A machine learning-based method and system for calculating formation pressure with multi-source data coupling

Abstract

The present invention relates to a method for calculating formation pressure by coupling multi-source data based on machine learning, comprising the following steps: S1 Collect logging data and mud logging data of multiple wells; S2 Simulate the actual formation pressure based on the Dc index, drilling efficiency - mechanical energy ratio monitoring, gas logging auxiliary judgment, measured pressure data, etc.; S3 Establish a univariate fitting equation between the formation pressure and the logging data; S4 Establish a bivariate fitting equation between the formation pressure and the logging - mud logging parameters; S5 Establish a formation pressure calculation model based on the logging - mud logging bivariate coupling data; S6 Calculate the formation pressure based on the formation pressure calculation model of the logging - mud logging bivariate coupling data; S7 Correct the formation pressure obtained in S6; S8 Establish a thermal map of the logging - mud logging parameters of each well; S9 Based on the results shown in the thermal map, eliminate variables using the correlation principle to obtain the final logging - mud logging parameter data set; S10 Establish a formation pressure calculation model based on the coupling of multi-source logging and mud logging data.

Description

Technical Field

[0001] The present invention belongs to the field of rock mechanics, and particularly relates to a method for calculating formation pressure by coupling multi-source data based on machine learning. Background Art

[0002] The basement of the Qiongdongnan Basin is mainly Indosinian and Yanshanian granite buried hills (age: 65 Ma). The formation and evolution of the granite buried hills in the Qiongdongnan Basin have been affected by multi-stage tectonic movements in the Indosinian, Yanshanian, and Himalayan periods, providing good conditions for the formation of weathered crusts and internal fractured reservoirs in the buried hills. The Qiongdongnan buried hills are divided into three buried hill belts: the southern buried hill belt, the central buried hill belt, and the northern buried hill belt. Generally, from the southern buried hill belt to the northern buried hill belt, the burial depth, burial time, and current seawater depth are all decreasing, and the development position of the buried hills is rising.

[0003] Currently, the research on overpressure in sandstone and mudstone formations at home and abroad has gradually become mature. In recent years, formation pressure monitoring has been carried out in the Yingqiong Basin, and there are preliminary means for monitoring formation pressure based on the causes of pressure. The monitoring accuracy of formation pressure is relatively high in formations with conventional lithology and common pressure causes (such as undercompaction, fluid expansion, and pressure conduction). However, in recent years, the bedrock buried hills have gradually become one of the hot areas of exploration and development. The evaluation techniques for the formation pressure of special lithology in the bedrock buried hills of the Qiongdongnan Basin are relatively limited, seriously restricting the evaluation accuracy of formation pressure.

[0004] Currently, the evaluation of overpressure in buried hill formations mainly uses electrical logging acoustic waves and mud logging Dc index. However, these two parameters cannot accurately indicate the overpressure in buried hill formations. The main reasons are as follows: 1) The framework of the buried hill formation is dense, and the acoustic wave velocity mainly depends on the rock framework and the degree of fracture development. The overpressure of pore fluid has little effect on the acoustic wave velocity, and it is impossible to effectively indicate overpressure using electrical logging acoustic waves; 2) The mud logging Dc and Sigma indices are mainly used for on-site quantitative monitoring of overpressure caused by undercompaction of sedimentary rocks. However, there are some problems in finding the existence of overpressure that can be indicated in bedrock formations, that is, the mechanism of change of mud logging parameters is not clear, and the accuracy of the quantitative evaluation results is relatively poor. Therefore, the present invention proposes a method for calculating formation pressure by coupling multi-source data based on machine learning, comprehensively considering the interaction between the drill bit and the buried hill formation, and optimizing or constructing a mud logging data volume that can effectively characterize the overpressure in the buried hill formation, providing a theoretical basis for the accurate evaluation of formation pressure in the Yingqiong Basin and buried hill formations. Summary of the Invention

[0005] A method for calculating formation pressure by coupling multi-source data based on machine learning, characterized by comprising the following steps:

[0006] S1: Collect the logging and mud logging data of multiple wells. The logging parameters mainly include acoustic travel time and density, and the mud logging parameters mainly include GR, resistivity, drilling rate, drill string rotation speed, MSE, Dc index, lateral bit work, longitudinal bit work, equivalent circulating density (ECD) of drilling fluid, and equivalent density of overlying formation pressure (OBG).

[0007] S2: Based on the Dc index, monitoring of drilling efficiency - mechanical specific energy, gas logging assisted judgment, measured formation pressure data, etc., simulate the actual formation pressure.

[0008] S3: Plot the cross - plot between logging data and formation pressure values, and use the method of unary linear regression to obtain the unary fitting equation between formation pressure values and logging data.

[0009] S4: Plot the cross - plot between logging data, mud logging data and formation pressure values, and use the method of binary linear regression to obtain the binary fitting equation between formation pressure values and logging - mud logging parameters.

[0010] S5: Integrate the binary fitting equation between formation pressure values and logging - mud logging parameters to establish a formation pressure calculation model based on logging - mud logging binary coupled data.

[0011] S6: Based on the formation pressure calculation model of logging - mud logging binary coupled data, calculate the formation pressure, including PP→DEN&Dc, PP→DEN&MSE, PP→VP&Dc, PP→VP&MSE.

[0012] S7: For the new formation pressure obtained in S6, use the program for comprehensive correction to obtain PP - correction.

[0013] S8: Based on the collected logging - mud logging data and PP - correction data, establish a heat map of logging - mud logging parameters for each well to obtain the correlation between logging - mud logging parameters and formation pressure.

[0014] S9: Based on the results shown in the heat map, use code to delete one of the two parameters with strong correlation to obtain the data set of logging - mud logging parameters after elimination.

[0015] S10: Based on the data set obtained after elimination, use machine learning algorithms for data training to establish a formation pressure calculation model based on the coupling of multiple logging - mud logging data. The machine learning algorithms are random forest method, multiple linear regression method, decision tree method, and support vector machine method.

[0016] In S5, it is characterized in that the model of the new formation pressure calculation method based on the coupling of logging - mud logging binary data is:

[0017] P p= m × VP + n × logging parameters (Dc, SIGA, MSE, longitudinal work, lateral work) + O

[0018] P p = j × DEN + k × logging parameters (Dc, SIGA, MSE, longitudinal work, lateral work) + L

[0019] Where: m, n, O, j, k, and L are coefficients and need to be determined by block.

[0020] In the said S6, the PP→DEN&Dc refers to the new formation pressure calculated by using the logging density and the Dc index of logging parameters and the formation pressure calculation model of logging-logging binary coupling data; the PP→DEN&MSE refers to the new formation pressure calculated by using the logging density and the MSE of logging parameters and the formation pressure calculation model of logging-logging binary coupling data; the PP→VP&Dc refers to the new formation pressure calculated by using the acoustic travel time of logging parameters and the Dc index of logging parameters and the formation pressure calculation model of logging-logging binary coupling data; the PP→VP&MSE refers to the new formation pressure calculated by using the acoustic travel time of logging parameters and the MSE of logging parameters and the formation pressure calculation model of logging-logging binary coupling data.

[0021] In the said S7, first identify and delete outliers, and then check the data set to delete the rows with null values and NaN values by using the detect outliers method. Then synthesize several new formation pressure results to obtain PP - correction.

[0022] In the said S8, the results shown in the heat map are mainly distinguished by colors to indicate the degree of correlation. The darker the color, the greater the correlation coefficient.

[0023] In the said S9, use code to delete the attributes with a correlation (absolute value) above 0.9.

[0024] In the said S10, synthesize the results obtained from the heat map and the machine learning algorithm, and select the model with the best effect (the highest prediction correlation coefficient).

[0025] The present invention thus provides a system for calculating formation pressure by coupling multi-source logging and well logging data based on machine learning, which is characterized in that it includes modules for executing steps S2 to S10 in the method according to any one of claims 1 to 8. Specifically, it includes a data entry module for entering data from step S1; a data processing module for performing data calculations in steps S2 to S10 (specifically, including a binary data fitting module for performing data calculations in steps S2 to S6; a pp-correction module for performing data correction and elimination of strong correlations in steps S7 to S9; a multi-source data coupling module for outputting the data coupling model obtained in step S10); and a result output module for outputting the results obtained by the data processing module. The above can be implemented by obtaining relevant computer programs through computational language programming.

[0026] Preferably, the computer program of the multi-source data coupling formation pressure calculation module based on the random forest method in the multi-source data coupling module is as follows:

[0027]

[0028] Optionally, it further includes a result display module, such as displaying through a screen or a remote terminal, etc.

[0029] Compared with the prior art, the present invention has the following advantages: The present invention discloses a method for calculating formation pressure by coupling multi-source data based on machine learning, which makes full use of logging and well logging data. By establishing a formation pressure calculation model based on the coupling of logging-well logging multi-source data through a multi-source data fitting module, it uses logging data to evaluate and eliminate the influence of fracture development degree on rock-breaking efficiency. Further, according to the analysis of well logging data, the interaction between the drill bit and the buried hill formation is comprehensively considered, and a well logging data volume that can effectively characterize the overpressure of the buried hill formation is constructed. The method for calculating formation pressure by coupling multi-source data based on machine learning disclosed by the present invention, through the pp-correction module, comprehensively considers the results of heat map analysis, eliminates the influence of strongly correlated parameters, simplifies the data model, and has higher applicability. The method of the present invention comprehensively considers the interaction between the drill bit and the buried hill formation, establishes a new method for monitoring the formation pressure of the buried hill, and can realize the quantitative evaluation of the overpressure of the buried hill. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Flowchart of the method for calculating formation pressure by coupling multi-source logging and well logging data based on machine learning.

[0031] Figures 2-1 to 2-3 The logging and well logging data collected in step 1.

[0032] Figure 3 Crossplot of logging data and formation pressure.

[0033] Figure 4Crossplot of logging data and formation pressure.

[0034] Figures 5-1 to 5-3 Formation pressure calculated by the formation pressure calculation model based on the well logging - mud logging binary coupled data.

[0035] Figure 6 Data processing process in Step 7.

[0036] Figures 7-1 to 7-3 PP - correction result in Step 7.

[0037] Figure 8 Thermodynamic diagram of the correlation between well logging and mud logging data.

[0038] Figure 9 Bar chart of attribute correlation (without absolute value).

[0039] Figure 10 Prediction result of the random forest method.

[0040] Figure 11 Prediction result of the multiple linear regression method.

[0041] Figure 12 Prediction result of the decision tree method.

[0042] Figure 13 Prediction result of the support vector machine method. Detailed implementation mode

[0043] The following combines the drawings and cases to make a more specific description of the multi - data coupled formation pressure calculation method based on machine learning of the present invention, where the flow chart of the formation pressure calculation method is as Figure 1 shown. The following uses an example to elaborate the specific implementation process.

[0044] Step 1, select a buried - hill formation overpressure well A, and query the corresponding well logging parameters VP, DEN and mud logging parameters GR, resistivity, drilling rate, drill string rotation speed, MSE, Dc index, lateral bit work, longitudinal bit work, ECD, OBG. The specific collection situation is as Figures 2-1 to 2-3 .

[0045] Step 2, simulate the actual formation pressure based on the Dc index, drilling efficiency - mechanical specific energy monitoring, gas logging auxiliary judgment, measured formation pressure data, etc. The specific steps are as follows:

[0046] 1) For the same well A, select the well logging and mud logging data used for calculating the formation pressure;

[0047] 2) Calculate the formation pressure at different depths based on the Dc index and drilling efficiency - mechanical specific energy to monitor the formation pressure;

[0048] 3) Assist in judging the formation pressure based on the gas logging data;

[0049] 4) Obtain the formation pressure at the test point through on-site actual testing;

[0050] Through the mutual calibration of the formation pressure results obtained by the Dc index, drilling efficiency - mechanical energy ratio monitoring, gas logging auxiliary judgment, and measured pressure data, the final simulated actual formation pressure is obtained.

[0051] Step 3: Draw a cross-plot between the logging data and the formation pressure value, and use the method of unary linear regression to obtain a unary fitting equation between the formation pressure value and the logging data. The specific steps are as follows:

[0052] 1) Based on the logging data collected in Step 1 and the formation pressure data calculated in Step 2, draw a cross-plot of the logging data and the formation pressure of Well A, as Figure 3 .

[0053] 2) Use the method of unary linear regression to obtain a binary fitting equation between the formation pressure value and the logging data.

[0054] Step 4: Draw a cross-plot between the logging parameters, mud logging data and the formation pressure value, and use the method of binary linear regression to obtain a binary fitting equation between the formation pressure value and the logging and mud logging parameters. The specific steps are as follows:

[0055] 1) Based on the mud logging data collected in Step 1 and the formation pressure data calculated in Step 2, draw a cross-plot of the mud logging data and the formation pressure of Well A, as Figure 4 .

[0056] 2) Use the method of binary linear regression to obtain a binary fitting equation between the formation pressure value and the logging and mud logging parameters, as shown in the following table:

[0057] Table 1 Binary fitting equation between the formation pressure value of Well A and the logging and mud logging parameters

[0058]

[0059] Step 5: Integrate the binary fitting equation between the formation pressure value and the logging - mud logging parameters to establish a formation pressure calculation model based on the binary coupled data of logging - mud logging.

[0060] P p = m×VP + n×mud logging parameters (Dc, SIGA, MSE, longitudinal work, transverse work) + O

[0061] P p = j×DEN + k×mud logging parameters (Dc, SIGA, MSE, longitudinal attack, transverse attack) + L Step 6: Based on the formation pressure calculation model of the binary coupled data of logging - mud logging, calculate the formation pressure ( Figures 5-1 to 5-3), including PP→DEN&Dc, PP→DEN&MSE, PP→VP&Dc, PP→VP&MSE. The specific steps are as follows:

[0062] 1) For the same well A, substitute the corresponding logging data collected in step 1 into the binary fitting equation obtained in step 4.

[0063] 2) Calculate the new formation pressure: PP→DEN&Dc, PP→DEN&MSE, PP→VP&Dc, PP→VP&MSE, as shown in Figure 5.

[0064] Step 7, for the new formation pressure obtained in S6, use the program to perform comprehensive correction to obtain PP-corrected, as Figure 6 shown. The specific steps are as follows:

[0065] 1) Identify and delete outliers, and then check the data set and use the detect outliers method to delete the rows with null values and NaN values, as Figure 6 .

[0066] 2) Then synthesize several new formation pressure results to obtain PP-corrected, see Figures 7-1 to 7-3 .

[0067] Step 8, based on the collected logging data and PP-corrected data, establish a thermal map of logging parameters for each well. The specific steps are as follows:

[0068] 1) For the same well A, read its logging data and delete the empty columns that are all NaN values.

[0069] 2) Perform attribute analysis on the data after deleting outliers, and draw a thermal map and a bar chart of attribute correlations.

[0070] Step 9, based on the results shown in the thermal map, use the code to delete one of the two parameters with strong correlation to obtain the data set of logging parameters after elimination. The specific steps are as follows:

[0071] 1) It can be seen from the thermal map of logging data correlations ( Figure 8 ) that for well A, the parameters "ROP and ECD" (correlation coefficient is -0.94), "ROP and OBG" (correlation coefficient is -0.96), and "ECD and OBG" (correlation coefficient is 0.95) have strong correlations, and only one of them can be selected in the subsequent parameter selection.

[0072] 2) Use the code to delete the parameters with strong correlation attributes (the absolute value of the correlation coefficient is above 0.9) ( Figure 9 ).

[0073] Step 10: Based on the dataset obtained after elimination, data training is carried out using four machine learning algorithms. The results are as Figure 10 shown in the prediction results of the random forest method. Figure 11 shown in the prediction results of the multiple linear regression method. Figure 12 shown in the prediction results of the decision tree method. Figure 13 shown in the prediction results of the support vector machine method. By comparing the correlation coefficients of each prediction result, the machine learning algorithm with the best effect is selected as the random forest method (correlation coefficient is 0.781). The finally established calculation method for the multi-data coupled formation pressure of Well A is the multi-data coupled formation pressure calculation based on the random forest method. The calculation program in its multi-data coupling module is as follows:

[0074]

Claims

1. A method for calculating formation pressure by coupling multi-source logging data based on machine learning, characterized in that, It includes the following steps: S1: Collect logging-while-drilling data from multiple wells. The logging parameters include acoustic travel time VP and density DEN, and the logging-while-drilling parameters include gamma ray logging GR, resistivity RT, drilling rate, drill string rotation speed, mechanical specific energy MSE, Dc index, bit lateral work, bit vertical work, equivalent circulating density ECD, and overburden pressure OBG; S2: Based on the Dc index, drilling efficiency - mechanical specific energy monitoring, gas logging assisted judgment, and measured formation pressure data, simulate the actual formation pressure; S3: Draw a crossplot between the logging data and the formation pressure value, and use the method of unary linear regression to obtain a unary fitting equation between the formation pressure value and the logging data; S4: Draw a crossplot between the logging data, logging-while-drilling data, and the formation pressure value, and use the method of binary linear regression to obtain a binary fitting equation between the formation pressure value and the logging-logging-while-drilling parameters; S5: Integrate the binary fitting equation between the formation pressure value and the logging-logging-while-drilling parameters to establish a formation pressure calculation model based on the binary coupled data of logging and logging-while-drilling; S6: Formation pressure calculation model based on well logging - mud logging binary coupling data to calculate formation pressure, including formation pressure calculation P based on density logging - Dc exponent binary coupling P →DEN&Dc, the fitting equation is P P = 0.125DEN - 0.016Dc + 2.092, formation pressure calculation P based on density logging - mechanical specific energy binary coupling P →DEN&MSE, the fitting equation is P p = 0.151DEN + (4.6E - 08)MSE + 2.140, formation pressure calculation P based on acoustic travel time logging - Dc exponent binary coupling P →VP&Dc, the fitting equation is P P = 0.001VP - 0.006Dc + 1.681, formation pressure calculation P based on acoustic travel time logging - mechanical specific energy binary coupling P →VP&MSE, the fitting equation is P p = 0.001VP + (6.57E - 08)MSE + 1.656; Among them, DEN is the density in the logging parameters; Dc is the ratio of the corrected actual drilling rate to the theoretical drilling rate; S7: For the formation pressure obtained in S6, perform comprehensive correction using the program to obtain P P - Correction, and the process is as follows: 1) Identify and delete outliers, and then check the dataset to delete the rows with null values and NaN values using the detect outliers method; 2) Then, by synthesizing the new formation pressure results, P is obtained. P - Correction; S8: Based on the collected logging-while-drilling data and P P -corrected data, establish a thermal map of logging-while-drilling parameters for each well to obtain the correlation magnitude between logging-while-drilling parameters and formation pressure; S9: Based on the results shown in the heatmap, use code to delete one of the two parameters with strong correlation to obtain a dataset of logging and logging-while-drilling parameters after elimination; S10: Based on the dataset obtained after elimination, perform data training using a machine learning algorithm; establish a formation pressure calculation model based on the multi-data coupling of logging and logging-while-drilling; the machine learning algorithm is selected from the random forest method, multiple linear regression method, decision tree method, or support vector machine method.

2. The method according to claim 1, characterized in that, In the above S8, the results shown in the heatmap distinguish the correlation magnitude by color, and the darker the color, the greater the correlation coefficient.

3. The method according to claim 1, characterized in that, In the above S9, use code to delete the parameters with an absolute value of the correlation coefficient above 0.

9.

4. The method according to claim 1, wherein In the above S10, it is characterized in that the machine learning algorithm is the random forest method.

5. A system for calculating formation pressure by coupling multi-source logging data based on machine learning, characterized in that, It includes a module for executing steps S2 to S10 in the method described in any one of claims 1 to 4.

6. The system according to claim 5, wherein, It includes a data entry module for entering the data from step S1; A data processing module for performing the data calculations in steps S2 to S10; And a result output module for outputting the results obtained by the data processing module.

7. The system according to claim 6, characterized in that, The data processing module for performing the data calculations in steps S2 to S10 includes a binary data fitting module for performing the data calculations in steps S2 to S6; P P - A correction module for performing data correction and strong correlation elimination in steps S7 to S9; A multi-data coupling module for outputting the data coupling model obtained in step S10.

8. The system according to any one of claims 5 to 7, characterized in that, It also includes a result display module.

Images