Reservoir fluid property discrimination method based on logging data

Abstract

The invention discloses a reservoir fluid property discrimination method based on logging data, and relates to the technical field of natural gas exploration, and the method comprises the following steps: S1, collecting logging gas logging data and logging data, and carrying out the depth matching of the logging gas logging data and the logging data, and taking the data as a depth data set; s2, generating an interpretation conclusion label; s3, utilizing the interpretation conclusion label to extract a depth section trend parameter; s4, extracting a training set and a test set; s5, training and verifying the reservoir fluid property prediction model; s6, reservoir fluid property discrimination is completed. According to the reservoir fluid property distinguishing method based on the logging data, the advantages of logging and well logging technologies are integrated, potential information provided by the logging gas logging data and the well logging data is fully utilized, the accurate coincidence rate of reservoir fluid property recognition is increased, and the method is easy and convenient to operate, small in manual intervention and high in practicability. The efficiency and objectivity of the interpretation method can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of natural gas exploration, and in particular to a reservoir fluid property identification method based on well logging data. Background Art

[0002] In oil exploration, fluid property identification helps determine the type of fluid in the reservoir (such as oil layer, gas layer, water layer and dry layer, etc.), which is crucial for evaluating the value and potential of oil and gas reservoirs. In addition, understanding fluid properties can help geologists and petroleum engineers estimate the production capacity of the reservoir. Therefore, accurate evaluation of reservoir fluid properties is extremely important for oil exploration.

[0003] In current oil exploration, the evaluation of reservoir fluid properties usually relies on the combination of logging data and well logging data. Logging data can be monitored in real time during the drilling process to obtain first-hand physical information and provide timely and direct reference for geological engineers. Well logging data indirectly characterizes fluid properties by using logging instruments to obtain geophysical properties such as conductivity, acoustic properties, and radioactivity of rock formations. These two technologies each have their own unique advantages and are usually used in combination to improve the accurate evaluation of fluid properties.

[0004] The advantage of logging technology lies in its real-time and directness. It can continuously obtain formation information during the drilling process, which is difficult to replace by other methods. However, logging data also has some limitations, such as it is difficult to reflect reservoir physical property information. Well logging technology is known for its high precision and multi-parameter measurement capabilities. Through logging instruments, high-resolution reservoir physical properties and fluid property data can be obtained. However, logging technology is expensive, requires professional equipment and technicians, and the data interpretation is complex and may be affected by formation conditions, resulting in certain uncertainties.

[0005] Although logging and well logging technologies have their own advantages, the current evaluation methods usually establish corresponding standards for each, making it difficult to effectively combine the data of the two for comprehensive evaluation. This separate evaluation method may lead to insufficient information utilization, thus affecting the rapid and accurate evaluation of the reservoir. Therefore, it is urgent to propose a reservoir fluid property discrimination method based on logging and well logging data to achieve rapid evaluation of reservoir fluid properties. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a reservoir fluid property identification method based on well logging data.

[0007] The technical solution of the present invention is: a method for determining reservoir fluid properties based on well logging data comprises the following steps:

[0008] S1, collecting mud logging gas measurement data and well logging data, and performing depth matching on the mud logging gas measurement data and the well logging data as a depth data set;

[0009] S2, generate explanation conclusion labels for deep datasets;

[0010] S3, extracting depth segment trend parameters using interpretation conclusion labels;

[0011] S4, extracting a training set and a test set from the depth data set using the depth segment trend parameter;

[0012] S5, using the training set and the test set to train and verify the reservoir fluid property prediction model in sequence;

[0013] S6. Use the verified reservoir fluid property prediction model to complete the reservoir fluid property identification.

[0014] Furthermore, in S1, the calculation formula for depth matching is:

[0015]

[0016] d = round(y) + 1;

[0017] In the formula, c is the top depth after treatment, x is the original top depth, d is the bottom depth after treatment, y is the original bottom depth, and round(·) is used to round y. Indicates rounding down.

[0018] Furthermore, S3 includes the following sub-steps:

[0019] S31. Extract the mean, variation amplitude, peak value, standard deviation and slope of the depth data set using the interpretation conclusion label;

[0020] S32, calculating a comprehensive distance metric for the depth dataset;

[0021] S33, calculating the interaction parameter of the depth data set according to the comprehensive distance metric of the depth data set;

[0022] S34, calculating the stability parameter of the depth data set;

[0023] S35. The mean, variation amplitude, peak value, standard deviation, slope, comprehensive distance metric, interaction parameter and stability parameter of the depth data set are used as depth segment trend parameters.

[0024] Furthermore, in S32, the calculation formula of the comprehensive distance metric D of the depth data set is:

[0025] D=D depth (x i ,y i )+D G+R (x i ,y i );

[0026] In the formula, x i Represents the i-th parameter in the depth dataset, y i Indicates the i-th parameter in the data sequence to be matched, D depth (·) represents the depth parameter distance, D G+R (·) indicates the distance converted from similarity.

[0027] Furthermore, in S33, the calculation formula of the interaction parameter z of the depth data set is:

[0028] z=log(1+|TG|·D 2 -exp(-D)·sin(TG));

[0029] Where log(·) represents the logarithmic function, TG represents the total gas measurement, exp(·) represents the exponential function, and D represents the comprehensive distance metric of the depth dataset.

[0030] Further, in S34, the depth data set includes stability parameters SI of 2 to 5 data points. 2～5 The calculation formula is:

[0031]

[0032] In the formula, ω i represents the weight of the i-th parameter in the depth dataset, x i represents the i-th parameter in the depth dataset, Mean represents the mean of the depth dataset, n represents the number of parameters in the depth dataset, Stand deviation represents the standard deviation of the depth dataset, and II(·) represents the indicator function;

[0033] In S34, the stability parameter SI for the depth data set contains more than 5 data points >5 The calculation formula is:

[0034]

[0035] In the formula, median represents the median of the depth data set, x1 represents the first parameter in the depth data set, x2 represents the second parameter in the depth data set, x3 represents the third parameter in the depth data set, and x n Represents the nth parameter in the depth dataset.

[0036] Furthermore, in S4, a data subset having pressure sampling or formation test conclusions is extracted from the depth data set using the depth segment trend parameter, and the data subset is split into a test set and a training set.

[0037] Furthermore, in S5, the regularization function Ω(fk ) is:

[0038]

[0039] In the formula, f k represents the model of the kth tree, γ represents the first regularization parameter, T represents the number of leaf nodes in the tree, j represents the leaf node index, β represents the adjustment weight, λ represents the second regularization parameter, ω j represents the weight of model learning, α represents the third conventional regularization parameter, and h j represents the thickness characteristic, and exp(·) represents the exponential function.

[0040] The beneficial effects of the present invention are as follows: the reservoir fluid property identification method based on mud logging data provided by the present invention combines the advantages of mud logging and well logging technology, fully utilizes the potential information provided by mud logging gas measurement data and well logging data, improves the accuracy and compliance rate of reservoir fluid property identification, and the method is simple to operate, requires less manual intervention, and can improve the efficiency and objectivity of the interpretation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The flowchart of the method for determining reservoir fluid properties based on logging data is shown in FIG. DETAILED DESCRIPTION

[0042] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0043] like Figure 1 As shown, the present invention provides a reservoir fluid property identification method based on well logging data, comprising the following steps:

[0044] S1, collecting mud logging gas measurement data and well logging data, and performing depth matching on the mud logging gas measurement data and the well logging data as a depth data set;

[0045] S2, generate explanation conclusion labels for deep datasets;

[0046] S3, extracting depth segment trend parameters using interpretation conclusion labels;

[0047] S4, extracting a training set and a test set from the depth data set using the depth segment trend parameter;

[0048] S5, using the training set and the test set to train and verify the reservoir fluid property prediction model in sequence;

[0049] S6. Use the verified reservoir fluid property prediction model to complete the reservoir fluid property identification.

[0050] In S1, the exploration well data with both gas logging and logging while drilling projects in the same block in recent years are collected, and the exploration wells must have oil and gas displays. The data is mainly stored in the form of a table output by the sensors of the field logging instrument and the logging instrument. The gas logging and logging data are both data rows of continuous depth points. Among them, the resolution of logging gas logging data is 1 data row per meter, and the data columns include gas logging total volume (TG), methane (C1), ethane (C2), propane (C3), isobutane (iC4), normal butane (nC4), isopentane (iC5) and normal pentane (nC5); the resolution of logging data is 1 data row per 0.1 meter, and the data columns include natural gamma, resistivity, acoustic wave time difference, neutron and density. The gas logging and logging display segment data is collected, and the data columns include top depth, bottom depth, oil occurrence, rock cutting fluorescence level and interpretation conclusion; the logging interpretation segment data is collected, and the data columns include top depth, bottom depth and interpretation conclusion.

[0051] In S2, interpretation conclusions that have been verified by pressure sampling or formation testing are preferred; for formations with a reservoir thickness of 1m to 2m, the logging interpretation conclusion with a higher resolution is used. For formations with a reservoir thickness of more than 2m, the logging interpretation conclusion is used. The interpretation conclusion labels are oil layer, water layer and dry layer, respectively, and are marked with serial numbers 0, 1, and 2.

[0052] In the embodiment of the present invention, in S1, the calculation formula for performing depth matching is:

[0053]

[0054] d = round(y) + 1;

[0055] In the formula, c is the top depth after treatment, x is the original top depth, d is the bottom depth after treatment, y is the original bottom depth, and round(·) is used to round y. Indicates rounding down.

[0056] In this embodiment of the present invention, S3 includes the following sub-steps:

[0057] S31. Extract the mean, variation amplitude, peak value, standard deviation and slope of the depth data set using the interpretation conclusion label;

[0058] S32, calculating a comprehensive distance metric for the depth dataset;

[0059] S33, calculating the interaction parameter of the depth data set according to the comprehensive distance metric of the depth data set;

[0060] S34, calculating the stability parameter of the depth data set;

[0061] S35. The mean, variation amplitude, peak value, standard deviation, slope, comprehensive distance metric, interaction parameter and stability parameter of the depth data set are used as depth segment trend parameters.

[0062] In the embodiment of the present invention, in S32, the calculation formula of the comprehensive distance metric D of the depth data set is:

[0063] D=D depth (x i ,y i )+D G+R (x i ,y i );

[0064] In the formula, x i Represents the i-th parameter in the depth dataset, y i Indicates the i-th parameter in the data sequence to be matched, D depth (·) represents the depth parameter distance, D G+R (·) indicates the distance converted from similarity.

[0065] The present invention selects typical display segment data verified by pressure sampling or formation testing in adjacent wells as alignment objects. These data constitute a sequence called X = {x1, x2, ..., x n}, and the data sequence to be matched is called sequence Y = {y1, y2, ..., y m}, considering the difference in length between the two sequences and that distance calculation does not rely solely on a single depth measurement method, but should be based on natural gamma, resistivity and other parameters to comprehensively evaluate the similarity between the sequences. A matching strategy suitable for the data characteristics of the study area is adopted. For the different thicknesses of the basic depth segments, a depth adjustment factor f is introduced in sections, namely, thin layers after 2 meters, medium-thick layers from 2 meters to 5 meters, and thick layers above 5 meters. d , defined as Where, d i and d j Respectively represent display segment x i and i The depth value of , v and p are constants that control the smoothness and attenuation amplitude of the large difference logarithmic power decay. Therefore, the depth parameter distance D formula is D depth (x i ,y i )=f d ·|d i -d j |.

[0066] In the process of drilling, natural gamma and resistivity logging curves can reflect the formation properties, so a feature space can be constructed to establish a measure of similarity. The depth, natural gamma and resistivity of the sequence X are regarded as points in a high-dimensional feature space, and each point can be represented as a vector X. i =[d i ,G i ,R i ], each point in the sequence Y can be represented as a vector Y i =[d j ,G j ,R j ]. Where d i , G i , R j Represent the depth, natural gamma value and resistivity value of X series respectively. j ,G j ,R j Represent the depth, natural gamma value and resistivity value of the Y series respectively.

[0067] After constructing the feature space, different weight coefficients are used to establish the comprehensive similarity S of natural gamma and resistivity between points according to the reservoir characteristics of different study areas. The formula is: In the formula, S(x i ,y i ) represents the natural gamma and resistivity at a certain point, α and β are weight coefficients (the sum of the two is 1).

[0068] Convert similarity into distance D G+R , the formula is Finally, for D depth and D G+R The sum is obtained to obtain the comprehensive distance metric D.

[0069] In the embodiment of the present invention, in S33, the calculation formula of the interaction parameter z of the depth data set is:

[0070] z=log(1+|TG|·D 2 -exp(-D)·sin(TG));

[0071] Where log(·) represents the logarithmic function, TG represents the total gas measurement, exp(·) represents the exponential function, and D represents the comprehensive distance metric of the depth dataset.

[0072] In the embodiment of the present invention, in S34, the depth data set includes stability parameters SI of 2 to 5 data points. 2～5 The calculation formula is:

[0073]

[0074] In the formula, ωi represents the weight of the i-th parameter in the depth dataset, x i represents the i-th parameter in the depth dataset, Mean represents the mean of the depth dataset, n represents the number of parameters in the depth dataset, Stand deviation represents the standard deviation of the depth dataset, and II(·) represents the indicator function;

[0075] In S34, the stability parameter SI for the depth data set contains more than 5 data points >5 The calculation formula is:

[0076]

[0077] In the formula, median represents the median of the depth data set, x1 represents the first parameter in the depth data set, x2 represents the second parameter in the depth data set, x3 represents the third parameter in the depth data set, and x n Represents the nth parameter in the depth dataset.

[0078] In the embodiment of the present invention, in S4, a data subset having pressure sampling or formation test conclusions is extracted from the depth data set using the depth segment trend parameter, and the data subset is split into a test set and a training set.

[0079] In the embodiment of the present invention, in S5, the regularization function Ω(f k ) is:

[0080]

[0081] In the formula, f k represents the model of the kth tree, γ represents the first regularization parameter, T represents the number of leaf nodes in the tree, j represents the leaf node index, β represents the adjustment weight, λ represents the second regularization parameter, ω j represents the weight of model learning, α represents the third conventional regularization parameter, and h j represents the thickness characteristic, and exp(·) represents the exponential function.

[0082] In an embodiment of the present invention, an XGBoost machine learning algorithm is selected to establish a reservoir fluid property prediction model, and the preferred model parameters are n_estimators: 108, max_depth: 7, min_child_weight: 4, gamma: 0, subsample: 0.8, colsample_bytree: 0.8, and learning_rate: 0.1.

[0083] In order to accelerate the convergence of the model in the early stage of training and make fine adjustments in the later stage of training to avoid overfitting, a learning rate strategy is adopted to adapt to the changes in model complexity through curvature information, specifically:

[0084] First, the second-order derivative (curvature) of the loss function at round t is calculated, which can be approximated by using the cumulative sum of gradients, specifically: In the formula, g i (t) represents the gradient of the i-th sample at the t-th iteration. Then adjust the learning rate. The learning rate η(t) is dynamically adjusted according to the curvature value K(t), and its calculation formula is In the formula, η0 represents the initial learning rate and β represents the adjustment weight. Then the learning rate is smoothly adjusted. In the formula, γ represents the smoothing weight. Finally, the dynamic step size is adjusted. In a specific iteration cycle, the loss change rate is calculated. If the loss decreases faster, the learning rate is appropriately increased, otherwise it is reduced. Specifically,

[0085]

[0086] Set к, ∈ and δ as adjustment coefficients. If r(t)>к, it means that the loss decreases rapidly and the learning rate can be increased. If r(t)<∈, it means that the loss decreases slowly and the learning rate should be reduced. Specifically,

[0087] Input the test set data into the trained model for verification.

[0088] By comparing the prediction results with the labels of the test set, the model performance is evaluated using precision, recall, and F1-score. The evaluation results are shown in Table 1, indicating that the model performance is good.

[0089] Table 1

[0090]

[0091]

[0092] Input recent block data (data not involved in training) into the model for prediction, and run the model to obtain specific reservoir fluid property classification results. The classification result conformity statistics are shown in Table 2.

[0093] Table 2

[0094] well location Correct classification level Classification error layer Total number of layers Correct rate (%) 1 Well 1 0 1 100 2 Wells 39 12 51 76.47 3 wells 6 1 7 85.71 4 Wells 19 1 20 95

[0095] The results show that in practical applications, the model classification accuracy is high and can meet the needs of actual applications.

[0096] In order to improve the generalization ability of the model and prevent the model from overfitting, a custom regularization function is designed. L1 and L2 regularization are combined, and different smoothing functions are used for different layer thicknesses (when the thickness is less than 2m, the regularization is strong; when the thickness is between 4m and 5m, the regularization is reduced; when the thickness is greater than 5m, the regularization is minimal).

[0097] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A reservoir fluid property identification method based on well logging data, characterized in that: The following steps are involved: S1, collecting mud logging gas measurement data and well logging data, and performing depth matching on the mud logging gas measurement data and the well logging data as a depth data set; S2, generate explanation conclusion labels for deep datasets; S3, extracting depth segment trend parameters using interpretation conclusion labels; S4, extracting a training set and a test set from the depth data set using the depth segment trend parameter; S5, using the training set and the test set to train and verify the reservoir fluid property prediction model in sequence; S6. Use the verified reservoir fluid property prediction model to complete the reservoir fluid property identification.

2. The reservoir fluid property identification method based on logging data according to claim 1, characterized in that: In S1, the calculation formula for depth matching is: d = round(y) + 1; In the formula, c is the top depth after treatment, x is the original top depth, d is the bottom depth after treatment, y is the original bottom depth, and round(·) is used to round y. Indicates rounding down.

3. The reservoir fluid property identification method based on logging data according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31. Extract the mean, variation amplitude, peak value, standard deviation and slope of the depth data set using the interpretation conclusion label; S32, calculating a comprehensive distance metric for the depth dataset; S33, calculating the interaction parameter of the depth data set according to the comprehensive distance metric of the depth data set; S34, calculating the stability parameter of the depth data set; S35. The mean, variation amplitude, peak value, standard deviation, slope, comprehensive distance metric, interaction parameter and stability parameter of the depth data set are used as depth segment trend parameters.

4. The reservoir fluid property identification method based on well logging data according to claim 3 is characterized in that: In S32, the calculation formula of the comprehensive distance metric D of the depth data set is: D=D depth (x i ,y i )+D G+R (x i ,y i ); In the formula, x i Represents the i-th parameter in the depth dataset, y i Indicates the i-th parameter in the data sequence to be matched, D depth (·) represents the depth parameter distance, D G+R (·) indicates the distance converted from similarity.

5. The reservoir fluid property identification method based on logging data according to claim 3 is characterized in that: In S33, the calculation formula of the interaction parameter z of the depth data set is: z=log(1+|TG|·D 2 -exp(-D)·sin(TG)); Where log(·) represents the logarithmic function, TG represents the total gas measurement volume, exp(·) represents the exponential function, and D represents the comprehensive distance metric of the depth dataset.

6. The reservoir fluid property identification method based on logging data according to claim 3 is characterized in that: In S34, the depth data set contains 2 to 5 data points of stability parameter SI 2～5 The calculation formula is: In the formula, ω i represents the weight of the i-th parameter in the depth dataset, x i represents the i-th parameter in the depth dataset, Mean represents the mean of the depth dataset, n represents the number of parameters in the depth dataset, Stand deviation represents the standard deviation of the depth dataset, and II(·) represents the indicator function; In S34, the depth data set contains stability parameters SI of more than 5 data points >5 The calculation formula is: In the formula, median represents the median of the depth data set, x1 represents the first parameter in the depth data set, x2 represents the second parameter in the depth data set, x3 represents the third parameter in the depth data set, and x n Represents the nth parameter in the depth dataset.

7. The reservoir fluid property identification method based on logging data according to claim 1, characterized in that: In S4, a data subset having pressure sampling or formation test conclusions is extracted from the depth data set using the depth segment trend parameter, and the data subset is split into a test set and a training set.

8. The reservoir fluid property identification method based on well logging data according to claim 1, characterized in that: In S5, the regularization function Ω(f k ) is: In the formula, f k represents the model of the kth tree, γ represents the first regularization parameter, T represents the number of leaf nodes in the tree, j represents the leaf node index, β represents the adjustment weight, λ represents the second regularization parameter, ω j represents the weight of model learning, α represents the third conventional regularization parameter, and h j represents the thickness characteristic, and exp(·) represents the exponential function.

Images