Ultra-shallow low-contrast reservoir fluid identification method and system

Abstract

The invention belongs to the technical field of oil exploration and development, and relates to an ultra-shallow low-contrast reservoir fluid identification method and system, and the method comprises the steps: obtaining all logging and logging original response data of a target layer of a target region; calculating derivative parameters according to the logging and logging original response data; selecting characteristics of different types of fluids from the derivative parameters; preprocessing the selected feature data to obtain standardized feature data, and dividing the standardized feature data into a training set and a test set; training and verifying three models of a bagging tree, an LGBM model and an XGBoost model through a training set by taking the selected features as input and the fluid type as output; dynamically integrating the trained and tested three models into an integrated model through a voting mechanism, and verifying the integrated model through a test set; and inputting logging and logging response data and derivative parameters of a reservoir section with an unknown fluid type into the integrated model, and predicting the fluid type. According to the method, the fluid property of the ultra-shallow low-contrast reservoir in the sea area can be quickly and accurately identified.

Description

Technical Field

[0001] The invention belongs to the field of geophysical logging technology, relates to reservoir fluid identification technology, and specifically relates to an ultra-shallow low-contrast reservoir fluid identification method and system. Background Art

[0002] In current oil and gas exploration, ultra-shallow gas reservoirs in the sea are of great significance for increasing natural gas reserves and production due to their shallow burial and easy exploitation. Typical representatives of this type of gas reservoir include the ultra-shallow natural gas reservoir in the LD22-1 area of ​​Ledong, southeastern Qiongdong, which has the characteristics of fine lithology, high mud content, high formation water saturation and formation water mineralization, and the development of thin / low-resistance gas layers, resulting in low gas layer contrast and the "mirror" effect of the "acoustic time difference-density-neutron" three-porosity curve is not obvious. Existing technical methods, including intersection plate method, overlap map, porosity difference ratio and sensitive parameter method based on logging response or derived parameters, are not effective in thin / low-resistance / low-contrast gas layers. Special logging methods such as array acoustic wave and nuclear magnetic resonance logging can improve fluid identification effects, but the construction conditions in the sea area are harsh, the cost is high, and the quality of measurement data is limited, so their scope of use is limited. For example, in extremely deep gas-bearing reservoirs, the acoustic wave attenuation while drilling is obvious, which makes it difficult to effectively extract the acoustic wave time difference, thereby reducing the universality of these methods.

[0003] In recent years, machine learning methods have provided a new solution for fluid identification. Machine learning methods can efficiently process complex data, quickly establish nonlinear mapping relationships between conventional logging data and test data, and make full use of various existing original logging echoes and derived parameters to extract and highlight the differences between different fluid properties, thereby improving the recognition accuracy and efficiency of fluid properties. Common machine learning methods for fluid identification include single identification methods (for example: neural networks, random forests, XGBOOST, etc.) and dynamic classification committee methods based on a combination of multiple methods, which are used for fluid identification in different regions. However, these methods still have the following problems in the identification of fluids in ultra-shallow low-contrast reservoirs in the sea: (1) Existing models are usually based on training data from a specific region and cannot be directly applied to low-contrast fluid identification in ultra-shallow undiagenetic reservoirs such as the South China Sea. It is necessary to redesign the dataset and train the model for this region.

[0004] (2) Current methods mostly use a single type of measurement data (such as logging curves) as input, without fully considering the logging response and its derived parameters. There is a lack of systematic extraction of fluid sensitive characteristics. Direct input of raw logging data may result in limited fluid identification accuracy. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for identifying fluids in ultra-shallow low-contrast reservoirs. For all original logging responses and derived parameters, the correlation between features and target variables is maximized, and the redundancy between features is minimized. The most important features related to the fluid properties are selected, and multi-task machine learning is combined to quickly and accurately identify the fluid properties of ultra-shallow low-contrast reservoirs in the sea area, thereby solving the problem of fluid identification in ultra-shallow / low gas-resistance layers.

[0006] In a first aspect, the present invention provides a method for identifying fluid in an ultra-shallow low-contrast reservoir, the steps of which are: Data acquisition steps: obtain all logging and mud logging original response data of the target layer in the target area; Parameter calculation steps: Calculate the derived parameters based on the original response data of well logging and mud logging; Feature selection step: using mRMR algorithm to select features of different types of fluids from derived parameters; Data processing steps: preprocess the selected features to obtain standardized features, and divide the standardized features into training sets and test sets; Model training and validation steps: The selected features are used as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type is used as the output of the three models. The three models are trained and validated through the training set; Integration step: The three models of bagged trees, LGBM and XGBoost after training and testing are dynamically integrated into an integrated model through a voting mechanism, and the integrated model is verified through the test set; Fluid identification step: input the logging and mud logging response data and derived parameters of the reservoir section with unknown fluid type into the integrated model, and predict the fluid type through the integrated model.

[0007] In some embodiments, the original response data of well logging and mud logging of the target area's clear gas testing layer segment and the reservoir segment of the fluid type corresponding to the pressure measurement are obtained.

[0008] In some embodiments, the raw logging response data include natural gamma, deep lateral resistivity, shallow lateral resistivity, flushing zone resistivity, longitudinal wave time difference, compensated neutron, compensated density, and wellbore curve; the raw logging response data include gas measurement data, geothermal gradient, gas flow rate analyzed by gas measurement equipment and conversion coefficient of degasser displacement, well depth, drilling fluid outlet temperature, drilling time, sum of drilling pump circulation displacement and booster pump displacement, deviation compression coefficient of gas under formation conditions, cumulative effective thickness of reservoir, drilling wellbore diameter, and well fluid density. The gas measurement data include total hydrocarbons, alkane components, and carbon dioxide.

[0009] In some embodiments, in the parameter calculation step, the derived parameters calculated include P-wave velocity, P-wave impedance, shale content, apparent formation density porosity after shale correction, apparent formation density porosity, apparent formation neutron porosity after shale correction, apparent formation neutron porosity, apparent formation acoustic porosity after shale correction, apparent formation acoustic porosity, three-porosity difference, excavation index, total hydrocarbon methane ratio, and underground unit rock gas content, wherein:

[0010]

[0011]

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] 2 C=T g / C1

[0021] In the formula, , unit is m / s; is the longitudinal wave impedance, in Pa; is the relative value of natural gamma, dimensionless; is the natural gamma logging value of the target layer, in API; is the natural gamma logging value of pure lithologic formations, in API units, obtained based on the actual conditions of the target area, with a typical value of 0; is the natural gamma logging value of pure mudstone formation, in API units, obtained according to the actual situation of the target area, with a typical value of 150; is an empirical coefficient related to the age of the strata, 3.7 for new strata and 2 for old strata; is the formation mud content, unit is %; is the apparent formation density porosity corrected for mud quality, in %; is the measured density value, in g / cm 3 , read out by the logging response DEN; is the density of the rock skeleton in the formation, in g / cm 3 , obtained from the core experimental test in the target area, with a typical value of 2.71; is the density of mudstone, in g / cm 3 , obtained according to the actual situation of the target area, the typical value is 2.62; is the density of mud filtrate, in g / cm 3 , obtained according to the actual situation of the target area, the typical value is 1; is the apparent formation neutron porosity corrected for mud content, in %; is the measured formation neutron porosity, unit: % is the neutron porosity value of the formation rock skeleton, which is obtained based on the actual situation of the target area, and the typical value is 100%; is the neutron porosity value of the mud filtrate, which is obtained according to the actual situation of the target area, and the typical value is 100%; is the neutron porosity value of mudstone, which is obtained based on the actual situation of the target area, and the typical value is 25%; is the apparent formation acoustic porosity corrected for mud quality, in %; is the measured P-wave time difference, in us / m, obtained from the logging response DT; is the acoustic time difference of the rock skeleton of the formation, in us / m, obtained according to the actual situation of the target area, with a typical value of 47.5; is the acoustic time difference of mud filtrate, in us / m, obtained according to the actual situation of the target area, and the typical value is 189; is the mudstone acoustic wave time difference, in us / m, obtained according to the actual situation of the target area, with a typical value of 85; is the apparent formation density porosity without mud correction; is the apparent formation neutron porosity without mud correction; It is the apparent formation sonic porosity without mud correction; is the difference of three porosities, unit is %; is the defined mining index, dimensionless; C is the peak-to-base ratio, dimensionless; is the content of all hydrocarbon-containing substances measured by the instrument, C1 is the methane content; is the gas content of the underground unit rock; C n Represents the gas measurement value of each hydrocarbon component, n=1,2,3,4,5; E n Represents the degassing efficiency of each component, dimensionless; kIt is the conversion factor for the gas flow rate and degasser displacement analyzed by the gas measurement equipment; ROP is measured by the surface logging equipment and represents drilling time, which means the time required for the drill bit to penetrate one meter of the formation, and the unit is min / m; Flow is the displacement, which is equal to the sum of the circulation displacement of the drilling pump and the displacement of the booster pump, and the unit is L / Min; is the deviation compressibility coefficient of gas under formation conditions; G t is the geothermal gradient, in units of o C / m, h is the well depth in m; is the drilling fluid outlet temperature, in o C; is the cumulative effective thickness of the reservoir, in m; is the borehole diameter in mm, which is directly obtained from the drill bit size data; is the drilling fluid density, in g / cm 3 .

[0022] In some embodiments, in the feature selection step, the specific steps of selecting features using the mRMR algorithm are: Correlation calculation steps: Calculate the correlation between each feature and the target variable, and the correlation indicator is mutual information; Redundancy calculation step: calculate the redundancy between each pair of features; Feature selection step: Select features through the following objective function;

[0023] in, is the selected feature set, It is a feature With the target variable The mutual information between It is a feature With features The mutual information between Is a collection size; Update step: continuously update the feature set through iteration , in each iteration, select the until the preset stop condition is met.

[0024] In some embodiments, in the feature selection step, different types of fluids are divided into 2-8 categories according to the target area.

[0025] In some embodiments, the data processing step further comprises: Data cleaning step: perform data cleaning on the selected features and delete outliers; Standardization step: Use the Z-score method to standardize all features after data cleaning to obtain standardized features; Dataset division step: divide the standardized features into training set and test set according to the set ratio.

[0026] In some embodiments, in the model training and verification steps, a cross-validation method is used to perform model training verification. During training and verification, the hyperparameters of the bagged tree are set to the number of base tree models and the maximum depth, the hyperparameters of the LGBM are set to the learning rate, the number of trees and the maximum depth, and the hyperparameters of the XGBoost are set to the learning rate, the number of trees and the maximum depth. The training set is divided into multiple subsets, wherein any one subset is selected as the verification set, and the other subsets are used as training sets. During the training and verification process, each model undergoes multiple rounds of verification, the results of the multiple rounds of verification are summarized, the average performance indicators of the multiple rounds are calculated, and the performance indicators of each model are calculated.

[0027] In some embodiments, the method by which the integrated model selects the most suitable model for fluid prediction through a voting mechanism is as follows: based on the input features and the performance of each model on a specific feature combination, the most suitable model for fluid prediction is dynamically selected, and then weights are assigned to different models, and weighted based on each performance in model verification. The prediction results of each model are weighted according to their weights to obtain the final prediction results of each model, and a weighted vote is performed on the final prediction results of the three models, and the final prediction result with the most votes is selected as the prediction result of the integrated model.

[0028] In a second aspect, the present invention provides an ultra-shallow low-contrast reservoir fluid identification system, which is used to implement the ultra-shallow low-contrast reservoir fluid identification method described in the first aspect of the present invention, comprising: Data acquisition module, to obtain all the original response data of well logging and mud logging in the target area and target layer; Parameter calculation module, which calculates derived parameters based on the original response data of well logging and mud logging; Feature selection module, which uses mRMR algorithm to select features of different types of fluids from derived parameters; The data processing module preprocesses the selected features to obtain standardized features, and divides the standardized features into a training set and a test set; In the model training and validation module, the selected features are used as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type is used as the output of the three models. The three models are trained and validated through the training set; The integration module dynamically integrates the trained and tested bagging tree, LGBM and XGBoost models into an integrated model through a voting mechanism; The fluid identification module inputs the logging and mud logging response data and derived parameters of the reservoir section of unknown fluid type into the integrated model, and predicts the fluid type through the integrated model.

[0029] Compared with the prior art, the advantages and positive effects of the present invention are: (1) The ultra-shallow low-contrast reservoir fluid identification method and system provided by the present invention have high accuracy and fast speed for ultra-shallow / low-resistance / low-contrast reservoir fluid identification.

[0030] (2) The ultra-shallow low-contrast reservoir fluid identification method and system provided by the present invention dynamically integrates three model learning methods, namely, bagging trees, LGBM, and XGBoost, to form an integrated model. The structure and parameters of the integrated model can be adjusted in real time according to the changes in the fluid. It has better adaptability and can maintain good recognition performance under different reservoir conditions.

[0031] (3) The ultra-shallow low-contrast reservoir fluid identification method and system provided by the present invention performs fluid identification through automated feature selection and dynamic integration, thereby reducing the impact of human factors on fluid identification results, and is capable of processing large-scale data streams in real time with high recognition accuracy and fast recognition speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the ultra-shallow low-contrast reservoir fluid identification method according to an embodiment of the present invention; Figure 2 This is a structural block diagram of the ultra-shallow low-contrast reservoir fluid identification system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the feature importance score ranking calculated using the mRMR algorithm in an embodiment of the present invention; Figure 4 It is a schematic diagram of the true positive rate and false negative rate results of the five-fold cross-validation confusion matrix of the target area bagging tree model according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the true positive rate and false negative rate results of the confusion matrix of the target area LGBM model of the embodiment of the present invention with a five-fold cross validation; Figure 6 This is a schematic diagram of the true positive rate and false negative rate results of the five-fold cross-validation confusion matrix of the target area XGBoost model in an embodiment of the present invention; Figure 7 This is a schematic diagram of the true positive rate and false negative rate results of the confusion matrix of the target area integration model test set according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the prediction results of unknown fluid types in other wells in the target area of ​​an embodiment of the present invention.

[0033] In the figure, 1. Data acquisition module, 2. Parameter calculation module, 3. Feature selection module, 4. Data processing module, 41. Data cleaning module, 42. Standardization module, 43. Data set partitioning module, 5. Model training and verification module, 6. Integration module, 7. Fluid identification module. DETAILED DESCRIPTION

[0034] The present invention will be described in detail below by way of exemplary embodiments in conjunction with the accompanying drawings. However, it should be understood that, without further description, elements, structures and features in one embodiment may also be beneficially combined in other embodiments.

[0035] The present invention provides a method and system for identifying fluids in ultra-shallow low-contrast reservoirs, which obtains all the original logging and mud logging response data of the target layer in the target area; calculates derived parameters based on the original logging and mud logging response data; selects the characteristics of different types of fluids from the derived parameters; pre-processes the selected characteristic data to obtain standardized characteristic data, and divides them into a training set and a test set; uses the selected characteristics as input and the fluid type as output, and verifies the three models of bagging tree, LGBM and XGBoost through training with the training set; dynamically integrates the three models after training and testing into an integrated model through a voting mechanism, and verifies the integrated model through the test set; inputs the logging and mud logging response data and derived parameters of the reservoir section of the unknown fluid type into the integrated model to predict the fluid type. The above-mentioned method and system for identifying fluids in ultra-shallow low-contrast reservoirs are described in detail below in conjunction with the accompanying drawings.

[0036] See also Figure 1 The first embodiment of the present invention provides an ultra-shallow low-contrast reservoir fluid identification method, the steps of which are: S1. Data acquisition step: Acquire all the original response data of well logging and mud logging in the target area and target layer.

[0037] Obtain the original logging and mud logging response data of the reservoir section with clear gas testing layer section and fluid type corresponding to the pressure measurement in the target area.

[0038] Specifically, the raw logging response data include natural gamma, deep lateral resistivity, shallow lateral resistivity, flushing zone resistivity, P-wave time difference, compensated neutron, compensated density, and caliper curve.

[0039] Specifically, the raw logging response data include gas measurement data, geothermal gradient, conversion coefficient of gas flow analyzed by gas measurement equipment and degasser displacement, well depth, drilling fluid outlet temperature, drilling time, sum of drilling pump circulation displacement and booster pump displacement, deviation compression coefficient of gas under formation conditions, cumulative effective thickness of reservoir, drilling wellbore diameter, and well fluid density. The gas measurement data include total hydrocarbons, alkane components, and carbon dioxide.

[0040] S2. Parameter calculation step: Calculate the derived parameters based on the original response data of logging and mud recording.

[0041] In the parameter calculation step, the calculated derived parameters include P-wave velocity, P-wave impedance, shale content, apparent formation density porosity after shale correction, apparent formation density porosity, apparent formation neutron porosity after shale correction, apparent formation neutron porosity, apparent formation acoustic porosity after shale correction, apparent formation acoustic porosity, three-porosity difference, excavation index, total hydrocarbon methane ratio, and underground unit rock gas content, wherein:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] 2 C=T g / C1

[0053] In the formula, , unit is m / s; is the longitudinal wave impedance, in Pa; is the relative value of natural gamma, dimensionless; is the natural gamma logging value of the target layer, in API; is the natural gamma logging value of pure lithologic formations, in API units, obtained based on the actual conditions of the target area, with a typical value of 0; is the natural gamma logging value of pure mudstone formation, in API units, obtained according to the actual situation of the target area, with a typical value of 150; is an empirical coefficient related to the age of the strata, 3.7 for new strata and 2 for old strata; is the formation mud content, unit is %; is the apparent formation density porosity corrected for mud quality, in %; is the measured density value, in g / cm 3 , read out by the logging response DEN; is the density of the rock skeleton in the formation, in g / cm 3 , obtained from the core experimental test in the target area, with a typical value of 2.71; is the density of mudstone, in g / cm 3 , obtained according to the actual situation of the target area, the typical value is 2.62; is the density of mud filtrate, in g / cm 3 , obtained according to the actual situation of the target area, the typical value is 1; is the apparent formation neutron porosity corrected for mud content, in %; is the measured formation neutron porosity, unit: % is the neutron porosity value of the formation rock skeleton, which is obtained based on the actual situation of the target area, and the typical value is 100%; is the neutron porosity value of the mud filtrate, which is obtained according to the actual situation of the target area, and the typical value is 100%; is the neutron porosity value of mudstone, which is obtained based on the actual situation of the target area, and the typical value is 25%; is the apparent formation acoustic porosity corrected for mud quality, in %; is the measured P-wave time difference, in us / m, obtained from the logging response DT; is the acoustic time difference of the rock skeleton of the formation, in us / m, obtained according to the actual situation of the target area, with a typical value of 47.5; is the acoustic time difference of mud filtrate, in us / m, obtained according to the actual situation of the target area, and the typical value is 189; is the mudstone acoustic wave time difference, in us / m, obtained according to the actual situation of the target area, with a typical value of 85; is the apparent formation density porosity without mud correction; is the apparent formation neutron porosity without mud correction; It is the apparent formation sonic porosity without mud correction; is the difference of three porosities, unit is %; is the defined mining index, dimensionless; C is the peak-to-base ratio, dimensionless; is the content of all hydrocarbon-containing substances measured by the instrument, C1 is the methane content; is the gas content of the underground unit rock; C n Represents the gas measurement value of each hydrocarbon component, n=1,2,3,4,5; E n Represents the degassing efficiency of each component, dimensionless;k It is the conversion factor for the gas flow rate and degasser displacement analyzed by the gas measurement equipment; ROP is measured by the surface logging equipment and represents drilling time, which means the time required for the drill bit to penetrate one meter of the formation, and the unit is min / m; Flow is the displacement, which is equal to the sum of the circulation displacement of the drilling pump and the displacement of the booster pump, and the unit is L / Min; is the deviation compressibility coefficient of gas under formation conditions; G t is the geothermal gradient, in units of o C / m, h is the well depth in m; is the drilling fluid outlet temperature, in o C; is the cumulative effective thickness of the reservoir, in m; is the borehole diameter in mm, which is directly obtained from the drill bit size data; is the drilling fluid density, in g / cm 3 .

[0054] S3, feature selection step: Use the mRMR algorithm to select features of different types of fluids from the derived parameters. The purpose of using the mRMR (Minimum Redundancy Maximum Relevance) algorithm is to find a set of features (Min-Redundancy) in the original feature set (i.e., derived parameters) that have the greatest correlation with the final output result (Max-Relevance) but the smallest correlation between the features (Min-Redundancy). The mRMR algorithm is based on mutual information, calculates the importance scores of the original input variables and sorts them in descending order. By using the mRMR algorithm for feature selection, on the one hand, the most representative features are selected from high-dimensional data, and on the other hand, the redundancy between features is minimized, so that the selected features can be used for model training and verification later, improving model performance and computational efficiency.

[0055] Specifically, the specific steps of selecting features using the mRMR algorithm are: S31, correlation calculation step: Calculate the correlation between each feature and the target variable. It should be noted that the commonly used correlation index is mutual information, and the present invention uses mutual information to measure the correlation.

[0056] S32, redundancy calculation step: calculate the redundancy between each pair of features. It should be noted that redundancy is measured by mutual information, and the mutual information calculation formula is:

[0057] In the formula, is a random variable and The mutual information of is a random variable and The joint probability distribution of is a random variable The marginal probability distribution of is a random variable The marginal probability distribution of .

[0058] S33, feature selection step: select features through the following objective function;

[0059] in, is the selected feature set, It is a feature With the target variable The mutual information between It is a feature With features The mutual information between Is a collection size; S34, Update step: continuously update the feature set through iteration , in each iteration, select the features until the preset stop condition is met (for example, the number of selected features reaches a preset value). For example, based on the sorting results of the mutual information of each feature, 1-m features in the sequence are selected as the selected feature set.

[0060] In some embodiments, different types of fluids are divided into 2-8 categories according to the target area, for example, they can be divided into hydrates, gas layers, water layers, gas-water layers, gas-water layers, etc.

[0061] S4. Data processing step: preprocess the selected features to obtain standardized features, and divide the standardized features into training set and test set.

[0062] Specifically, the data processing step further includes: S41, data cleaning step: perform data cleaning on the selected features and delete outliers. Specifically, an outlier is a value that obviously does not conform to the logging rules, for example, -9999 is an outlier because the value deviates from the normal data range. By removing outliers, the quality and consistency of the selected features are ensured.

[0063] S42, standardization step: Use the Z-score method to standardize all features after data cleaning to obtain standardized features.

[0064] (17) in, is the feature data to be processed, is the mean of the overall data; is the standard deviation of the overall data, and Z is the value after the data is standardized.

[0065] The Z-score method is used to unify the scale of features by subtracting the mean of the data and dividing by the standard deviation, so that the data distribution has zero mean and unit standard deviation. This method effectively reduces the impact of scale differences between different features on subsequent model training.

[0066] S43, data set division step: divide the standardized features into training set and test set according to the set ratio. For example: divide into training set and test set according to the ratio of 8:2. It should be noted that the ratio of training set to test set can be set according to actual needs. Use a fixed ratio segmentation algorithm to divide the data set to ensure the fairness of data set division and avoid the appearance of features not covered by the training set in the test set. Divide the data into training set and test set. The training set is used for subsequent model training to avoid data leakage and model overfitting. The test set is used for subsequent evaluation of the generalization ability of the model.

[0067] S5. Model training and verification steps: The selected features are used as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type is used as the output of the three models. The three models are trained and verified through the training set.

[0068] Specifically, the cross-validation method is used for model training and verification. During training and verification, the hyperparameters of the bagged tree are set as the number of base tree models and the maximum depth, the hyperparameters of the LGBM are set as the learning rate, the number of trees and the maximum depth, and the hyperparameters of the XGBoost are set as the learning rate, the number of trees and the maximum depth. The training set is divided into multiple subsets, and one of the subsets is selected as the verification set, and the other subsets are used as the training set. During the training and verification process, each model is verified for multiple rounds, and the results of multiple rounds of verification are summarized, and the average performance indicators of multiple rounds are calculated, and the performance indicators of each model are calculated. Performance indicators include but are not limited to true positive rate, false negative rate, and accuracy.

[0069]

[0070]

[0071]

[0072] In the formula, TP is the positive sample predicted correctly by the model, is a negative sample that is predicted incorrectly by the model, is the negative sample predicted correctly by the model, are positive samples that are incorrectly predicted by the model.

[0073] Taking the five-fold cross validation method as an example, during training and validation, the training set is divided into five subsets, one of which is selected as the validation set, and the other subsets are used as the training set. During the training and validation process, each model undergoes five rounds of validation, the results of the five rounds of validation are summarized, the average performance index of the five rounds is calculated, and the performance index of each model is calculated.

[0074] S6, integration step: The three models of bagged trees, LGBM and XGBoost after training and testing are dynamically integrated into an integrated model through a voting mechanism, and the integrated model is optimized through test set evaluation.

[0075] Specifically, the method by which the integrated model selects the most suitable model for fluid prediction through a voting mechanism is: Based on the input features and the performance of each model on a specific feature combination, the most suitable model is dynamically selected for fluid prediction.

[0076] Then, weights are assigned to different models based on their performance in model validation, and the prediction results of each model are weighted according to their weights to obtain the final prediction results of each model;

[0077] In the formula, is the weight of the model, is the prediction result of the model.

[0078] The final prediction results of the three models are weighted voted, and the final prediction result with the most votes is selected as the prediction result of the integrated model. For example, the weight of the bagging tree is 0.3, and the predicted fluid type is hydrate; the weight of LGBM is 0.33, and the predicted fluid type is mixed layer; the weight of XGBoost is 0.37, and the predicted fluid type is mixed layer; then the integrated model finally predicts that the fluid type is mixed layer.

[0079] Specifically, the overall fluid prediction performance of the integrated model is evaluated through the test set. The evaluation indicators include calculation accuracy, recall rate, true positive rate, etc. The model parameters are tuned according to the evaluation results, including adjusting hyperparameters, improving dynamic selection mechanism, etc., and then training and optimization are carried out according to the evaluation feedback loop until satisfactory model performance is achieved.

[0080] S7, fluid identification step: input the logging and mud logging response data and derived parameters of the reservoir section of unknown fluid type into the integrated model, and predict the fluid type through the integrated model.

[0081] The ultra-shallow low-contrast reservoir fluid identification method of the present invention uses the mRMR algorithm for all original logging responses and derived parameters, maximizes the correlation between features and target variables, and minimizes the redundancy between features, selects the most important features related to the fluid properties, and combines multi-task machine learning to quickly and accurately identify the fluid properties of ultra-shallow low-contrast reservoirs in the sea area, thereby solving the problem of ultra-shallow / low gas-resistance layer fluid identification.

[0082] See also Figure 2 The second embodiment of the present invention provides an ultra-shallow low-contrast reservoir fluid identification system, which is used to implement the ultra-shallow low-contrast reservoir fluid identification method described in the first aspect of the present invention, comprising: Data acquisition module 1, to obtain all the original response data of well logging and mud logging in the target area and target layer; Parameter calculation module 2, calculates derived parameters based on the original response data of well logging and mud logging; Feature selection module 3, using mRMR algorithm to select features of different types of fluids from derived parameters; Data processing module 4, preprocessing the selected features to obtain standardized features, and dividing the standardized features into a training set and a test set; Model training and validation module 5, using the selected features as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type as the output of the three models, and training and validating the three models through the training set; Integration module 6 dynamically integrates the trained and tested bagging tree, LGBM and XGBoost models into an integrated model through a voting mechanism; The fluid identification module 7 inputs the reservoir segment logging and mud logging response data and derived parameters of the unknown fluid type into the integrated model, and predicts the fluid type through the integrated model.

[0083] Specifically, in some embodiments, the data processing module 4 further includes: The data cleaning module 41 performs data cleaning on the selected features and deletes outliers; A standardization module 42 uses a Z-score method to standardize all features after data cleaning to obtain standardized features; The data set division module 43 divides the standardized features into a training set and a test set according to a set ratio.

[0084] The data processing module uses an outlier removal mechanism based on business rules to clean data, and realizes automatic screening by setting an outlier threshold (such as "-9999"). Standardization is achieved by calculating the mean and standard deviation to ensure that different features have similar numerical distributions after standardization, thereby improving the efficiency of model training. A fixed ratio splitting algorithm (such as 8:2) is used to divide the data set to ensure fairness in data set division and avoid features that are not covered by the training set in the test set.

[0085] The ultra-shallow low-contrast reservoir fluid identification system of the present invention uses the mRMR algorithm for all original logging responses and derived parameters, maximizes the correlation between features and target variables, and minimizes the redundancy between features, selects the most important features related to the fluid properties, and combines multi-task machine learning to quickly and accurately identify the fluid properties of ultra-shallow low-contrast reservoirs in the sea area, thereby solving the problem of fluid identification in ultra-shallow / low gas-resistance layers.

[0086] In order to verify the effectiveness of the method and system for identifying fluid in ultra-shallow low-contrast reservoirs described in the above embodiments of the present invention, the following specific embodiments are used for description.

[0087] Embodiment: Collect the original logging data of the target area's clear gas test layer segment and the reservoir segment of the fluid type corresponding to the pressure measurement. Calculate the derived parameters based on the original logging and measurement responses, and the calculated derived parameters include P-wave velocity, P-wave impedance, shale content, apparent formation density porosity after shale correction, apparent formation density porosity, apparent formation neutron porosity after shale correction, apparent formation neutron porosity, apparent formation acoustic porosity after shale correction, apparent formation acoustic porosity, three-porosity difference, excavation index, total hydrocarbon methane ratio, and underground unit rock gas content. Select the top 7 features with the highest importance scores, as shown in the attached figure. Figure 3 As shown, including longitudinal wave impedance , shallow lateral resistivity RLLS, natural gamma GR, excavation index , Gas content of underground unit rock , peak-to-base ratio C, three porosity differences , the remaining features are not selected to reduce the amount of calculation and increase the running speed of the model. Different types of fluids are divided into 6 categories according to the needs of the target area, including hydrates, mixed layers, gas layers, water layers, gas-water layers, and gas-water layers. Bag tree: Set the number of basic tree models to 100 and the maximum depth to 10. LGBM: Set the learning rate to 0.01, the number of trees to 500, the maximum depth to 6, and the minimum number of sample leaves to 60. XGBoost: Set the learning rate to 0.05, the number of trees to 500, the maximum depth to 6, and the minimum number of sample leaves to 8. Then perform a five-fold cross validation, divide the training set evenly into five subsets, select one of the subsets as the validation set, and the other four subsets as the training set, train each model and evaluate it, and finally, summarize the results of the five rounds of validation, calculate the average performance indicators of the five rounds, and then calculate the performance indicators of each model, including the true rate TPR and the false negative rate FNR. The confusion matrix results of the cross-validation of the three models are shown in the attached figure Figure 4 , Attachment Figure 5 , Attachment Figure 6 As shown, the confusion matrix results of the integrated model test set are as follows Figure 7 The prediction results of unknown fluid types in other wells in the target area are shown in Figure 8 As shown, the predicted fluid type corresponds well to the gas logging curve. It can be seen that the ultra-shallow low-contrast reservoir fluid identification method and system described in the above embodiment of the present invention can be effectively used for fluid identification in other unknown wells.

[0088] The above embodiments are used to explain the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for identifying fluid in ultra-shallow low-contrast reservoirs, characterized in that: The steps are: Data acquisition steps: obtain all logging and mud logging original response data of the target layer in the target area; Parameter calculation steps: Calculate the derived parameters based on the original response data of well logging and mud logging; Feature selection step: using mRMR algorithm to select features of different types of fluids from derived parameters; Data processing steps: preprocess the selected features to obtain standardized features, and divide the standardized features into training sets and test sets; Model training and validation steps: The selected features are used as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type is used as the output of the three models. The three models are trained and validated through the training set; Integration step: The three models of bagged trees, LGBM and XGBoost after training and testing are dynamically integrated into an integrated model through a voting mechanism, and the integrated model is verified through the test set; Fluid identification step: input the logging and mud logging response data and derived parameters of the reservoir section with unknown fluid type into the integrated model, and predict the fluid type through the integrated model.

2. The ultra-shallow low-contrast reservoir fluid identification method according to claim 1, characterized in that: In the data acquisition step, the original response data of well logging and mud logging of the target area's clear gas testing layer section and the reservoir section of the fluid type corresponding to the pressure measurement are obtained.

3. The ultra-shallow low-contrast reservoir fluid identification method according to claim 2, characterized in that: The raw logging response data include natural gamma, deep lateral resistivity, shallow lateral resistivity, flushing zone resistivity, longitudinal wave time difference, compensated neutron, compensated density, and wellbore curve; the raw logging response data include gas measurement data, geothermal gradient, gas flow analyzed by gas measurement equipment and conversion coefficient of degasser displacement, well depth, drilling fluid outlet temperature, drilling time, sum of drilling pump circulation displacement and booster pump displacement, deviation compression coefficient of gas under formation conditions, cumulative effective thickness of reservoir, drilling wellbore diameter, and well fluid density. The gas measurement data include total hydrocarbons, alkane components, and carbon dioxide.

4. The ultra-shallow low-contrast reservoir fluid identification method according to claim 3, characterized in that: In the parameter calculation step, the derived parameters calculated include P-wave velocity, P-wave impedance, shale content, apparent formation density porosity after shale correction, apparent formation density porosity, apparent formation neutron porosity after shale correction, apparent formation neutron porosity, apparent formation acoustic porosity after shale correction, apparent formation acoustic porosity, three-porosity difference, excavation index, total hydrocarbon methane ratio, and underground unit rock gas content, wherein: 2 C=T g / C1 In the formula, , unit is m / s; is the longitudinal wave impedance, in Pa; is the relative value of natural gamma, dimensionless; is the natural gamma logging value of the target layer, in API; It is the natural gamma logging value of pure lithologic formations, in API units, obtained based on the actual conditions of the target area; is the natural gamma logging value of pure mudstone formation, in API units, obtained based on the actual situation of the target area; is an empirical coefficient related to the age of the strata, 3.7 for new strata and 2 for old strata; is the formation mud content, unit is %; is the apparent formation density porosity corrected for mud quality, in %; is the measured density value, in g / cm 3 , read out by the logging response DEN; is the density of the rock skeleton in the formation, in g / cm 3 , obtained from core experimental tests in the target area; is the density of mudstone, in g / cm 3 , obtained according to the actual situation of the target area; is the density of mud filtrate, in g / cm 3 , obtained according to the actual situation of the target area; is the apparent formation neutron porosity corrected for mud content, in %; is the measured formation neutron porosity, unit: % is the neutron porosity value of the formation rock skeleton, which is obtained based on the actual situation of the target area; is the neutron porosity value of the mud filtrate, which is obtained according to the actual situation of the target area; is the neutron porosity value of mudstone, which is obtained according to the actual situation of the target area; is the apparent formation acoustic porosity corrected for mud quality, in %; is the measured P-wave time difference, in us / m, obtained from the logging response DT; is the acoustic time difference of the rock skeleton of the formation, in us / m, obtained according to the actual situation of the target area; is the acoustic time difference of mud filtrate, in us / m, obtained according to the actual situation of the target area; is the mudstone acoustic wave time difference, in us / m, obtained according to the actual situation of the target area, with a typical value of 85; is the apparent formation density porosity without mud correction; is the apparent formation neutron porosity without mud correction; It is the apparent formation sonic porosity without mud correction; is the difference of three porosities, unit is %; is the defined mining index, dimensionless; C is the peak-to-base ratio, dimensionless; is the content of all hydrocarbon-containing substances measured by the instrument, C1 is the methane content; is the gas content of the underground unit rock; C n Represents the gas measurement value of each hydrocarbon component, n=1,2,3,4,5; E n Represents the degassing efficiency of each component, dimensionless; k It is the conversion factor for the gas flow rate and degasser displacement analyzed by the gas measurement equipment; ROP is measured by the surface logging equipment and represents drilling time, which means the time required for the drill bit to penetrate one meter of the formation, and the unit is min / m; Flow is the displacement, which is equal to the sum of the circulation displacement of the drilling pump and the displacement of the booster pump, and the unit is L / Min; is the deviation compressibility coefficient of gas under formation conditions; G t is the geothermal gradient, in units of o C / m, h is the well depth in m; is the drilling fluid outlet temperature, in o C; is the cumulative effective thickness of the reservoir, in m; is the borehole diameter in mm, which is directly obtained from the drill bit size data; is the drilling fluid density, in g / cm 3 .

5. The ultra-shallow low-contrast reservoir fluid identification method according to claim 1, characterized in that: In the feature selection step, the specific steps of selecting features using the mRMR algorithm are: Correlation calculation steps: Calculate the correlation between each feature and the target variable, and the correlation indicator is mutual information; Redundancy calculation step: calculate the redundancy between each pair of features; Feature selection step: Select features through the following objective function; in, is the selected feature set, It is a feature With the target variable The mutual information between It is a feature With features The mutual information between Is a collection size; Update step: continuously update the feature set through iteration , in each iteration, select the until the preset stop condition is met.

6. The ultra-shallow low-contrast reservoir fluid identification method according to claim 1, characterized in that: In the feature selection step, different types of fluids are divided into 2-8 categories according to the target area.

7. The ultra-shallow low-contrast reservoir fluid identification method according to claim 1, characterized in that: The data processing step further comprises: Data cleaning step: perform data cleaning on the selected features and delete outliers; Standardization step: Use the Z-score method to standardize all features after data cleaning to obtain standardized features; Dataset division step: divide the standardized features into training set and test set according to the set ratio.

8. The ultra-shallow low-contrast reservoir fluid identification method according to claim 1, characterized in that: In the model training and verification steps, the cross-validation method is used for model training and verification. During training and verification, the hyperparameters of the bagged tree are set as the number of base tree models and the maximum depth, the hyperparameters of LGBM are set as the learning rate, the number of trees and the maximum depth, and the hyperparameters of XGBoost are set as the learning rate, the number of trees and the maximum depth. The training set is divided into multiple subsets, and one of the subsets is selected as the verification set, and the other subsets are used as the training set. During the training and verification process, each model undergoes multiple rounds of verification, the results of multiple rounds of verification are summarized, the average performance indicators of multiple rounds are calculated, and the performance indicators of each model are calculated.

9. The ultra-shallow low-contrast reservoir fluid identification method according to claim 8, characterized in that: The method by which the integrated model selects the most suitable model for fluid prediction through a voting mechanism is as follows: based on the input features and the performance of each model on a specific feature combination, the most suitable model is dynamically selected for fluid prediction, and then weights are assigned to different models, and weighted based on each performance in model verification. The prediction results of each model are weighted according to their weights to obtain the final prediction results of each model, and weighted voting is performed on the final prediction results of the three models, and the final prediction result with the most votes is selected as the prediction result of the integrated model.

10. An ultra-shallow low-contrast reservoir fluid identification system, used to implement the ultra-shallow low-contrast reservoir fluid identification method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, to obtain all the original response data of well logging and mud logging in the target area and target layer; Parameter calculation module, which calculates derived parameters based on the original response data of well logging and mud logging; Feature selection module, which uses mRMR algorithm to select features of different types of fluids from derived parameters; The data processing module preprocesses the selected features to obtain standardized features, and divides the standardized features into a training set and a test set; In the model training and validation module, the selected features are used as the input of the three models of bagging tree, LGBM and XGBoost, and the fluid type is used as the output of the three models. The three models are trained and validated through the training set; The integration module dynamically integrates the trained and tested bagging tree, LGBM and XGBoost models into an integrated model through a voting mechanism; The fluid identification module inputs the logging and mud logging response data and derived parameters of the reservoir section of unknown fluid type into the integrated model, and predicts the fluid type through the integrated model.

Images