A method for determining the original and current water saturation of a flooded layer.

Abstract

This invention relates to a method for determining the original and current water saturation of a water-flooded layer. It primarily addresses the challenge of existing well logging interpretation methods, which rely solely on well logging amplitude information and conventional data analysis techniques, failing to further improve the accuracy of well logging interpretation. The method includes: well logging data format conversion, curve extraction and curve redirection; reconstruction of the natural gamma curve; construction of a mudstone baseline from the spontaneous potential curve; multi-dimensional feature extraction from the merged CSV well logging curves; selection of a suitable algorithm; training using prepared sample data; and outputting models for the original and current water saturation. The extracted feature data from the well to be predicted is input into the model to predict the original and current water saturation at each depth point within the predicted thickness of the well. This method can further improve the accuracy of well logging interpretation, providing technical support for detailed reservoir geological description and precise tapping of remaining oil potential.

Description

Technical fields:

[0001] This invention relates to the field of oilfield logging evaluation technology, and in particular to a method for determining the original water saturation and current water saturation of a water-flooded layer. Background technology:

[0002] Quantitative calculation of water-flooded reservoir parameters mainly involves determining the production parameters, with residual oil saturation as the core. Among these, the original water saturation and current water saturation are crucial parameters for quantitatively evaluating water-flooded reservoirs. Extensive research has been conducted by scholars both domestically and internationally, resulting in numerous mature theoretical and empirical calculation formulas based on rock volumetric physics models. These parameters reflect reservoir characteristics from different perspectives; they are related yet distinct, all closely related to reservoir pore volume and controlled by pore geometry, morphology, and distribution.

[0003] Previously, the determination of the original water saturation mainly adopted the "core calibration logging" technology. Mathematical statistics methods were applied, guided by the logging response characteristics of the reservoir and the conductivity mechanism of the water-flooded layer, and constrained by geological conditions. Interpretation models of the original water saturation parameters of thick oil layers were established according to different oil layer groups and different thickness types. This enabled a detailed quantitative evaluation of the water flooding degree of thick oil layers, meeting the needs of oilfield development and potential tapping.

[0004] Water saturation remains a challenging issue in well logging interpretation. For many years, the domestic and international well logging academic community has conducted in-depth research on saturation models for argillaceous sandstone reservoirs. Each model reflects different characteristics of argillaceous sandstone reservoirs from a certain perspective, or develops different understandings of the conductivity of argillaceous sandstone to some extent. The resistivity interpretation models used to determine saturation are mostly empirical models established using parallel conductivity theory and Archie's formula or extended Archie formula. These include the Hossin equation, Simandoux equation, Indonesia equation, WS model, DW model, SB model, HB model, and SATORI model.

[0005] However, due to the complexity of formation water mixtures under various displacement conditions after water flooding of oil reservoirs, the electrical response patterns are significantly affected. As the degree of water flooding increases, the electrical response sensitivity of high and ultra-high water flooded layers and inefficient circulating layers decreases. Conventional logging information interpretation methods and data analysis techniques alone are insufficient to further improve the interpretation accuracy. Furthermore, conventional interpretation methods primarily use one-dimensional or two-dimensional linear models, failing to fully exploit the multi-dimensional potential information within and between curves. These methods also lack scalability, have rigid patterns, poor real-time performance, and long update cycles. New interpretation methods are needed to identify remaining oil and support precise potential tapping.

[0006] With the continuous development of computer technology, the research and application of information technology and automation technology have gradually gained importance in oilfield scientific research and production. Decades of development and production in Daqing Changyuan have accumulated rich and standardized geological, testing, and dynamic data, providing a solid material foundation for the application of big data analysis technology. How to leverage big data analysis technology to discover the hidden value of data, improve the utilization rate of various data in the oilfield production process, and achieve data-driven oil discovery, digital identification, prediction, and optimization will inevitably become an important way to guide the intelligent development goals of oilfield production and reduce costs and increase efficiency. Summary of the Invention:

[0007] This invention overcomes the problem in existing well logging interpretation methods that rely solely on well logging amplitude information and conventional data analysis techniques, which are insufficient to further improve the accuracy of well logging interpretation. Instead, it provides a method for determining the original and current water saturation of water-flooded layers. This method can guide the automatic interpretation of single-well vertical to layer-by-layer and multi-well horizontal comparisons, reducing labor intensity for interpreting water-flooded layers in new wells and reviewing old wells, and providing technical support for detailed reservoir geological description and precise tapping of remaining oil potential.

[0008] The present invention solves its problem through the following technical solution: a method for determining the original water saturation and current water saturation of a flooded layer, comprising:

[0009] Part 1: Model Training and Learning

[0010] (1) Extract the curves required for modeling, convert the format of the sample well logging data, and redirect the curves;

[0011] (2) Preprocess the converted sample well logging data;

[0012] (3) Reconstruct the natural gamma curves in the sample well logging data; construct the mudstone baseline from the spontaneous potential curves in the sample well logging data;

[0013] (4) Core data preprocessing; depth matching of preprocessed core data with logging curves;

[0014] (5) The processed logging curve data is merged with the thickness data, oil layer group information and core analysis data that are matched with the logging curve depth stored in dbf or Excel format, and merged into a CSV sample well logging curve file.

[0015] (6) Perform multidimensional feature extraction on the merged CSV logging curves;

[0016] (7) Select a suitable AI algorithm, train it using the prepared sample data, and output the original water saturation and the current water saturation model;

[0017] Part Two: Model Prediction

[0018] (1) Extract the curves required for modeling, convert the format of the logging data of the well to be predicted, and redirect the curves;

[0019] (2) Preprocess the converted logging data of the well to be predicted;

[0020] (3) Reconstruct the natural gamma curve in the logging data of the well to be predicted, and construct the mudstone baseline in the spontaneous potential curve in the logging data of the well to be predicted;

[0021] (4) The processed logging curve data is merged with the thickness data and oil layer group information stored in dbf or Excel format and merged into a csv logging curve file of the well to be predicted;

[0022] (5) Extract multidimensional features from the logging curves of the wells to be predicted in the merged CSV file;

[0023] (6) Input the feature data extracted from the well to be predicted into the trained model, and predict the original water saturation and the current water saturation for each depth point within the thickness to be predicted in the well to obtain preliminary prediction results.

[0024] Preferred methods for format conversion of sample well logging data and well logging data to be predicted include:

[0025] Extract the corresponding curves from the LAS format file (or TXT format, 716 format, etc.) of each well and convert them into CSV files. Then, convert any character data in the files into numerical data.

[0026] Preferably, the method for preprocessing the converted sample well logging data and the well logging data to be predicted includes:

[0027] The method for handling missing values ​​is to use a linear interpolation algorithm to fill in the missing values ​​in the sampling points based on the adjacent sampling points;

[0028] Truncate outliers;

[0029] Invalid values ​​are deleted.

[0030] Preferably, the method for constructing a mudstone baseline from the spontaneous potential curves in the sample well logging data and the well logging data to be predicted includes:

[0031] The baseline walking method based on the spontaneous potential extrema within a window range constructs a mudstone baseline through heuristic search: First, a window length is set, and spontaneous potential extrema are searched within the first window length. After finding the first maximum point, the window length is extended downwards as a boundary to search for extrema again, and the two searched extrema are connected. This process is repeated to construct the mudstone baseline.

[0032] Preferred core data preprocessing methods include:

[0033] Core data standardization: Except for permeability, all other parameters of the core data are uniformly converted into percentage units;

[0034] Remove missing values: Only remove null values ​​corresponding to the current prediction parameters;

[0035] Remove abnormal core analysis sample data: such as removing data whose original water saturation value is greater than the current water saturation value.

[0036] Preferably, multidimensional feature extraction is performed on the well logging curves of the merged CSV sample wells and the well logging curves of the well to be predicted, including: extracting the numerical features of the original curves, transforming the original curves, normalizing the curves, and extracting the maximum, minimum, average, median, variance, curve difference feature values, curve skewness, and kurtosis of each original well logging curve within a certain step size window.

[0037] Preferably, the AI ​​algorithm includes traditional learning algorithms and deep learning algorithms;

[0038] The traditional learning algorithms include: Linear algorithm, LightGBM (gradient boosting machine algorithm), Xgboost (gradient boosting algorithm), Random Forest, AdaBooost (iterative algorithm), etc.

[0039] The deep learning algorithms mentioned include: fully connected neural networks (DNN), long short-term memory (LSTM) neural networks, and combinations of convolutional neural networks and LSTM (CNN+LSTM), etc. Among them, CNN (Convolutional Neural Network) is a type of convolutional neural network.

[0040] Preferably, a method for post-processing the obtained preliminary prediction results to obtain accurate data on the original water saturation and the current water saturation includes:

[0041] Specify the data rationality of the above prediction results. The specification principle is as follows: If the predicted SWI ≥ 100, then SWI is re-assigned 100; if the predicted SWI ≤ 10, then SWI is re-assigned 10; if the predicted SW ≥ 100, then SW is re-assigned 100; if the predicted SW ≤ 0, then SW is re-assigned 0; if the predicted SW < SWI, then the value of SWI is assigned to SW, that is, SWI = SW;

[0042] Where: SW: current water saturation; SWI: original water saturation.

[0043] The present invention may have the following beneficial effects when compared with the above background technology:

[0044] Compared with other methods for determining the original water saturation and current water saturation of water-flooded layers, the significant advantages of the present invention are reflected in:

[0045] (1) The method for determining the original water saturation and current water saturation of water-flooded reservoirs based on big data analysis technology discovers the hidden value of data through multi-dimensional feature mining technologies such as multi-dimensional data correlation and cognitive computing, breaks through the limitations of traditional one-dimensional and two-dimensional linear models, and improves the utilization rate of various types of information;

[0046] (2) The interpretive method has stronger expandability, can be re-trained for the data of new blocks, the sample library and the model can be updated in real time, is autonomous and controllable, and is more adaptable to the constantly changing complex geological conditions;

[0047] (3) The method for determining the original water saturation and current water saturation of water-flooded reservoirs based on big data analysis technology has higher accuracy and less manual intervention;

[0048] (4) The new method realizes the intelligent identification of the original water saturation and current water saturation of water-flooded reservoirs, does not require parameter cards to be set, can process multiple wells in batches, saves a large amount of labor costs, thus greatly improving work efficiency and reducing manual labor intensity. It provides technical support for fine description of reservoir geology, precise tapping of remaining oil, etc. Description of the Drawings:

[0049] The drawings here are incorporated into the specification and form a part of this specification. These drawings show embodiments in line with the present invention and are used together with the specification to explain the technical solutions of the present invention.

[0050] Through the description of the embodiments of the present invention with reference to the following drawings, the above and other objects, features and advantages of the present invention become clearer. In the drawings:

[0051] Figure 1 It is the water-flooded interpretation result diagram of Well Bei 2-362-Jian P25 in Embodiment 1 of the present invention;

[0052] Figure 2 This is a diagram illustrating the results of water flooding interpretation of Well North 2-350-Jian 45 in Embodiment 2 of the present invention;

[0053] Figure 3 This is a diagram illustrating the results of water flooding interpretation of Well North 2-357-Jian 82 in Embodiment 3 of the present invention;

[0054] Figure 4 This is a schematic block diagram illustrating the method for determining the original water saturation and current water saturation of the flooded layer according to an embodiment of the present invention. Detailed implementation method:

[0055] The present invention will now be described in detail with reference to the accompanying drawings and exemplary embodiments. However, it should be understood that, without further description, the layering descriptions, arrangements, and features of one embodiment may be advantageously incorporated into other embodiments.

[0056] Figure 4 A schematic block diagram illustrating a method for determining the initial and current water saturation of a flooded layer according to an embodiment of the present invention is shown. Figure 4 As shown, a method for determining the original and current water saturation of a flooded layer is described, and its specific implementation plan consists of two parts:

[0057] Part 1: Model Training and Learning

[0058] (1) Extract the curves required for modeling, perform format conversion on the sample well logging data, and redirect the curves; the methods for format conversion of sample well logging data and logging data of the well to be predicted include:

[0059] Extract the corresponding curves from the LAS format file (or TXT format, 716 format, etc.) of each well and convert them into CSV files. Then, convert any character data in the files into numerical data.

[0060] (2) Preprocessing the converted sample well logging data; methods for preprocessing the converted sample well logging data and the logging data of the well to be predicted include:

[0061] The missing values ​​are handled by using a linear interpolation algorithm to fill in the missing values ​​in the sampling points with adjacent sampling points; outliers are truncated; and invalid values ​​are deleted.

[0062] (3) Reconstruct the natural gamma curves in the sample well logging data; construct the mudstone baseline from the spontaneous potential curves in the sample well logging data; including:

[0063] To address the anomaly issue of GR curves in water-flooded layers, a GR reconstruction model is constructed. Unaffected curves are input into the GR reconstruction model to generate reconstructed values ​​to replace outliers. To address the significant differences in formation water salinity across different strata and the shift in spontaneous potential (SP) mudstone baselines after water flooding of oil-bearing layers, requiring improved accuracy in SP amplitude difference calculations, a heuristic search method based on SP extreme points within a window is used to construct mudstone baselines. The method is as follows: First, based on a pre-defined window length, extreme points of spontaneous potential are searched within the first window length. After finding the first maximum point, the window length is extended downwards as a boundary to search for extreme points again, and these two searched extreme points are connected. This process is repeated to construct the mudstone baseline.

[0064] (4) Core data preprocessing; depth matching of preprocessed core data with logging curves; core data preprocessing methods include:

[0065] This includes core data standardization: converting all parameters in the core data except permeability to percentage units. Furthermore, to ensure the reliability of the modeling sample data, it is necessary to clean up abnormal core analysis sample data, such as removing missing values: only removing null values ​​corresponding to the current prediction parameters; and removing abnormal core analysis sample data: since the original water saturation samples in core analysis are calculated based on regional experience, some original water saturation may be higher than the current water saturation, and these original water saturation values ​​need to be removed.

[0066] (5) Merge the processed logging curve data with the thickness data, reservoir information, and core analysis data (stored in DBF or Excel format) and matched with the logging curve depth. Since the reservoir and thickness data are given according to depth segments, the reservoir information needs to be discretized by depth point, aligned with the logging curve at the same depth, and then merged. Furthermore, perform depth matching between the core analysis data and the logging curve data. The method is as follows: for each core sampling point, find its nearest logging curve sampling point within a certain threshold range and perform depth matching. Finally, integrate all the above data into a single CSV sample well logging curve file.

[0067] (6) Perform multidimensional feature extraction on the merged CSV well logging curves; perform multidimensional feature extraction on the merged CSV sample well logging curves, including:

[0068] Based on well logging curves, this study enhances the dimensionality of well logging data by extracting and fusing multi-dimensional features. Deep reinforcement learning is used to represent these multi-information features, bridging the mapping from the data domain to the feature domain. This overcomes the limitations of primarily one-dimensional or two-dimensional linear models in interpretation methods, improving the utilization rate of various information types. The well logging curve features to be extracted include the following:

[0069] (a) Original curve amplitude characteristics: micropotential RMN, microgradient RMG, deep lateral RLLD, shallow lateral RLLS, microsphere RXO, lithological density DEN, acoustic wave HAC, spontaneous potential SP, natural gamma GR;

[0070] (b) Curve Transformations: RMN-RMG, LOG(RLLD), LOG(RLLS), DSP, HAC-DEN

[0071] (c) Normalize the original curve and use the min-max method to map the sample feature values ​​to the range (0, 1).

[0072] (d) Using the current depth point as the center, simultaneously take the average, maximum, minimum, median, and variance of the curve within a step window of 18, 16, 14, 12, and 10 sampling points upwards and downwards (taking the RMN curve as an example, taking RMN_window_18_avg, RMN_window_18_max, RMN_window_18_min......RMN_window_10_avg, RMN_window_10_max, RMN_window_10_min).

[0073] (e) Differential Feature Diff: The current depth point is subtracted from the logging curve values ​​of the upper and lower 1-19 depth points respectively;

[0074] (f) Curve deflection: Calculates the deflection value within a 10-20 step window of the current depth;

[0075] (g) Kurtosis: Calculates the kurtosis value within a 10-20 step window of the current depth;

[0076] (h) Curve depth context features: Considering the correlation on the curve depth, and making full use of well logging information, the reservoir parameter prediction for the current sampling point is also related to the N adjacent upper / lower sampling points. The AI ​​model can consider these correlations and learn automatically. For example, it can take the data from the 10 upper / lower sampling points above / below the current sampling point, for a total of 21 data points, to represent the characteristics of the point.

[0077] (7) Select a suitable AI algorithm, train it using the prepared sample data, and output the original water saturation and the current water saturation model; the AI ​​algorithm includes traditional learning algorithms and deep learning algorithms;

[0078] Traditional learning algorithms include: Linear Learning, LightGBM (Gradient Boosting Machine Learning), XGBoost, Random Forest, AdaBoost, etc.; deep learning algorithms include: DNN (Fully Connected Neural Network), LSTM (Long Short-Term Memory), CNN+LSTM (a combination of Convolutional Neural Network and Long Short-Term Memory), etc. An algorithm can be selected based on requirements. Sample feature data is input into the algorithm for training, and the model is output. If the model accuracy does not meet requirements, the algorithm parameters can be adjusted.

[0079] Part Two: Model Prediction

[0080] The first step involves converting the logging data format, extracting curves, and redirecting them. This primarily involves extracting the corresponding curves from the LAS format file (or TXT, 716 format, etc.) of each well, converting them into CSV files, and converting any character data in the files into numerical data. Since the names of the same logging curve may differ between different wells, a standardization process is performed, redirecting the data to a unified name.

[0081] The second step involves processing invalid, outlier, and missing values ​​in the well logging data. Invalid, outlier, and missing values ​​all act as noise, directly impacting the machine learning performance. First, invalid values ​​are processed by deleting columns containing only invalid values ​​and then removing rows containing only invalid values, primarily cleaning up entire columns or rows with values ​​of 0 or -9999. Next, missing values ​​are processed using linear interpolation to fill in missing values ​​from adjacent sampling points, ensuring the integrity of the data used for machine learning. Finally, outliers are handled by truncation, allowing machine learning to learn from high-quality data and ensuring the model's accuracy and generalization ability.

[0082] The third step involves reconstructing the natural gamma curve and constructing the spontaneous potential mudstone baseline. To address the anomaly issue of the GR curve in the water-flooded layer, a GR reconstruction model is constructed. Unaffected curves are input into the GR reconstruction model to generate reconstructed values ​​to replace outliers. To address the significant differences in formation water salinity across different strata and the shift in the spontaneous potential mudstone baseline after water flooding of the oil layer, requiring improved accuracy in calculating the SP amplitude difference, a heuristic search is used to construct the mudstone baseline through a SP baseline walking method based on SP extreme points within a window range.

[0083] The fourth step is to merge the processed logging curve data with the thickness data and oil layer group information stored in dbf or Excel format after discretization.

[0084] Step 5: Feature extraction. The method for feature extraction is the same as that in Step 6 of the first part.

[0085] Step 6: Input the extracted feature data into the pre-trained model in, and predict the original water saturation and the current water saturation for each depth point within the thickness to be predicted in the well to be predicted, obtaining preliminary prediction results.

[0086] Step 7: Post-processing of prediction results. Specify the data rationality of the above prediction results. The specification principle is: if the predicted SWI≥100, then SWI is re-assigned 100; if the predicted SWI≤10, then SWI is re-assigned 10; if the predicted SW≥100, then SW is re-assigned 100; if the predicted SW≤0, then SW is re-assigned 0; if the predicted SW<SWI, then the value of SWI is assigned to SW, that is, SWI = SW (SW: current water saturation; SWI: original water saturation).

[0087] Example 1

[0088] This example uses Well Bei 2-362-Jian P25 in the northern part of the Daqing Placanticline to illustrate the method of the present invention.

[0089] Figure 1 is the waterflood interpretation result chart of Well Bei 2-362-Jian P25. The first track from the left on the chart is the deep lateral RLLD, shallow lateral RLLS, and microspherical resistivity curve RXO; the second track is the micro-gradient RMG and micro-potential curve RMN; the third track is the well diameter CAL, spontaneous potential SP, and natural gamma curve GR; the fourth track is the lithology density DEN and acoustic wave curve HAC. The original log curve values used in the calculation can be directly read from the chart; the fifth track on the chart is the depth track; the sixth track is the oil reservoir group; the seventh track is the effective thickness track; the eighth track with bar lines is the original water saturation from core analysis, and the continuous curve is the predicted original water saturation; the ninth track with bar lines is the current water saturation from core analysis, and the continuous curve is the predicted current water saturation. The specific calculation steps are as follows:

[0090] Step 1: First, extract the set curves RMN, RMG, RLLD, RLLS, RXO, DEN, HAC, SP, GR of Well Bei 2-362-Jian P25 from the Las file and convert them into a Csv file, and convert the possible character data in the file into numerical data.

[0091] Step 2: Clean the invalid values, missing values, and abnormal values of the log curves of Well Bei 2-362-Jian P25. For this well, it is necessary to remove the data with log curve values of 0 and -9999 above 930 meters and below 1210 meters in depth, and there are no missing values and abnormal values in the target interval.

[0092] The third step is data integration. The processed well logging data and core analysis data are merged with the thickness data and reservoir information stored in DBF or Excel format. Since the reservoir group and thickness data are given according to depth segments, the reservoir group information needs to be discretized according to depth points, aligned with the well logging data at the same depth, and then merged into a single CSS file.

[0093] The fourth step is multi-dimensional feature extraction of the logging curve. The original curve features need to be processed as follows:

[0094] (1) Original curve amplitude characteristics:

[0095] Micropotential RMN, microgradient RMG, deep lateral RLLD, shallow lateral RLLS, microsphere RXO, lithological density DEN, acoustic wave HAC, spontaneous potential SP, natural gamma GR;

[0096] (2) Curve transformation: RMN-RMG, LOG(RLLD), LOG(RLLS), DSP, HAC-DEN

[0097] (3) Normalize the original curve and use the min-max method to map the sample feature values ​​to the range (0, 1).

[0098] (4) Taking the current depth point as the center, simultaneously take the average, maximum, minimum, median, and variance of the curve within the step window of 18, 16, 14, 12, and 10 sampling points upwards and downwards;

[0099] Taking the RMN curve as an example, we take the following values ​​respectively:

[0100] RMN_window_18_avg、

[0101] RMN_window_18_max、

[0102] RMN_window_18_min......

[0103] RMN_window_10_avg、

[0104] RMN_window_10_max、

[0105] RMN_window_10_min;

[0106] (5) Differential feature Diff: The current depth point is subtracted from the logging curve values ​​of the upper and lower 1-19 depth points respectively;

[0107] (6) Curve deflection: Calculates the deflection value within a 10-20 step window of the current depth;

[0108] (7) Kurtosis of the curve: Calculate the kurtosis value within a window of 10 - 20 steps at the current depth.

[0109] (8) Curve depth context features: Considering the correlation in curve depth and making full use of logging information, for the prediction of reservoir parameters at the current sampling point, there is also a certain correlation with N adjacent upper / lower sampling points. The AI model can consider these correlations and learn automatically. For example, take the data of 10 upper / lower sampling points of the current sampling point, a total of 21 data to represent the features of this point.

[0110] The fifth step is model prediction. Input all the new data generated in the fourth step and the eigenvalue of the original curve into the established model, and predict the original water saturation and the current water saturation for each depth point within the effective thickness of the test well respectively to obtain the preliminary prediction results.

[0111] The sixth step is post - processing of the prediction results.

[0112] Specify the data rationality of the above prediction results. The specification principle is:

[0113] If the predicted SWI ≥ 100, then re - assign SWI as 100; if the predicted SWI ≤ 10, then re - assign SWI as 10; if the predicted SW ≥ 100, then re - assign SW as 100; if the predicted SW ≤ 0, then re - assign SW as 0; if the predicted SW < SWI, then assign the value of SWI to SW, that is, SWI = SW; (SW: current water saturation; SWI: original water saturation).

[0114] Post - process the prediction results according to the above specified principle and output the processed results. Obtain the accurate data of the original water saturation and the current water saturation.

[0115] The detailed prediction results of the original water saturation and the current water saturation of this well are shown in Table 1.

[0116] Table 1 Prediction results of the original water saturation and the current water saturation of the water - flooded zones in Well Bei 2 - 362 - Check P25

[0117]

[0118]

[0119]

[0120] To verify the feasibility of the method in this embodiment, the newly predicted original and current water saturation of the Bei 2-362-Jian P25 well were compared with the original and current water saturation from core analysis. The absolute error of the original water saturation was 3.37%, and the absolute error of the current water saturation was 5.67%. This is consistent with the core analysis and meets the requirements of the reserve specification: the absolute error of the original water saturation should be less than 5.0%, and the absolute error of the current water saturation should be less than 8.0%, indicating that the method in this embodiment is feasible.

[0121] Example 2

[0122] This embodiment uses the Daqing Changyuan North 2-350-Jian 45 well as an example to illustrate the method of the present invention. The well logging interpretation results for this well are shown in the figure below. Figure 2 Using the same calculation steps as in Example 1, the detailed prediction results of the original water saturation and current water saturation of the well are shown in Table 2.

[0123] Table 2. Results of Original and Predicted Water Saturation of the Water-Flooded Layer in Well Bei 2-350-Jian 45

[0124]

[0125]

[0126]

[0127]

[0128] To verify the feasibility of the method in this embodiment, the newly predicted original and current water saturation of well Bei 1-350-Jian 45 were compared with the original and current water saturation from core analysis. The absolute error of the original water saturation was 3.35%, and the absolute error of the current water saturation was 6.31%. This is consistent with the core analysis and meets the requirements of the reserve specification: the absolute error of the original water saturation should be less than 5.0%, and the absolute error of the current water saturation should be less than 8.0%, indicating that the method in this embodiment is feasible.

[0129] Example 3

[0130] Taking well Bei 2-357-Jian 82 as an example, the interpretation results of this well are shown in the figure below. Figure 3 Using the same calculation steps as in Example 1, the detailed prediction results of the original water saturation and current water saturation of the well are shown in Table 3.

[0131] Table 3. Results of Original and Predicted Water Saturation of the Water-Flooded Layer in Well Bei 2-357-Jian 82

[0132]

[0133]

[0134]

[0135] To verify the feasibility of the method in this embodiment, the newly predicted original and current water saturation of well Bei 2-357-Jian 82 were compared with the original and current water saturation from core analysis. The absolute error of the original water saturation was 4.26%, and the absolute error of the current water saturation was 5.69%. This is consistent with the core analysis and meets the requirements of the reserve specification: the absolute error of the original water saturation should be less than 5.0%, and the absolute error of the current water saturation should be less than 8.0%, indicating that the method in this embodiment is feasible.

[0136] This invention provides a method for determining the original and current water saturation of water-flooded layers based on big data analysis. This method is applied to the interpretation of infill wells in the Daqing Changyuan reservoir, guiding automatic interpretation of single-well vertical to layer analysis and multi-well horizontal comparison. It reduces labor intensity for interpreting water-flooded layers in new wells and reviewing old wells. Furthermore, it improves the quality of interpretation of newly drilled wells each year, reduces the workload of manual interpretation, and increases work efficiency, providing technical support for detailed reservoir geological description and precise tapping of remaining oil potential.

[0137] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, equivalent substitutions, and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for determining the original water saturation and the present water saturation of a watered-out zone, characterized by: The method comprises the following steps: First part: model training and learning Extracting according to the required curves for modeling, format conversion of the sample well logging data, and curve reorientation; Pretreatment of the converted sample well logging data; Reconstruction of the natural gamma ray curve in the sample well logging data; and construction of a shale baseline for the natural potential curve in the sample well logging data; Core data pretreatment; and depth matching of the pretreated core data and the logging curves; Merging the processed logging curve data with the thickness data, oil reservoir group information stored in the dbf format or Excel format, and the core analysis data matched in depth with the logging curves to form a csv sample well logging curve file; Multi-dimensional feature extraction on the merged csv logging curves; Selecting a suitable AI algorithm, training with the prepared sample data, and outputting the original water saturation and current water saturation models; Second part: model prediction Extracting according to the required curves for modeling, format conversion of the well logging data to be predicted, and curve reorientation; Pretreatment of the converted well logging data to be predicted; Reconstruction of the natural gamma ray curve in the well logging data to be predicted, and construction of a shale baseline for the natural potential curve in the well logging data to be predicted; Merging the processed logging curve data with the thickness data and oil reservoir group information stored in the dbf format or Excel format to form a csv well logging curve file to be predicted; Multi-dimensional feature extraction on the merged csv well logging curves to be predicted; Inputting the extracted feature data of the well to be predicted into the trained model to predict the original water saturation and current water saturation of each depth point in the thickness to be predicted, and obtaining a preliminary prediction result; Post-processing of the obtained preliminary prediction result to obtain accurate data of the original water saturation and current water saturation; The specific method comprises: Specifying the data rationality of the above prediction result, and the principle is: if the predicted SWI is greater than or equal to 100, then the SWI is revalued as 100; if the predicted SWI is less than or equal to 10, then the SWI is revalued as 10; if the predicted SW is greater than or equal to 100, then the SW is revalued as 100; if the predicted SW is less than or equal to 0, then the SW is revalued as 0; and if the predicted SW is less than the SWI, then the value of the SWI is assigned to the SW, i.e., SWI = SW. Wherein: SW is the current water saturation; and SWI is the original water saturation.

2. The method for determining original water saturation and present water saturation of a watered-out reservoir according to Claim 1, wherein, The method for format conversion of the sample well logging data and the well logging data to be predicted comprises the following steps: Extracting and converting the corresponding curves in the las format, txt format or 716 format file of each well into a csv file, and converting the character type data in the file into numerical type data.

3. The method for determining original water saturation and present water saturation of a watered-out reservoir according to claim 1, characterized in that, The method for pretreatment of the converted sample well logging data and the well logging data to be predicted comprises the following steps: The missing value processing method is to supplement the missing values in the sampling points according to the adjacent sampling points by using a linear interpolation algorithm; Abnormal values are processed by truncation; Invalid values are deleted.

4. The method for determining original and present water saturation of a watered-out reservoir according to Claim 1, wherein, The method for constructing a mudstone baseline for spontaneous potential curves in sample well logging data and well logging data to be predicted is based on a spontaneous potential baseline wandering method of spontaneous potential extreme points in a window range, and a mudstone baseline is constructed through heuristic search, specifically including: First, a window length is set, spontaneous potential extreme points are searched for in the first window length, after a first maximum value point is found, the two extreme points searched for are connected again by extending a window length downward from the limit; a mudstone baseline is constructed by analogy.

5. The method for determining original and present water saturation of a watered-out reservoir according to Claim 1, wherein, The core data preprocessing method includes: Core data standardization: all core data parameters except permeability are converted into percentage units; Remove missing values: only remove the null values corresponding to the current predicted parameters; Remove abnormal core analysis sample data: such as removing data with original water saturation values greater than the current water saturation value.

6. The method for determining original and present water saturation of a watered-out reservoir according to claim 1, wherein, Multi-dimensional feature extraction is performed on the merged csv sample well logging curves and well logging curves of the well to be predicted, including: extracting original curve numerical features, original curve transformation, curve normalization, maximum, minimum, average, median, variance, curve difference feature value, curve skewness, and kurtosis of each original logging curve in a certain step window.

7. The method for determining original and present water saturation of a watered-out reservoir according to Claim 1, wherein, The AI algorithm includes a traditional learning algorithm and a deep learning algorithm.

8. The method for determining original water saturation and present water saturation of a watered-out layer according to claim 7, characterized in that, The traditional learning algorithm includes: Linear, LightGBM, Xgboost, Random Forest, and AdaBooost. The deep learning algorithm includes: a fully connected neural network, a long short-term neural network, and a combination of a convolutional neural network and a long short-term neural network.

Images