predicted point according to the seismic attribute features based on a predetermined first sandstone thickness prediction model; obtaining a second sandstone thickness prediction result of the to-be-predicted point according to the three-dimensional seismic data body based on a predetermined second sandstone thickness prediction model; respectively performing trend constraint face conversion on the first sandstone thickness prediction result and the second sandstone thickness prediction result to obtain first trend constraint face data corresponding to the first sandstone thickness prediction result and second trend constraint face data corresponding to the second sandstone thickness prediction result; obtaining a river facies distribution map corresponding to the to-be-predicted point according to the well point facies type, the first trend constraint face data, and the second trend constraint face data; obtaining a predicted sandstone thickness of the to-be-predicted point according to the first sandstone thickness prediction result, the second sandstone thickness prediction result, the river facies distribution map, and the actual sandstone thickness.

2. The method of claim 1, wherein, The obtaining of the predicted sandstone thickness of the to-be-predicted point according to the first sandstone thickness prediction result, the second sandstone thickness prediction result, the river facies distribution map, and the actual sandstone thickness comprises: determining a product value of each actual sandstone thickness of a depth position of the to-be-predicted point and a preset sandstone thickness weight, wherein the to-be-predicted point is located at the depth position and has a distance less than a preset distance threshold from the to-be-predicted point, and summing the product values to obtain an actual sandstone thickness influence value; determining a product value of the first sandstone thickness prediction result of each to-be-predicted point and a preset first sandstone thickness weight, wherein the to-be-predicted point has a distance less than the preset distance threshold from the to-be-predicted point, and summing the product values to obtain a first sandstone thickness prediction result influence value; determining a product value of the second sandstone thickness prediction result of each to-be-predicted point and a preset second sandstone thickness weight, wherein the to-be-predicted point has a distance less than the preset distance threshold from the to-be-predicted point, and summing the product values to obtain a second sandstone thickness prediction result influence value; determining a product value of the river facies in the river facies distribution map and a preset river facies weight, wherein the river facies has a distance less than the preset distance threshold from the to-be-predicted point, and summing the product values to obtain a river facies influence value; adding the actual sandstone thickness influence value, the first sandstone thickness prediction result influence value, the second sandstone thickness prediction result influence value, and the river facies influence value to obtain the predicted sandstone thickness.

3. The method of claim 1, wherein, The determination of the first sandstone thickness prediction model comprises: obtaining a first initial sandstone thickness prediction model, wherein the first initial sandstone thickness prediction model comprises a plurality of first initial sandstone thickness prediction processing layers, each of the first initial sandstone thickness prediction processing layers comprises a plurality of first basic thickness prediction models, and an input of a current processing layer is an average value of all the basic thickness prediction models in a previous processing layer; obtaining a plane seismic attribute feature and an actual sandstone thickness of each depth position of the target well; determining, based on the first initial sandstone thickness prediction model, a first sandstone thickness prediction value of each depth position of the target well output by each of the first basic thickness prediction models in each of the first initial sandstone thickness prediction processing layers according to the plane seismic attribute feature; determining a mean value of all the first sandstone thickness prediction values corresponding to each of the first initial sandstone thickness prediction processing layers to obtain a first sandstone thickness prediction value mean value corresponding to each of the first initial sandstone thickness prediction processing layers; determining, according to the first sandstone thickness prediction value and the first sandstone thickness prediction value mean value, a first dispersion degree value corresponding to each of the first initial sandstone thickness prediction processing layers; determining, according to all the first sandstone thickness prediction values corresponding to each of the first initial sandstone thickness prediction processing layers and the actual sandstone thickness, a first deviation coefficient value corresponding to each of the first initial sandstone thickness prediction processing layers; determining, according to a product of the first dispersion degree value and the first deviation coefficient value corresponding to each of the first initial sandstone thickness prediction processing layers, a first evaluation value corresponding to each of the first initial sandstone thickness prediction processing layers; performing, according to the first evaluation value, screening processing on each of the first basic thickness prediction models in all the first initial sandstone thickness prediction processing layers in sequence until screening of all the first basic thickness prediction models in all the first initial sandstone thickness prediction processing layers is completed, retaining a first model in a case of screening the first model and removing a second model in a case of screening the second model, to obtain the first sandstone thickness prediction model, wherein the first model is a first basic thickness prediction model whose first evaluation value in a case of removing the first basic thickness prediction model is less than a first evaluation value in a case of retaining the first basic thickness prediction model, and the second model is the first basic thickness prediction model whose first evaluation value in a case of removing the first basic thickness prediction model is greater than or equal to a first evaluation value in a case of retaining the first basic thickness prediction model.

4. The method of claim 3, wherein, The method further comprises: According to the accuracy rate index corresponding to the first initial sandstone thickness prediction processing layer, the plurality of first initial sandstone thickness prediction processing layers in the first sandstone thickness prediction model are sequentially screened, the first processing layer is retained in the case of screening the first processing layer, the second processing layer is removed and the screening is stopped in the case of screening the second processing layer, and the first sandstone thickness prediction model is obtained, wherein the first processing layer is a processing layer with an accuracy rate index greater than or equal to the last processing layer, the second processing layer has an accuracy rate index less than the last processing layer, the accuracy rate index is the ratio of the number of first basic thickness prediction models reaching a preset prediction standard in the first initial sandstone thickness prediction processing layer to the total number of all first basic thickness prediction models in the first initial sandstone thickness prediction processing layer, and the preset prediction standard is that the difference between the first sandstone thickness prediction value and the actual sandstone thickness is less than a preset deviation threshold.

5. The method of claim 3, wherein, The first deviation coefficient value corresponding to each first initial sandstone thickness prediction processing layer is determined according to all the first sandstone thickness prediction values and the actual sandstone thickness corresponding to each first initial sandstone thickness prediction processing layer, including: determining the deviation sum value of the deviation of all the first sandstone thickness prediction values and the actual sandstone thickness corresponding to each processing layer; a preset constant is added on the basis of the deviation sum value to obtain a corrected deviation sum value; the reciprocal of the corrected deviation sum value is taken to obtain the first deviation coefficient value corresponding to each processing layer.

6. The method of claim 1, wherein, The determination of the second sandstone thickness prediction model includes: obtaining a second initial sandstone thickness prediction model, wherein the second initial sandstone thickness prediction model includes a plurality of second initial sandstone thickness prediction processing layers, each second initial sandstone thickness prediction processing layer includes a plurality of second basic thickness prediction models, and the input of the current processing layer is the average value of all the basic thickness prediction models in the last processing layer; obtaining the target well seismic data characteristics and the actual sandstone thickness of each depth position of the target well; determining the second sandstone thickness prediction value output by each second basic thickness prediction model in each second initial sandstone thickness prediction processing layer according to the target well seismic data characteristics based on the second initial sandstone thickness prediction model; determining the mean value of all the sandstone thickness prediction values corresponding to each second initial sandstone thickness prediction processing layer to obtain the second sandstone thickness prediction value mean value corresponding to each second initial sandstone thickness prediction processing layer; according to the second sandstone thickness prediction value and the second sandstone thickness prediction value mean value, the second dispersion degree value corresponding to each second initial sandstone thickness prediction processing layer is obtained; determining the second deviation coefficient value corresponding to each second initial sandstone thickness prediction processing layer according to all the sandstone thickness prediction values and the actual sandstone thickness corresponding to each second initial sandstone thickness prediction processing layer; determining the second evaluation value corresponding to each second initial sandstone thickness prediction processing layer according to the second dispersion degree value and the second deviation coefficient value; According to the second evaluation value, each second basic thickness prediction model in all the second initial sandstone thickness prediction processing layers is sequentially screened until all the second basic thickness prediction models in all the second initial sandstone thickness prediction processing layers are screened, a third model is retained in the case of screening the third model, and a fourth model is removed in the case of screening the fourth model, to obtain the second sandstone thickness prediction model, wherein the third model is a second basic thickness prediction model whose second evaluation value in the case of removing each second basic thickness prediction model is less than the second evaluation value in the case of retaining the second basic thickness prediction model, and the fourth model is a second basic thickness prediction model whose second evaluation value in the case of removing each second basic thickness prediction model is greater than or equal to the second evaluation value in the case of retaining the second basic thickness prediction model.

7. The method of claim 6, wherein, The method further comprises: According to the accuracy index corresponding to the second initial sandstone thickness prediction processing layer, a plurality of second initial sandstone thickness prediction processing layers in the second sandstone thickness prediction model are sequentially screened, the third processing layer is retained in the case of screening the third processing layer, and the fourth processing layer is removed and the screening is stopped in the case of screening the fourth processing layer, to obtain the first sandstone thickness prediction model, wherein the third processing layer is a second initial sandstone thickness prediction processing layer whose accuracy index is greater than or equal to that of the last processing layer, the fourth processing layer is a second initial sandstone thickness prediction processing layer whose accuracy index is less than that of the last processing layer, the accuracy index is a ratio of a number of second basic thickness prediction models in the second initial sandstone thickness prediction processing layer that meet a preset prediction standard to a total number of all second basic thickness prediction models in the second initial sandstone thickness prediction processing layer, and the preset prediction standard is that a difference between the second sandstone thickness prediction value and the actual sandstone thickness is less than a preset deviation threshold.

8. The method of claim 1, wherein, The method further comprises: Cumulative porosity, permeability, and lithology sensitivity curves of the target reservoir are obtained; Based on a clustering algorithm, the target reservoir is divided into oil production grades according to the cumulative porosity, the permeability, the lithology sensitivity curves, and the predicted sandstone thickness.

9. An apparatus for intelligent prediction of reservoir sandstone thickness, characterized by, Comprise: a memory configured to store instructions; and a processor configured to call the instructions from the memory and enable the method for intelligent prediction of reservoir sandstone thickness according to any one of claims 1 to 8 when the instructions are executed.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing a machine to perform the method for intelligent prediction of reservoir sandstone thickness according to any one of claims 1 to 7. The machine-readable storage medium stores instructions for causing a machine to perform the method for intelligent prediction of reservoir sandstone thickness according to any one of claims 1 to 7.

Images