A data-driven rock mechanics logging prediction method based on uncertainty analysis

Abstract

The present invention relates to a data-driven rock mechanics logging prediction method based on uncertainty analysis, comprising: calculating the coefficient of variation of different depth intervals of a logging curve; determining a sample interval based on the coefficient of variation; obtaining a probability density function of each sample interval; simulating the probability density function of each sample interval by a Monte Carlo simulation method to obtain simulated data that conforms to the probability density function of each group of sample intervals; inputting the simulated data as a data set into a prediction model based on data driving to obtain the probability density function, mean and variance of the rock mechanics parameters corresponding to each group of sample intervals; setting a confidence interval, regarding the results within the confidence level interval as credible prediction results, and regarding the mean of the fitting results as the best estimate, to obtain a rock mechanics parameter profile with credibility.

Description

Technical Field

[0001] The present invention relates to the technical field of rock mechanics parameter prediction, and in particular to a data-driven rock mechanics logging prediction method based on uncertainty analysis. Background Art

[0002] Reservoir rock mechanics parameters are key parameters for solving petroleum geological problems and are the bridge connecting geological engineering. It is very important to accurately evaluate reservoir rock mechanics parameters. As a complex natural geological body, the rock in the formation itself is a highly complex uncertain system. The rock mechanics parameters obtained based on relevant data such as logging data and indoor experimental results are uncertain, so the true value of the parameter will be within a distribution function area. However, in most calculation models, the results obtained are all definite values, and the uncertainty of the results cannot be analyzed, that is, the prediction results do not consider the influence of uncertainty, which makes the final prediction error larger. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a data-driven rock mechanics logging prediction method based on uncertainty analysis.

[0004] The present invention is achieved through the following technical solutions:

[0005] A data-driven rock mechanics logging prediction method based on uncertainty analysis includes the following steps:

[0006] Calculate the coefficient of variation of the well logging curve at different depth intervals;

[0007] Based on the coefficient of variation, determine the sample interval;

[0008] Get the probability density function of each sample interval;

[0009] The Monte Carlo simulation method is used to simulate the probability density function of each sample interval to obtain simulated data that conforms to the probability density function of each group of sample intervals;

[0010] The simulation data is input as a data set into a data-driven prediction model, and the prediction model outputs corresponding rock mechanics parameters;

[0011] Based on the prediction results output by the prediction model, the probability density function, mean value and variance of the rock mechanical parameters corresponding to each group of sample intervals are obtained;

[0012] A confidence interval is set, and the results within the confidence level interval are regarded as credible prediction results. At the same time, the mean of the prediction results is regarded as the best estimate, and a rock mechanics parameter profile with credibility is obtained.

[0013] Among them, the coefficient of variation of different depth intervals is calculated using the following formula:

[0014]

[0015] In the above formula, γ is the coefficient of variation, ΔH is the sample interval, N is the total number of logging curve samples, and H is i is the depth of a certain point, P(H i ) is the logging value corresponding to the depth point.

[0016] Preferably, ΔH=0.5.

[0017] Optionally, the method for obtaining the probability density function of each sample interval is as follows: after selecting the sample interval range, the logging curve is divided according to the sample interval, and is moved sequentially according to the sampling interval to obtain a data set of each center point; assuming that the distribution of each data set approaches a normal distribution, the mean and variance of each center point data set are obtained to obtain the probability density function of the corresponding sample interval.

[0018] The logging curve includes corresponding logging curve parameters after feature screening or parameters obtained from changes in the logging curve.

[0019] Optionally, the 5% and 95% values ​​are regarded as the lower confidence limit and the upper confidence limit, respectively, and the values ​​in between are regarded as the 90% confidence level interval.

[0020] Optionally, the method for constructing the data-driven prediction model includes the following steps:

[0021] Obtaining raw data, including the indoor measured values ​​of rock mechanical parameters and the parameter values ​​of the corresponding logging curves;

[0022] Preprocess the raw data, including data cleaning and feature screening:

[0023] Segment the preprocessed data;

[0024] The segmented data is divided into a training set and a test set by using the K-fold cross-validation method;

[0025] An algorithm model is selected, and the algorithm model is trained using a training set, with the input being the preprocessed logging parameter value and the output being the predicted value of the rock mechanics parameter; the algorithm model is evaluated and optimized using a test set, and the optimal prediction model obtained is used as the data-driven prediction model.

[0026] Optionally, the algorithm model is one of RF, GPR, and Auto-sklearn models.

[0027] Preferably, the rock mechanics parameters include tensile strength, compressive strength, elastic modulus, Poisson's ratio, and brittleness index.

[0028] Compared with the prior art, this application has at least the following beneficial effects:

[0029] Due to the existence of uncertainty, the true value of rock mechanics parameters will be distributed in a certain area. The prediction result of this application is not a single value, but a corresponding confidence interval. The values ​​within this interval can be regarded as valid prediction values, making the prediction result more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 It is a flow chart of the data-driven rock mechanics logging prediction method based on uncertainty analysis in the embodiment;

[0032] Figure 2 is the tensile strength predicted by the data-driven prediction model in the embodiment;

[0033] Figure 3 is the compressive strength predicted by the data-driven prediction model in the embodiment;

[0034] Figure 4 is the elastic modulus predicted by the data-driven prediction model in the embodiment;

[0035] Figure 5 is the Poisson's ratio predicted by the data-driven prediction model in the embodiment;

[0036] Figure 6 is a brittleness index predicted by a data-driven prediction model in an embodiment;

[0037] Figure 7 Comparison chart of correlation coefficients of rock mechanical parameters predicted by different models;

[0038] Figure 8 Comparison chart of average relative errors of rock mechanical parameters predicted by different models;

[0039] Fig. 9 This is the predicted profile of a well in the study area in the example. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0041] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other.

[0042] like Figure 1 As shown, the present embodiment discloses a data-driven rock mechanics logging prediction method based on uncertainty analysis, comprising the following steps:

[0043] S1. Select the well logging curve of the target formation in the study area and calculate its coefficient of variation at different depth intervals, as shown in formula (1).

[0044]

[0045] In the above formula, γ is the coefficient of variation, ΔH is the sample interval, N is the total number of logging curve samples, and H is i is the depth of a certain point, P(H i ) is the logging value corresponding to the depth point.

[0046] S2, based on the coefficient of variation of the logging data, determine the value of the sample interval ΔH.

[0047] The calculation results of the coefficient of variation of different depth intervals are shown in Table 1. It can be seen that as the sample interval increases, its coefficient of variation also gradually increases, indicating that the greater the degree of discreteness between the data, the greater the difference in its characteristics. Comprehensively comparing the coefficient of variation of different depth intervals, as the sample interval increases, the rate of change of its coefficient of variation gradually tends to be stable and close to a constant, which is called the sill value. When the coefficient of variation value exceeds the sill value, it is considered that spatial correlation does not exist. When the sample interval changes from 0.875 to 1, its rate of change has become stable. When the sample interval is 0.5m, its γ(i) / γ(1) is less than 3, and the rate of change is significantly larger than that of the subsequent intervals, indicating that there is spatial correlation in this distance, and the data points in this interval are more than those in the previous sample intervals, which can reflect the data distribution characteristics of this sample interval. Therefore, ΔH=0.5 is selected as the sample interval of the study area, and the probability density function in this interval is the probability density function of the corresponding interval center depth.

[0048] Table 1: Variation of coefficient of variation with sample interval

[0049]

[0050] S3. Obtain the probability density function of each sample interval.

[0051] After selecting the sample interval range, the logging curve parameters are divided according to the sample interval and moved in sequence according to the sampling interval to obtain the data set of each center point.

[0052] Assuming that the distribution of each data set is close to the normal distribution, therefore, by obtaining the mean and variance of each center point data set, we can get the probability density function of the corresponding sample interval.

[0053] It is worth noting that the logging curve parameters in this step include the corresponding logging curve parameters after feature screening or the parameters obtained from the logging curve changes, such as depth, effective stress, natural gamma, neutron, acoustic wave time difference, density, deep resistivity, etc.

[0054] S4, using the Monte Carlo simulation method to simulate the probability density function of each sample interval, to obtain simulated data that conforms to the probability density function of each group of sample intervals, and the simulated data conforms to the probability distribution form of the corresponding sample interval. The simulated data is a new data set containing uncertainty.

[0055] It is worth noting that the number of Monte Carlo simulations is reasonably set according to needs.

[0056] In some embodiments, the number of Monte Carlo simulations is set to 100 times.

[0057] S5. The simulation data is input as a data set into a data-driven prediction model. The prediction model is a data-driven model with the best prediction effect. The output result of the prediction model is the corresponding rock mechanics parameters, such as tensile strength, compressive strength, elastic modulus, Poisson's ratio, and brittleness index.

[0058] Based on the prediction results output by the prediction model, the probability density function, mean and variance of the rock mechanical parameters corresponding to each group of sample intervals are obtained.

[0059] In some embodiments, the number of Monte Carlo simulations is set to 100 times, so each group of sample intervals has 100 sets of simulated data. The 100 sets of simulated data in each group of sample intervals are input as data sets into the data-driven prediction model. In a rock mechanics parameter, 100 prediction results can be obtained in the same sample interval. Since the distribution of the simulated data sets in the previous text is normal distribution, the corresponding probability density function is determined by the mean and variance. Therefore, the average value of the 100 prediction results is regarded as the theoretical normal distribution mean of each group of sample intervals, and its variance is regarded as the theoretical normal distribution variance of each group of sample intervals, so that the probability density function, mean and variance of the rock mechanics parameter corresponding to each group of sample intervals can be obtained.

[0060] S6, setting the confidence interval, considering the results within the confidence level interval as credible prediction results, and considering the mean of the corresponding rock mechanics parameter prediction results as the best estimate, to obtain a rock mechanics parameter profile with credibility.

[0061] Optionally, the 5% and 95% values ​​are regarded as the lower confidence limit and the upper confidence limit respectively, and the values ​​in between are regarded as the 90% confidence level interval, and the results in the confidence level interval are considered to be credible prediction results.

[0062] The rock mechanics parameters predicted in this embodiment are tensile strength, compressive strength, elastic modulus, Poisson's ratio, and brittleness index.

[0063] Correlation coefficient R 2 , root mean square error RMSE, mean relative error MRE, mean absolute error MAE to evaluate the prediction results, and the correlation coefficient R 2 The higher the value, the smaller the root mean square error (RMSE), mean relative error (MRE), and mean absolute error (MAE), indicating better prediction results. The evaluation results of the five rock mechanical parameters, namely tensile strength, compressive strength, elastic modulus, Poisson’s ratio, and brittleness index, are shown in Table 2. The prediction results of the corresponding rock mechanical parameters are shown in Figure 2-Figure 6 shown.

[0064] Table 2: Results of data-driven rock mechanics assessment based on uncertainty analysis

[0065]

[0066] Among them, the correlation coefficient R 2 The calculation formulas for the root mean square error RMSE, the mean relative error MRE, and the mean absolute error MAE are shown in formulas (2) to (5).

[0067]

[0068]

[0069]

[0070]

[0071] In the above formula, y i is the measured value of the corresponding rock mechanical parameters; is the predicted value of the corresponding rock mechanical parameters; n is the number of tests.

[0072] Among them, tensile strength refers to the ability of rock to withstand tensile force, which indicates the maximum bearing capacity of rock under tensile conditions; compressive strength refers to the ability of rock to withstand compressive force, which is the ultimate strength of rock failure; elastic modulus is a parameter to measure the sensitivity of rock to stress changes. For formation fracturing, a higher rock elastic modulus will reduce the degree of rock deformation, making the rock more difficult to fracture; rock Poisson's ratio also reflects the deformation characteristics of rock when subjected to stress, indicating the degree of lateral deformation of rock; brittleness index is used to indicate the destructive properties of rock during stress. Generally, the minerals of rock affect brittleness. When the formation is fractured, a high brittleness index makes it easier to form cracks, and the size of the brittleness index affects the complexity of crack formation.

[0073] In order to verify the prediction effect of the improved model, the prediction effects of RF, GPR, and Auto-sklearn models were compared. The comparison results are as follows: Figure 7 , Figure 8 Compared with the prediction of five rock parameters, namely tensile strength, compressive strength, elastic modulus, Poisson's ratio and brittleness index, the improved prediction model based on uncertainty analysis achieved excellent results in the prediction of compressive strength, elastic model and brittleness index, proving that the model has a strong performance.

[0074] This method was applied to a well in the study area. The results are as follows Fig. 9 shown.

[0075] Among them, Figure 1 As shown, the method for constructing a data-driven prediction model includes the following steps:

[0076] A. Obtaining original data, including the indoor measured values ​​of five rock mechanical parameters and the parameter values ​​of the corresponding logging curves;

[0077] B. Preprocess the raw data: including data cleaning and feature screening;

[0078] C. Segment the preprocessed data;

[0079] D. Use the K-fold cross-validation method to divide the segmented data into a training set and a test set;

[0080] E. Select an algorithm model, such as RF, GPR, or Auto-sklearn model. The input is the preprocessed logging parameter value, and the output is tensile strength, compressive strength, elastic modulus, Poisson's ratio, and brittleness index. Use the training set to train the algorithm model, and use the test set to evaluate and optimize the algorithm model to obtain the optimal prediction model.

[0081] Due to the existence of uncertainty, the true value of rock mechanics parameters will be distributed in a certain area. The prediction result of this application is not a single value, but a corresponding confidence interval. The values ​​within this interval can be regarded as valid prediction values, making the prediction result more reliable.

[0082] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data-driven rock mechanics logging prediction method based on uncertainty analysis, characterized in that: The following steps are involved: Calculate the coefficient of variation of the well logging curve at different depth intervals; Based on the coefficient of variation, determine the sample interval; Get the probability density function of each sample interval; The Monte Carlo simulation method is used to simulate the probability density function of each sample interval to obtain simulated data that conforms to the probability density function of each group of sample intervals; The simulation data is input as a data set into a data-driven prediction model, and the prediction model outputs corresponding rock mechanics parameters; Based on the prediction results output by the prediction model, the probability density function, mean and variance of the rock mechanical parameters corresponding to each group of sample intervals are obtained; wherein, the average value of the prediction results is regarded as the theoretical normal distribution mean of each group of sample intervals, and its variance is regarded as the theoretical normal distribution variance of each group of sample intervals, so that the probability density function, mean and variance of the rock mechanical parameters corresponding to each group of sample intervals can be obtained; Set the confidence interval, regard the results within the confidence level interval as credible prediction results, and regard the mean of the prediction results as the best estimate to obtain a rock mechanics parameter profile with credibility; The coefficient of variation for different depth intervals was calculated using the following formula: In the above formula, γ is the coefficient of variation, ΔH is the sample interval, N is the total number of logging curve samples, and H is i is the depth of a certain point, P(H i ) is the logging value corresponding to the depth point.

2. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 1 is characterized in that: ΔH=0.

5.

3. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 1 is characterized in that: The method to obtain the probability density function of each sample interval is: After selecting the sample interval range, the logging curve is divided according to the sample interval and moved in sequence according to the sampling interval to obtain the data set of each center point; assuming that the distribution of each data set approaches the normal distribution, the mean and variance of each center point data set are obtained to obtain the probability density function of the corresponding sample interval.

4. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 3 is characterized in that: The logging curve includes corresponding logging curve parameters after feature screening or parameters obtained from the change of the logging curve.

5. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 1 is characterized in that: The 5% and 95% values ​​are regarded as the lower confidence limit and upper confidence limit respectively, and the values ​​in between are regarded as the 90% confidence level interval.

6. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 1 is characterized in that: The method for constructing the data-driven prediction model comprises the following steps: Obtaining raw data, including the indoor measured values ​​of rock mechanical parameters and the parameter values ​​of the corresponding logging curves; Preprocess the raw data, including data cleaning and feature screening: Segment the preprocessed data; The segmented data is divided into a training set and a test set by using the K-fold cross-validation method; An algorithm model is selected, and the algorithm model is trained using a training set, with the input being the preprocessed logging parameter value and the output being the predicted value of the rock mechanics parameter; the algorithm model is evaluated and optimized using a test set, and the optimal prediction model obtained is used as the data-driven prediction model.

7. The data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 6 is characterized in that: The algorithm model is one of the RF, GPR, and Auto-sklearn models.

8. A data-driven rock mechanics logging prediction method based on uncertainty analysis according to claim 1, 6 or 7, characterized in that: The rock mechanics parameters include tensile strength, compressive strength, elastic modulus, Poisson's ratio, and brittleness index.

Images