A method and system for real-time identification of lithology during drilling

Abstract

The present invention provides a method and system for real-time identification of lithology during drilling, and belongs to the technical field of oil and gas exploration and development. The method comprises: (1) collecting data of neighboring wells; (2) constructing a feature set and a label set: constructing a feature set and a label set using data of neighboring wells; (3) training a basic model for lithology identification: training the basic model for lithology identification using the feature set and the label set; (4) correcting the basic model for lithology identification in real time; (5) real-time identification of lithology: using the corrected basic model for lithology identification to perform real-time identification of lithology and obtain the category of lithology. Based on the data of drilled neighboring wells, the present invention uses a random forest method to establish a basic model for lithology identification, and uses a probability model to correct the basic model, thereby avoiding the prediction error caused by differences in geological and engineering conditions in the deployment of the model in new wells, greatly improving the accuracy and timeliness of lithology identification, and providing a basis for increasing drilling speed, reducing drill bit wear, and increasing the life of downhole tools.

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas exploration and development, and in particular relates to a method and system for real-time identification of lithology during drilling. Background Art

[0002] During oil drilling, the adjustment of drilling parameters is mainly based on the judgment of downhole lithology. For formations with drastic lithology changes, timely adjustment of drilling parameters is of great significance for increasing drilling speed, reducing drill bit wear and increasing downhole tool life.

[0003] Chinese patent publication CN105888657B discloses a method for identifying the lithology of sedimentary rocks using element logging. It uses element logging measurement data as parameters, analyzes the chemical composition and diagenetic principles of various lithologies to establish a set of lithology identification models, and solves the problem of lithology identification while drilling in element logging. This method determines the lithology of rocks by analyzing the chemical composition of rock cuttings, but relies on the return of rock cuttings from the wellhead during drilling. The timeliness is reduced due to the delay time, which is particularly obvious in deep and ultra-deep wells. Chinese patent publication CN107489417A discloses a sandstone and mudstone lithology identification method based on XRF logging rock and mineral composition inversion method. Its main steps include: first, establishing the theoretical basis and interpretation principles of sandstone and mudstone rock and mineral composition inversion, then calculating the lithology characteristic quantity parameters, lithology characteristic quantity index and derived interpretation ratio parameters, establishing the evaluation criteria and parameter combination for sandstone and mudstone lithology identification, and establishing the sandstone and mudstone lithology identification interpretation evaluation model and process. The influence of subjective judgment factors is greatly eliminated, and rapid and quantitative lithology identification and evaluation are achieved. The Chinese patent publication CN111007064A discloses a logging lithology intelligent identification method based on image recognition, which includes the following steps: Step 1: Construct a mineral category intelligent identification method based on a neural network; Step 2: Apply the above intelligent identification method to identify mineral categories such as quartz, feldspar, and rock fragments; Step 3: Reinforce learning and updating of the intelligent identification neural network, and update the neural network method according to the diagnosis results; Step 4: Identify the edge of rock particles; Step 5: Use the neural network method to identify the mineral type inside each particle; Step 6: Name each particle according to the rock name naming method.

[0004] At present, there are five main methods for judging downhole lithology during drilling:

[0005] 1) Cable logging: After a certain section of the wellbore is completed, cable logging tools such as natural gamma and neutron porosity are lowered. After the electrical logging data is downloaded, the lithology is interpreted through professional software and algorithms. This method can only obtain downhole lithology results after the well section is completed, which is not timely and has little guiding significance for the actual drilling process;

[0006] 2) Logging while drilling: Installing a logging while drilling sub near the drill bit can detect downhole lithology and other information in real time. However, due to the distance between the measuring sub and the drill bit, it cannot reflect the lithology information at the drill bit, and the cost is high, making it difficult to promote and apply on a large scale.

[0007] 3) Geological logging intuitive judgment: The judgment of lithology at the drilling site is mainly based on the cuttings returned from the wellhead. However, since it takes a certain amount of time for the cuttings to move from the drill bit at the bottom of the well to the wellhead, and the delay time increases with the increase of well depth, this method has poor timeliness;

[0008] 4) Experience judgment: Comprehensively judge the downhole lithology based on logging parameters (drilling pressure, rotation speed, displacement, pump pressure, torque, drilling time), which requires high experience of engineers and is difficult to achieve quantitative evaluation.

[0009] 5) Data science methods: The relationship between logging parameters and lithology is established through fuzzy mathematics, multivariate regression or neural network methods. This type of method can achieve high accuracy within a certain application range, but it ignores the differences in engineering and geological environments between wells, and is prone to large deviations in the deployment of new well models. Summary of the invention

[0010] The purpose of the present invention is to solve the above-mentioned problems existing in the prior art, to provide a method and system for real-time identification of lithology during drilling, and to solve the problem of difficulty in real-time acquisition of downhole lithology information during drilling.

[0011] The present invention is achieved through the following technical solutions:

[0012] The first invention of the present invention provides a method for real-time identification of lithology during drilling, the method comprising:

[0013] (1) Collect data from adjacent wells;

[0014] (2) Constructing feature sets and label sets: Constructing feature sets and label sets using adjacent well data;

[0015] (3) Training the basic model for lithology identification: using the feature set and label set to train the basic model for lithology identification;

[0016] (4) Real-time correction of the basic model for lithology identification;

[0017] (5) Real-time identification of lithology: The lithology is identified in real time using the corrected lithology identification basic model to obtain the lithology category.

[0018] A further improvement of the present invention is that the operation of step (1) comprises:

[0019] Collect time series logging data and drilling data of adjacent wells to be drilled, including vertical depth, inclined depth, displacement, drilling pressure, rotation speed, hook load, inlet flow, drilling fluid density, mechanical penetration rate, rotary table torque, drill bit size and drill bit model.

[0020] A further improvement of the present invention is that the operation of step (2) comprises:

[0021] (21) Filter the data related to drilling conditions based on the inclination depth, displacement, rotation speed and drilling pressure in the logging data;

[0022] (22) calculating the torque variation coefficient according to the turntable torque, and adding the torque variation coefficient feature item to form the original feature set;

[0023] (23) Process the original feature set and construct a new feature set and label set;

[0024] (24) concatenating the new feature sets and label sets of all neighboring wells into feature sets and label sets respectively;

[0025] (25) The drill bit model data in the feature set and the lithology data in the label set are converted into numbers to obtain the final feature set and the final label set.

[0026] A further improvement of the present invention is that the operation of step (21) comprises:

[0027] Select the logging data that meets the following conditions from the logging data;

[0028] Q i >1L / s and MD i >MD i-1 And RPM i >1rpm and WOB i >1kN

[0029] Among them, Q i is the displacement corresponding to a certain point; MD i is the oblique depth corresponding to a certain point; MD i-1 The slope depth corresponding to the previous point of a certain point; RPM i is the speed corresponding to a certain point; WOB i is the drilling pressure corresponding to a certain point.

[0030] A further improvement of the present invention is that the operation of step (22) includes:

[0031] The step length W of the inclined depth is set, and W is used as the sliding window. The torque variation coefficient in each well section of each well is calculated using the following formula:

[0032]

[0033] The length of each well section is W meters;

[0034] Where V T is the torque variation coefficient of the rotary table torque in the well section; σ T is the RMS torque of the rotary table in the well section, in kN.m; is the average value of the rotary table torque in the well section, in kN.m;

[0035] The calculated torque variation coefficient is then assigned to each point in the well section;

[0036] Finally, a column of feature items is added to the feature set of each well to form the original feature set. The value of the feature item is the torque variation coefficient of each point in the well.

[0037] A further improvement of the present invention is that the operation of step (23) includes:

[0038] Re-interpolate the slant depth into an arithmetic sequence with a step size of S;

[0039] Take S as the window, take the average value of each feature item in the original feature set in each window to get the average value of each feature item, and use the average value as the value of the feature item in the window, fill it into the feature set to form a new feature set;

[0040] Use lithology as the label set.

[0041] A further improvement of the present invention is that the operation of step (24) includes:

[0042] The new feature sets and label sets of all adjacent wells are shuffled according to the oblique depth;

[0043] The feature sets of all neighboring wells after shuffling are concatenated into a feature set, and the label sets of all neighboring wells after shuffling are concatenated into a label set.

[0044] A further improvement of the present invention is that the operation of step (25) includes:

[0045] In the feature set obtained in step (24), the data of the drill bit model is One-Hot encoded, the data of the drill bit model is converted into the corresponding One-Hot code, and the data of the drill bit model in the feature set is replaced line by line with the corresponding One-Hot code to obtain the final feature set;

[0046] In the label set obtained in step (24), the lithology data is One-Hot encoded, the lithology data is converted into the corresponding One-Hot code, and the lithology data in the label set is replaced line by line with the corresponding One-Hot code to obtain the final label set.

[0047] A further improvement of the present invention is that the operation of step (3) comprises:

[0048] The random forest prediction model is used to construct a basic model for lithology identification. The final feature set obtained in step (2) is used as an independent variable and the final label set is used as a dependent variable to train the basic model for lithology identification to obtain a trained basic model for lithology identification.

[0049] A further improvement of the present invention is that the operation of step (4) comprises:

[0050] During the drilling process, the real-time logging data and drilling data of the well to be drilled are processed in real time using the same method as in step (2) to obtain the final feature set of the well to be drilled; the final feature set of the well to be drilled is input into the lithology identification basic model obtained in step (3), and the lithology identification basic model outputs the lithology prediction result in real time;

[0051] During the drilling process, the lithology analysis results of the drilled section of the well to be drilled and the corresponding logging data and drilling data are collected, and the collected lithology analysis results are used as lithology measurement results;

[0052] At regular intervals or every time drilling reaches a certain depth, a probability correction model is established using the lithology prediction results and the corresponding lithology measured results, and the coefficients of the probability correction model are determined to complete the correction of the lithology identification model.

[0053] A further improvement of the present invention is that the probability correction model is as follows:

[0054] Assuming that there are m types of lithology to be predicted, the output of the basic lithology identification model is the probability p of each lithology. 1 ,p 2 ,p 3 ,…,p m , and the lithology probability corresponding to the measured lithology result at a certain depth is y 1 ,y 2 ,y 3 ,…,y m ;

[0055] The probability correction model is as follows:

[0056]

[0057] Wherein, P(y=1|p) is the probability after correction, p is the probability before correction, and A and B are the coefficients of the probability correction model respectively.

[0058] A further improvement of the present invention is that the operation of step (5) comprises:

[0059] Substituting the lithology prediction result as the probability before correction into the probability correction model obtained in step (4) to calculate the probability after correction;

[0060] The maximum probability among the corrected probabilities is found, and the category corresponding to the maximum probability is the category of the identified lithology, and real-time lithology identification is now achieved.

[0061] The second aspect of the present invention provides a lithology identification system, which includes: a historical data processing unit, a basic model building unit, a real-time data processing unit, a model correction and lithology identification unit, and a result output unit.

[0062] The historical data processing unit is used to collect and store the logging data and drilling data of the adjacent wells, and to construct a feature set and a label set using the logging data and drilling data;

[0063] The basic model building unit is connected to the historical data processing unit, and uses the feature set and label set constructed by the historical data processing unit to train the lithology identification basic model;

[0064] The real-time data processing unit collects the logging data and drilling data of the well to be drilled in real time, and constructs a feature set using the logging data and drilling data; and collects the lithology analysis results of the drilled well section of the well to be drilled and the corresponding logging data and drilling data in real time;

[0065] The model correction and lithology identification unit is connected to the real-time data processing unit and the basic model establishment unit respectively, and uses the lithology analysis results of the drilled section of the well to be drilled collected by the real-time data processing unit to correct the lithology identification basic model established by the basic model establishment unit in real time, and uses the corrected lithology identification basic model to obtain the lithology category in real time;

[0066] The result output unit is connected to the model correction and lithology identification unit, and outputs the lithology category obtained by the model correction and lithology identification unit, for example, it can be visually displayed in text form.

[0067] Compared with the prior art, the beneficial effects of the present invention are as follows: based on the data of drilled adjacent wells, the present invention uses the random forest method to establish a basic model for lithology identification based on logging data and drilling data, and uses a probability model to correct the basic model, thereby avoiding the prediction error of the model in the deployment of new wells due to differences in geological and engineering conditions, greatly improving the accuracy and timeliness of lithology identification, and providing a basis for increasing drilling speed, reducing drill bit wear and increasing the life of downhole tools. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The figure is a flowchart of the steps of the method of the present invention.

[0069] Figure 2 This is a schematic diagram of the One-Hot coding method for lithology and drill bits.

[0070] Figure 3 Schematic diagram of drilling condition identification based on real-time logging parameters.

[0071] Figure 4 It is a feature dataset and label set of a drilled adjacent well.

[0072] Figure 5 A comparison table of the identified lithology and the measured lithology for a certain well section.

[0073] Figure 6 It is a schematic diagram of the structure of the system of the present invention. DETAILED DESCRIPTION

[0074] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0075] Since lithology data is discrete and discontinuous data, such as sandstone, mudstone, argillaceous limestone, conglomerate, granite, etc., and the drilling parameters and the downhole lithology response law are highly nonlinear, it is difficult to use mathematical methods to establish the theoretical relationship between drilling parameters and response parameters and lithology. Machine learning classification technology has the ability to classify discrete data and has a strong mapping ability for high-dimensional nonlinear problems.

[0076] The present invention designs a real-time lithology discrimination method and system based on random forest classification and probability model correction, which can realize real-time discrimination of lithology at the drill bit during drilling, effectively improve the response speed of the drilling site to downhole conditions, help to improve drilling speed and efficiency, and increase the life of the drill bit.

[0077] In order to solve the problem of difficulty in obtaining downhole lithology information in real time during drilling, the present invention proposes a method and system for lithology identification while drilling based on random forest. The present invention is based on the logging data of the drilled adjacent wells or well sections, and uses the random forest algorithm and the probability model correction method to establish a lithology identification while drilling model, taking into account the response relationship between multiple characteristics and the lithology being drilled, including depth, drilling pressure, rotation speed, hook load, displacement, drilling fluid density, mechanical drilling speed, turntable torque, turntable torque variation coefficient, drill bit size, and drill bit model. In particular, the identification deviation caused by new drilling projects and differences in geological conditions is taken into account, which effectively improves the response speed and accuracy of the real-time judgment of downhole lithology during the drilling process, and optimizes the drilling speed, reduces drill bit wear, and increases the life of downhole tools by timely adjusting the drilling parameters.

[0078] Specifically, Figure 1 As shown, the embodiments of the method of the present invention are as follows

[0079] [Example 1]

[0080] The method comprises:

[0081] (1) Collect data from neighboring wells: Collect time series logging data and drilling data from neighboring wells to be drilled.

[0082] The time series logging data and drilling data include: vertical depth, inclined depth, displacement, drilling pressure, rotation speed, hook load, inlet flow, drilling fluid density, mechanical penetration rate, rotary table torque, drill bit size and drill bit model.

[0083] (2) Constructing a feature set and a label set: constructing a feature set and a label set using the time series logging data and drilling data of the adjacent wells, specifically including:

[0084] (21) Filtering data related to drilling conditions based on the inclination depth, displacement, rotation speed and drilling pressure in the logging data: Filtering logging data that meets the following conditions from the logging data, the data used later are the logging data filtered out in this step and the corresponding drilling data;

[0085] Q i >1L / s and MD i >MD i-1 And RPM i >1rpm and WOB i >1kN

[0086] Among them, Q i is the displacement corresponding to a certain point, in L / s; MD i is the oblique depth corresponding to a certain point, in meters; MD i-1 The slant depth corresponding to the previous point of a certain point, in meters; RPM i is the speed corresponding to a certain point, in rpm; WOB i is the drilling pressure corresponding to a certain point, in kN. Figure 3 In order to screen the logging data, the data belonging to the drilling conditions are extracted.

[0087] (22) calculating the torque variation coefficient based on the turntable torque data;

[0088] The calculation is performed using a sliding window method: the step length W of the oblique depth sequence data set is set, W is used as the sliding window, and the torque variation coefficient within the sliding window is calculated well by well:

[0089]

[0090] Where V T is the torque variation coefficient of the rotary table torque in the well section; σ T is the RMS torque of the rotary table in the well section, in kN.m; It is the average value of the rotary table torque in the well section, in kN.m.

[0091] For each adjacent well, the torque variation coefficient of multiple well sections is calculated in each adjacent well, and the length of each well section is W meters, that is, each well section in each adjacent well is calculated in turn.

[0092] Since the amount of data collected is huge, a sliding window is used to divide the well section, and the torque variation coefficient of each well section is calculated to reduce the amount of data. Then, the calculated torque variation coefficient is assigned to all the data in the sliding window. For example, assuming that the sliding window W is 10m, a torque variation coefficient is calculated every 10m. Assuming that there are 10 points in the sliding window, the calculated torque variation coefficient is assigned to the numerical item V corresponding to these 10 points. T Then, a column of feature items is added to the feature set of each well to form the original feature set. The value of the feature item is the torque variation coefficient of each point in the well.

[0093] (23) Process the original feature set and construct a new feature set and label set.

[0094] The slant depth is re-interpolated into an arithmetic sequence with a step size of S (that is, the slant depth is arranged as 0, S, 2S, and 3S), and S is used as a window. In each window, the value of each feature item in the original feature set is averaged to obtain the average value of each feature item, and the average value is filled into the new feature set accordingly. That is, the average value of all values ​​of each feature item in the same S is taken, and the average value is used as the value of the feature item in the window and filled into the feature set to form a new feature set.

[0095] The lithology is taken as the label set, that is, each S in the new feature set corresponds to a row of feature items, and each S in the label set corresponds to a lithology.

[0096] The feature items in the new feature set include: vertical depth, inclined depth, arrangement, drilling pressure, rotation speed, hook load, inlet flow, drilling fluid density, mechanical penetration rate, rotary table torque, drill bit size, drill bit model and torque variation coefficient. The reason why it is called a new feature set is that the inclined depth has changed due to interpolation;

[0097] (24) The new feature sets and label sets of all adjacent wells are shuffled in order according to the slant depth (for example, the original slant depth is in ascending order, such as 1 2 3 4 5, and after shuffling the order according to the slant depth, it may be: 4 1 3 2 5), and the feature sets of all adjacent wells after the shuffled order are concatenated into a feature set, and the label sets of all adjacent wells after the shuffled order are concatenated into a label set.

[0098] (25) The drill bit model data and lithology data are converted into numbers to obtain the final feature set and the final label set:

[0099] like Figure 2As shown, in the feature set obtained in step (24), the data of the drill bit model is One-Hot encoded, and the data of the drill bit model in the feature set is converted into the corresponding One-Hot code to obtain the final feature set. Specifically, the data of the drill bit model is converted into encoded numbers. For example, the drill bit model "ST915" is converted into the code [1, 0, 0, 0, 0], and the values ​​corresponding to "ST915" in the feature items of the feature set are replaced with 1, 0, 0, 0, 0 row by row to form the final feature set.

[0100] In the label set obtained in step (24), the lithology data is One-Hot encoded, the lithology data is converted into numbers, and the lithology data in the label set is replaced with the corresponding One-Hot code line by line to obtain the final label set. Specifically, the lithology data is converted into coded numbers, for example, the lithology data such as "limestone" is converted into the code [0, 0, 0, 1, 0], and the values ​​corresponding to "limestone" in the feature items of the label set are replaced with 0, 0, 0, 1, 0 line by line. The final label set is formed.

[0101] (3) Training the basic model for lithology identification:

[0102] The random forest prediction model is used to construct a basic model for lithology identification. The final feature set obtained in step (2) is used as an independent variable and the final label set is used as a dependent variable to train the basic model for lithology identification to obtain a trained basic model for lithology identification.

[0103] Specifically, the random forest prediction model is a machine learning method parallel to the neural network. Its principle is briefly described as follows: input the feature set, train the random forest with the label set, and after training, input the feature set, and the random forest prediction model outputs the lithology prediction results. These are all implemented using existing methods and will not be repeated here.

[0104] The random forest model has the advantages of strong sample size adaptability, no need for feature normalization, few hyperparameters and easy parameter adjustment. The hyperparameter search and K-fold cross validation methods are used to determine the random forest hyperparameters, and the hyperparameters are optimized to obtain the optimized random forest hyperparameters. The hyperparameter optimization method can use grid search, random search and Bayesian optimization methods (all of which are implemented using existing methods and will not be repeated here).

[0105] The optimized random forest hyperparameters are used to train the random forest model using the final feature set and the final label set obtained in step (2) to obtain the basic model for lithology identification. The method of using the feature set, label set and random forest prediction model to train the basic model for lithology identification can be implemented using the random forest method, which will not be described in detail here. Figure 4 It is the sample data used to establish the basic model of lithology identification.

[0106] (4) Real-time correction of lithology identification model:

[0107] During the drilling process, the real-time logging data and drilling data of the well to be drilled are processed in real time using the same method as in step (2) to obtain the final feature set of the well to be drilled; the final feature set of the well to be drilled is input into the lithology identification basic model obtained in step (3), and the lithology identification basic model outputs the lithology prediction result in real time;

[0108] During the drilling process, the lithology analysis results of the drilled section of the well to be drilled and the corresponding logging data and drilling data are collected, and the collected lithology analysis results are used as lithology measurement results;

[0109] At regular intervals or every time drilling reaches a certain depth, a probability correction model is established using the lithology prediction results and the corresponding lithology measured results (i.e. the prediction results and lithology measured results corresponding to the same logging data and drilling data). The coefficients of the probability correction model are determined through an iterative algorithm to complete the correction of the lithology identification model.

[0110] Specifically, the probability correction model is as follows:

[0111] Assuming that there are m types of lithology to be predicted, the output of the basic lithology identification model is the probability p of each lithology. 1 ,p 2 ,p 3 ,…,p m , and the lithology probability corresponding to the measured lithology result at a certain depth is y 1 ,y 2 ,y 3 ,…,y m .

[0112] The output of the basic model for lithology identification is probability, y i The corresponding value is the probability that the lithology at this depth is lithology i. For example, the probabilities output by the basic model for lithology identification are 0.1, 0.2, 0.5, and 0.2, indicating that the probabilities of the lithology being sandstone, mudstone, siltstone, and limestone are 0.1, 0.2, 0.5, and 0.2, respectively. If the measured lithology result is a certain lithology, the measured lithology result needs to be converted into a probability form. For example, if the measured lithology result is mudstone, the probabilities of converting it into sandstone, mudstone, siltstone, and limestone are 0, 1, 0, and 0, respectively, and the sum of the probability components is equal to 1.

[0113] Based on the measured lithology data and lithology prediction data, the probability output by the lithology identification basic model is corrected using the following formula:

[0114]

[0115] The above formula is the probability correction model.

[0116] Wherein, P(y=1|p) is the probability after correction, p is the probability before correction (ie, the probability output by the basic model for lithology identification), and A and B are the coefficients of the probability correction model, respectively.

[0117] The measured lithology data are used as the probability after correction, the predicted lithology data corresponding to the measured lithology data are used as the probability before correction, minimizing the cross entropy is used as the loss function, and an iterative algorithm is used to optimize and determine the coefficients A and B of the probability correction model. These are all existing technologies and will not be described here.

[0118] (5) Real-time identification of lithology:

[0119] Substituting the lithology prediction result as the probability before correction into the probability correction model obtained in step (4) to calculate the probability after correction;

[0120] The maximum probability among the corrected probabilities is found, and the category corresponding to the maximum probability is the identified lithology type, and real-time lithology identification is now achieved.

[0121] During the drilling process, the drilling fluid will continuously carry the rock cuttings at the drill bit to the ground, and the rock cuttings will be processed and analyzed (the existing method can be used, which will not be repeated here). After that, the lithology is obtained, that is, the lithology analysis results of the drilled section to be drilled, which are used as the lithology measurement results. Since it takes some time to carry the rock cuttings from the well to the ground, and it also takes some time to process and analyze the rock cuttings, but the drilling process does not stop, when the lithology measurement results are obtained, the drill bit has drilled a distance, and the lithology measurement results of the newly drilled distance cannot be obtained immediately. At this time, the lithology recognition basic model is used to obtain the lithology prediction results of the newly drilled distance in real time, thus realizing the real-time recognition of the lithology.

[0122] In order to make the prediction results more accurate, the present invention continuously corrects the lithology identification basic model according to the obtained lithology measured results. Specifically, the present invention performs correction every period of time or every time a certain depth is drilled, that is, the newly obtained lithology measured results are used to recalculate the values ​​of parameters A and B in the probability correction model every period of time (for example, 24 hours) or every time a certain depth is drilled (for example, 3 meters), and then the new A and B values ​​are substituted into the probability correction model to obtain an updated probability correction model. In the subsequent prediction process, the updated probability correction model is used to correct the probability output by the lithology identification basic model to obtain the corrected probability. The updated probability correction model is used for correction until the next recalculation to obtain new A and B values, that is, the probability correction model is updated again. The number of lithology measured results used for correction can be designed according to actual needs, for example, the latest 100 groups of lithology measured results, and the next correction uses the newly obtained 100 groups of lithology measured results.

[0123] For example, the calibration is performed every 50 meters. During the drilling process, the lithology recognition basic model is used to obtain the lithology prediction results in real time. At the same time, the lithology measured results and the prediction results corresponding to the drilling part are used to obtain the parameters A and B in the probability correction model. That is, the lithology measured results are used as the corrected probability in the probability correction model, and the lithology prediction results are used as the pre-corrected probability in the probability correction model. Both are substituted into the probability correction model to obtain the values ​​of parameters A and B. The calibration within 50 meters uses the values ​​of A and B until the new values ​​of A and B are obtained again in the next 50 meters.

[0124] Figure 5 Comparison between the model identification results and the sand dredging interpretation results after time delay. Figure 5 It can be seen that the method of the present invention can achieve higher recognition accuracy.

[0125] The embodiments of the real-time lithology identification system provided by the present invention are as follows:

[0126] [Example 2]

[0127] like Figure 6 As shown, the second aspect of the present invention provides a lithology identification system, which includes: a historical data processing unit 10, a basic model building unit 20, a real-time data processing unit 30, a model correction and lithology identification unit 40, and a result output unit 50.

[0128] The historical data processing unit 10 is used to collect and store the logging data and drilling data of the adjacent wells, and use the logging data and drilling data to construct a feature set and a label set; specifically, the feature set and the label set are constructed by using step (2) in the above method;

[0129] The basic model building unit 20 is connected to the historical data processing unit 10, and uses the feature set and label set constructed by the historical data processing unit 10 to train the lithology recognition basic model; specifically, the lithology recognition basic model is trained using step (3) in the above method;

[0130] The real-time data processing unit 30 collects the logging data and drilling data of the well to be drilled in real time, and constructs a feature set using the logging data and drilling data; and collects the lithology analysis results of the drilled section of the well to be drilled and the corresponding logging data and drilling data in real time; specifically, the feature set is constructed using step (2) in the above method.

[0131] The model correction and lithology identification unit 40 is connected to the real-time data processing unit 30 and the basic model establishment unit 20 respectively, and uses the lithology analysis results of the drilled section of the well to be drilled collected by the real-time data processing unit 30 to correct the lithology identification basic model established by the basic model establishment unit 20 in real time, and uses the corrected lithology identification basic model to obtain the type of lithology in real time. Specifically, the lithology identification basic model established by the basic model establishment unit 20 is corrected in real time using step (4) in the above method, and the lithology category is obtained in real time using step (5) in the above method.

[0132] The result output unit 50 is connected to the model correction and lithology identification unit 40 to output the lithology category obtained by the model correction and lithology identification unit 40 , for example, it can be visually displayed in text form.

[0133] The present invention effectively improves the response speed and accuracy of the real-time judgment of downhole lithology during the drilling process. The response relationship between multiple characteristics and the lithology being drilled is considered, including depth, drilling pressure, rotation speed, hook load, displacement, drilling fluid density, mechanical drilling speed, rotary table torque, rotary table torque variation coefficient, drill bit size, and drill bit model. The basic model for lithology identification is established based on the random forest algorithm using the data of neighboring wells, and the basic model is corrected using the data of the marked well section being drilled, and the real-time logging and drill bit information are connected to realize the lithology identification of the rock being drilled.

[0134] Finally, it should be explained that the above technical solution is only one implementation method of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the application methods and principles disclosed in the present invention, and it is not limited to the method described in the above specific implementation method of the present invention. Therefore, the method described above is only preferred and does not have a restrictive meaning.

Claims

1. A method for real-time identification of lithology during drilling, characterized in that: The method comprises: Step (1) collecting adjacent well data; Step (2) constructing feature sets and label sets: constructing feature sets and label sets using adjacent well data; Step (3) training a basic model for lithology identification: using the feature set and the label set to train a basic model for lithology identification; Step (4) real-time correction of the basic model for lithology identification; Step (5) real-time identification of lithology: using the corrected lithology identification basic model to perform real-time identification of lithology and obtain the lithology category; The operation of step (1) includes: Collect time series logging data and drilling data of adjacent wells to be drilled, including vertical depth, inclined depth, displacement, drilling pressure, rotation speed, hook load, inlet flow, drilling fluid density, mechanical penetration rate, rotary table torque, drill bit size and drill bit model; The operation of step (2) includes: Step (21) screening data related to drilling conditions according to the inclined depth, displacement, rotation speed and drilling pressure in the logging data; Step (22) calculates the torque variation coefficient according to the turntable torque, and adds the torque variation coefficient feature item to form the original feature set; Step (23) processes the original feature set to construct a new feature set and label set; Step (24) concatenates the new feature sets and label sets of all adjacent wells into a feature set and a label set respectively; Step (25) converts the drill bit model data in the feature set and the lithology data in the label set into numbers to obtain a final feature set and a final label set; The operation of step (3) includes: A basic model for lithology identification is constructed using a random forest prediction model, and the final feature set obtained in step (2) is used as an independent variable and the final label set is used as a dependent variable to train the basic model for lithology identification to obtain a trained basic model for lithology identification; The operation of step (4) includes: During the drilling process, the real-time logging data and drilling data of the well to be drilled are processed in real time using the same method as in step (2) to obtain the final feature set of the well to be drilled; the final feature set of the well to be drilled is input into the lithology identification basic model obtained in step (3), and the lithology identification basic model outputs the lithology prediction result in real time; During the drilling process, the lithology analysis results of the drilled section of the well to be drilled and the corresponding logging data and drilling data are collected, and the collected lithology analysis results are used as lithology measurement results; At regular intervals or every time drilling reaches a certain depth, a probability correction model is established using the lithology prediction results and the corresponding lithology measured results, and the coefficients of the probability correction model are determined to complete the correction of the lithology identification model.

2. The method for real-time identification of lithology during drilling according to claim 1, characterized in that: The operation of step (21) includes: Select the logging data that meets the following conditions from the logging data; Q i >1L / s and MD i >MD i-1 And RPM i >1rpm and WOB i >1kN Among them, Q i is the displacement corresponding to a certain point; MD i is the oblique depth corresponding to a certain point; MD i-1 The slope depth corresponding to the previous point of a certain point; RPM i is the speed corresponding to a certain point; WOB i is the drilling pressure corresponding to a certain point.

3. The method for real-time identification of lithology during drilling according to claim 2, characterized in that: The operation of step (22) includes: The step length W of the inclined depth is set, and W is used as the sliding window. The torque variation coefficient in each well section of each well is calculated using the following formula: The length of each well section is W meters; Where V T is the torque variation coefficient of the rotary table torque in the well section; σ T is the RMS torque of the rotary table in the well section, in kN.m; is the average value of the rotary table torque in the well section, in kN.m; The calculated torque variation coefficient is then assigned to each point in the well section; Finally, a column of feature items is added to the feature set of each well to form the original feature set. The value of the feature item is the torque variation coefficient of each point in the well.

4. The method for real-time identification of lithology during drilling according to claim 3, characterized in that: The operation of step (23) includes: Re-interpolate the slant depth into an arithmetic sequence with a step size of S; Take S as the window, take the average value of each feature item in the original feature set in each window to get the average value of each feature item, and use the average value as the value of the feature item in the window, fill it into the feature set to form a new feature set; Use lithology as the label set.

5. The method for real-time identification of lithology during drilling according to claim 4, characterized in that: The operation of step (24) includes: Shuffle the order of the new feature sets and label sets of all adjacent wells according to the oblique depth; The feature sets of all neighboring wells after shuffling are concatenated into a feature set, and the label sets of all neighboring wells after shuffling are concatenated into a label set.

6. The method for real-time identification of lithology during drilling according to claim 5, characterized in that: The operation of step (25) includes: In the feature set obtained in step (24), the data of the drill bit model is One-Hot encoded, the data of the drill bit model is converted into the corresponding One-Hot code, and the data of the drill bit model in the feature set is replaced line by line with the corresponding One-Hot code to obtain the final feature set; In the label set obtained in step (24), the lithology data is One-Hot encoded, the lithology data is converted into the corresponding One-Hot code, and the lithology data in the label set is replaced line by line with the corresponding One-Hot code to obtain the final label set.

7. The method for real-time identification of lithology during drilling according to claim 6, characterized in that: The probability correction model is as follows: Assuming that there are m types of lithology to be predicted, the output of the basic lithology identification model is the probability p1, p2, p3, ..., p m , and the lithology probability corresponding to the measured lithology result at a certain depth is y1,y2,y3,…,y m ; The probability correction model is as follows: Wherein, P(y=1|p) is the probability after correction, p is the probability before correction, and A and B are the coefficients of the probability correction model respectively.

8. The method for real-time identification of lithology during drilling according to claim 7, characterized in that: The operation of step (5) includes: Substituting the lithology prediction result as the probability before correction into the probability correction model obtained in step (4) to calculate the probability after correction; The maximum probability among the corrected probabilities is found, and the category corresponding to the maximum probability is the category of the identified lithology.

9. A lithology identification system, used to implement the method for real-time lithology identification during drilling as claimed in claim 1, characterized in that: The lithology identification system comprises: a historical data processing unit, a basic model building unit, a real-time data processing unit, a model correction and lithology identification unit, and a result output unit; The historical data processing unit is used to collect and store the logging data and drilling data of the adjacent wells, and to construct a feature set and a label set using the logging data and drilling data; The basic model building unit is connected to the historical data processing unit, and uses the feature set and label set constructed by the historical data processing unit to train the lithology identification basic model; The real-time data processing unit collects the logging data and drilling data of the well to be drilled in real time, and constructs a feature set using the logging data and drilling data; and collects the lithology analysis results of the drilled well section of the well to be drilled and the corresponding logging data and drilling data in real time; The model correction and lithology identification unit is connected to the real-time data processing unit and the basic model establishment unit respectively, and uses the lithology analysis results of the drilled section of the well to be drilled collected by the real-time data processing unit to correct the lithology identification basic model established by the basic model establishment unit in real time, and uses the corrected lithology identification basic model to obtain the lithology category in real time; The result output unit is connected to the model correction and lithology identification unit, and outputs the lithology category obtained by the model correction and lithology identification unit.

Images