A method for drilling rate prediction before drilling with credibility considering uncertainty of multi-source information and application

Abstract

The present application belongs to the technical field of oil and gas drilling engineering, and particularly relates to a drilling speed prediction method with credibility considering multi-source information uncertainty and application, the drilling speed prediction method with credibility including the following steps: S1, multi-source data integration and mechanical drilling speed influencing factor definition; S2, multi-source uncertainty parameter probability distribution model establishment; S3, related random drilling scene generation and drilling engineering constraint filtering; S4, mechanical drilling speed prediction model calculation with cross-well residual correction; S5, output drilling speed prediction result with credibility. The present application can establish a multi-source uncertainty parameter model in the pre-drilling stage, propagate input parameter uncertainty to the mechanical drilling speed prediction result, and output a probabilistic prediction result, so as to improve the scientificity and reliability of pre-drilling speed decision.

Description

Technical Field

[0001] This invention relates to the field of oil and gas drilling engineering technology, specifically to a reliable mechanical drilling rate prediction method and its application that considers the uncertainty of multi-source information. Background Technology

[0002] Mechanical drilling speed It is an important parameter for evaluating drilling efficiency and predicting drilling cycle. In the pre-drilling stage, the mechanical rate of penetration (MRP) of the target well is usually predicted based on data from adjacent wells, target formations, lithological understanding, drill bit scheme, and planned drilling parameters to guide drill bit selection, drill string assembly design, drilling parameter window design, and speed-up scheme deployment.

[0003] Existing methods for predicting mechanical drilling speed mainly include empirical formula methods, rock drillability methods, and machine learning prediction methods. These methods typically calculate a single value using fixed input parameters. Predicted values ​​have some applicability when data patterns are stable. However, in the actual pre-drilling design stage, factors such as formation interfaces, lithological combinations, rock strength, abrasiveness, drilling pressure, rotational speed, displacement, pump pressure, drill bit wear, and tool efficiency all involve uncertainties. If only a deterministic value is output... Values ​​cannot be quantified. Engineering questions such as "What is the probability of drilling below the minimum economic drilling speed?", "What is the probability of a certain drill bit scheme reaching the design drilling speed?", and "Are the candidate speed-up schemes stable?"

[0004] Pre-drilling mechanical drilling rate prediction faces significant multi-source uncertainties, mainly including five categories: The first category is geological and rock drillability uncertainty, including formation interface depth error, layer thickness variation, lithological longitudinal and lateral variations, rock strength, abrasiveness, and fluctuations in drillability parameters; the second category is formation pressure and wellbore pressure environment uncertainty, including pore pressure, fracture pressure, collapse pressure, leakage pressure, equivalent circulation density, and changes in the safe density window; the third category is drill bit, drill string assembly, and speed-up tools uncertainty, including drill bit type, ... The first category includes: differences in coding, tooth profile, tooth arrangement, nozzle flow path, individual drill bit performance, drill bit wear and decay, and differences in the efficiency of drill string assemblies and speed-up tools; the second category includes: uncertainties in drilling engineering parameters and construction execution, including deviations in the execution of drilling pressure, rotational speed, displacement, torque, pump pressure, drilling fluid parameters, and field parameters; the third category includes: uncertainties in data cognition and model generalization, including insufficient coverage of adjacent well samples, differences in data distribution between different wells and different drilling operations, missing information on undrilled sections of the target well, and generalization errors in cross-well applications of the model.

[0005] Traditional mechanical drilling rate prediction methods typically treat the aforementioned factors as deterministic inputs, or only establish a single-value mapping relationship between drilling parameters and mechanical drilling rate within an existing sample range. This makes it difficult to describe the propagation effect of multi-source uncertainties on pre-drilling mechanical drilling rate prediction results. Therefore, the prediction results obtained by traditional methods usually appear as a single mechanical drilling rate curve or a single numerical value, failing to provide a comprehensive picture. The reliability interval, the probability of low mechanical drilling speed risk, and the reliability level are insufficient to fully meet the engineering decision-making needs of pre-drilling bit selection, drilling parameter window design, and speed-up tool selection.

[0006] Therefore, there is a need for a method that can establish a multi-source uncertainty parameter model in the pre-drilling stage, propagate the uncertainty of input parameters to the mechanical drilling rate prediction results, and output probabilistic prediction results, so as to improve the scientificity and reliability of pre-drilling rate decision-making. Summary of the Invention

[0007] This invention addresses the aforementioned technical problems and overcomes the deficiency of traditional deterministic mechanical drilling rate prediction methods in failing to characterize pre-drilling uncertainties. It proposes a reliable mechanical drilling rate prediction method implemented according to the steps of "probabilistic modeling → relevant sampling → drilling engineering constraint filtering → mechanical drilling rate prediction model calculation → cross-well residual calibration → well depth probability profile output". In a preferred embodiment, the mechanical specific energy... Leakage of the target variable can be avoided; after obtaining the prediction results, further risk assessment and scheme optimization can be carried out.

[0008] The "confidence-based prediction results" mentioned in this invention refer to the set of mechanical drilling rate prediction samples and their probability statistics formed after Monte Carlo random drilling scenario simulation and cross-well residual calibration, including at least... , And the P05-P95 confidence interval band between the two. Based on the prediction results, the following can also be calculated: , , and Furthermore, a comprehensive score for candidate solutions is generated; the above indicators are application indicators of the prediction results.

[0009] Unlike simply concatenating machine learning models with Monte Carlo simulations, this invention focuses on the feasibility of drilling engineering: First, random samples are limited to effective drilling scenarios that meet the requirements of drill bit-lithology compatibility, drilling parameter windows, displacement-pump pressure coupling, drilling fluid density windows, and tool efficiency ranges; second, in a preferred embodiment, mechanical specific energy is considered. Unknown factors may be introduced during the pre-drilling phase. The problem is that separate training phases for historical wells and pre-drilling prediction phases for target wells are set up. The calculation path is then used. Next, the model generalization residual is obtained by leaving one cross-well for verification in each well, and this residual is propagated together with the uncertainty of the input parameters. Finally, a pre-drilling prediction probability profile with credibility mechanical drilling rate is formed with well depth as the ordinate, so that the prediction results have engineering feasibility, model generalization error calibration capability and risk interpretation capability along well depth.

[0010] The specific technical solution of the present invention is as follows: A reliable mechanical drilling rate prediction method considering the uncertainty of multi-source information, comprising the following steps: S1. Multi-source data integration and definition of factors affecting mechanical drilling rate: The first... The factors influencing the mechanical drilling rate of the nth sample are defined as feature vectors, and the nth sample is used as the feature vector. The target value for each sample is defined as , This refers to the mechanical drilling speed; The factors affecting mechanical drilling speed are classified into five categories according to their sources of uncertainty, namely, multi-source uncertainty parameters; S2. Establishment of a probability distribution model for multi-source uncertainty parameters: The multi-source uncertainty parameters are classified into continuous parameters, discrete parameters, and interval parameters according to their numerical characteristics, and corresponding probability distribution models are established for each. S3. Generation of relevant random drilling scenarios and filtering of drilling engineering constraints: A correlation matrix is ​​established for the aforementioned multi-source uncertainty parameters to generate a stochastic drilling scenario with specified correlations. Drilling engineering constraint filtering is then applied to obtain the final result. Group of random drilling scenarios; S4. Calculation of mechanical drilling rate prediction model including cross-well residual calibration: Training a mechanical drilling rate prediction model; A cross-well residual sample library was established by adopting either a one-well-by-well or a one-block-by-block cross-well verification method. The result obtained in step S3 The initial random drilling scenario is input into the mechanical drilling rate prediction model to obtain the initial... The sample set is obtained by resampling residuals from the cross-well residual sample library and superimposing them onto the prediction results to obtain the calibrated result. Predicted sample set; Step S4 also includes Target variable leakage avoidance; S5. Output the pre-drilling prediction results of mechanical drilling rate with confidence level: Repeat steps S2 to S4 for each well depth location of the target well to obtain the location of each well depth. Predict the sample set; calculate the values ​​at each well depth. and This forms a probability profile for pre-drilling prediction of mechanical drilling speed with credibility.

[0011] Based on the prediction results output in step S5, further risk assessment and scheme optimization can be carried out.

[0012] Furthermore, the specific steps for step S1, multi-source data integration, and the definition of factors affecting mechanical drilling speed are as follows: In step S1, multi-source data available before drilling is integrated. The multi-source data includes basic well information, well structure, formation stratification, lithological description, logging parameters, logging-while-drilling parameters, rock mechanics parameters, drill bit and tool parameters, and measured mechanical drilling rate.

[0013] The first The factors influencing the mechanical drilling rate of each sample are defined as feature vectors. Target value Defined as :

[0014] In equations (1) and (2), Number the sample; For depth measurement or vertical depth, the unit is meters (m). For layer; Lithology; Well logging characteristics; Characteristics of rock mechanics; Features of a drill bit; Features of drilling tools and speed-up tools; Drilling pressure, unit: kN; Rotational speed, in r / min; Displacement, in L / min; This refers to riser pressure or pump pressure, in MPa. For the first Mechanical drilling rate for each sample, in m / h.

[0015] Further, the multi-source uncertainties in pre-drilling mechanical drilling rate prediction are classified into five categories: the first category is geological and rock drillability uncertainty; the second category is formation pressure and wellbore pressure environment uncertainty; the third category is drill bit, drill string assembly, and speed-up tools uncertainty; the fourth category is drilling engineering parameters and construction execution uncertainty; and the fifth category is data cognition and model generalization uncertainty. These five categories of uncertainties collectively affect the achievable range and prediction reliability of the mechanical drilling rate in the target formation.

[0016] Furthermore, step S2, the establishment of the multi-source uncertainty parameter probability distribution model, specifically involves: In step S2, probability distribution models are established for different types of mechanical drilling rate influencing factors. Continuous parameters can be obtained by statistically analyzing the mean, standard deviation, and quantiles of samples from adjacent wells in the same layer; discrete parameters can be obtained by obtaining the category probability through category frequency; interval parameters can be established by establishing a triangular distribution or a uniform distribution based on the design lower limit, upper limit, and most likely value.

[0017] For the five types of uncertainties, this invention establishes corresponding probabilistic representation models. For continuous parameters, normal, log-normal, triangular, uniform, Beta, or Gamma distributions are used for description; for discrete parameters, categorical probability distributions are used; and for interval parameters, uniform or triangular distributions are used. These uncertainties can not only be used as joint sampling inputs for mechanical drilling rate prediction, but their probability distribution results can also be output separately to demonstrate the impact of different types of uncertainties on pre-drilling mechanical drilling rate prediction results.

[0018] The probability distribution of continuous parameters is preferably determined as follows: first, outlier identification and physical boundary checks are performed on samples from adjacent wells in the same layer, and then candidate distributions such as normal distribution, log-normal distribution, Gamma distribution, Beta distribution, triangular distribution, and uniform distribution are fitted respectively; when the sample size meets the test requirements, one or more of the following are used as the basis for distribution selection: Kolmogorov-Smirnov test, Anderson-Darling test, Akaike Information Criterion (AIC), or Bayesian Information Criterion (BIC); when the sample size is insufficient or only the design parameter window is available, the triangular distribution determined by the lower limit, the most likely value, and the upper limit are preferred, or the uniform distribution determined by the lower limit and the upper limit are preferred.

[0019] The class probability of discrete parameters is determined by the class frequency of samples from adjacent wells in the same layer or at the same opening; when the target well has a clear design scheme, the drill bit type, The code, drill string combination type, and speed-up tool type can be used as input variables for candidate schemes. When the target well only has alternative schemes, a category probability distribution can be established according to the set of candidate schemes, and samples that do not meet the engineering fit relationship can be removed in the subsequent drilling engineering constraint filtering.

[0020] In the embodiments, the probability distributions of uncertainties in geological and rock drillability, formation pressure and wellbore pressure environment, drill bit and drilling tool and speed-up tools, drilling engineering parameters and construction execution, and data cognition and model generalization can be plotted respectively. Based on this, a probability profile of the pre-drilling prediction of mechanical drilling speed with credibility after the propagation of multi-source uncertainties can be plotted.

[0021] For continuous parameters Its mean and standard deviation can be calculated using the following formula:

[0022] That Quantiles are determined by the cumulative distribution function. Inverse function determined:

[0023] In equations (3) to (5), For the sample size, For sample number, As parameter number, For the first In the nth sample The possible values ​​of each parameter and The first The mean and standard deviation of each parameter, (·) is the first The inverse function of the cumulative distribution function with one parameter. For quantile levels, For the first One parameter Quantiles.

[0024] For discrete parameters , its first The class probability of the class value is:

[0025] In equation (6), Discrete parameters Take the first The number of samples for each class value, M is the total number of classes, and the denominator is the sum of the number of samples for each class.

[0026] For interval parameters, if the lower limit is known... Most likely value and upper limit A triangular distribution can be established, and its probability density function is:

[0027] Table 1 shows the preferred probability distribution types supported by this method and their applicable scenarios.

[0028] Table 1. Types and Parameter Definitions of Probability Distributions

[0029] Furthermore, step S3, the generation of relevant random drilling scenarios and the filtering of drilling engineering constraints, specifically involves: In step S3, considering that drilling engineering parameters, formation lithology parameters and tool parameters are not independent of each other, we do not directly sample each parameter independently. Instead, we first establish a correlation matrix, then generate a random drilling scenario with specified correlation, and further filter the drilling engineering constraints.

[0030] parameter and The correlation coefficient between them is defined as:

[0031] In equation (9), ( , ) is a parameter and covariance, and These are the standard deviations of the two parameters. The correlation coefficient between the two parameters is denoted as .

[0032] The original correlation matrix is ​​composed of all correlation coefficients. .like If the result is not positive definite, it is made positive definite through eigenvalue correction:

[0033] In equations (10) and (11), The original correlation matrix, This is the positive definite correlation matrix. The eigenvector matrix, to for eigenvalues, For the number of parameters, T is a minimal positive number used to correct non-positive eigenvalues. T is a mathematical symbol meaning the transpose of a matrix.

[0034]

[0035] Generate independent standard normal random vectors Subsequently, the relevant standard normal random vector for:

[0036] Then, random samples for each parameter are generated through probability integral transformation:

[0037] In equations (12) and (13), For the first A group of independent standard normal random vectors, The correlation is the standard normal random vector obtained by the correlation matrix transformation. for The There are 1 component, Φ(·) is the standard normal distribution function. (·) is the first The inverse function of the marginal cumulative distribution function with one parameter. For the generated first The first scene in the group A random sample with parameters.

[0038] To prevent the generation of random combinations that do not conform to engineering principles or are unenforceable on-site, this invention performs drilling engineering constraint filtering on random drilling scenarios. Let the first... Group of random scenarios Engineering inequality constraints are The equality constraint is Then the effective sample discriminant function is:

[0039] In equation (14), ( ) is the inequality constraint discriminant function. ( ) is the equality constraint discrimination function; it takes the value 1 if all corresponding constraints are satisfied, otherwise it takes the value 0. For the first Effective sample discrimination results for a group of random scenarios; when When =0, the random scenario is discarded or regenerated. The drilling engineering constraint filtering includes, but is not limited to, drill bit type and allowable lithology matching constraints, upper limit constraints on drilling pressure, upper limit constraints on rotation speed, constraints on the correspondence between displacement and pump pressure, constraints on the applicable range of drilling fluid parameters, constraints on the safe window of wellbore pressure, and constraints on the range of tool efficiency.

[0040] Preferably, the drill bit-lithology compatibility constraint can be determined by the compatibility matrix between drill bit type and lithology category. When the sampled drill bit type is not suitable for the target lithology, the sample is discarded. The drill pressure and rotation speed constraints are jointly determined by the drill bit manufacturer's recommended range, drill string assembly strength, and field design parameter window. The displacement and pump pressure correspondence constraint is determined by the hydraulic calculation model, the measured relationship of adjacent wells, or the fitting function. The wellbore pressure safety window constraint is jointly determined by pore pressure, fracture pressure, collapse pressure, leakage pressure, and equivalent circulation density. The tool efficiency range constraint is determined by the speed-up tool manual, the application effect of adjacent wells, or field experience statistics.

[0041] Through the above drilling engineering constraint filtering, Monte Carlo sampling does not obtain arbitrary mathematical random combinations, but rather effective random drilling scenarios that can represent the feasible working conditions before drilling the target well, thereby avoiding the distortion of the pre-drilling prediction probability profile containing the reliability of mechanical drilling rate due to unreasonable parameter combinations.

[0042] like Figure 1 As shown, this invention categorizes the multi-source uncertainties in pre-drilling mechanical drilling rate prediction into five types: geological and rock drillability, formation pressure and wellbore pressure environment, drill bit and drilling tools and speed-up tools, drilling engineering parameters and construction execution, and data cognition and model generalization, and represents each type using probability distributions. Through Monte Carlo joint sampling, parameter correlation constraints, and drilling engineering constraint filtering, multiple sets of random drilling scenarios are generated and input into the mechanical drilling rate prediction model, ultimately obtaining the rate of drilling along the well depth. , and Confidence interval band. This figure illustrates the propagation process of various input uncertainties to the final pre-drilling prediction result containing confidence in the mechanical drilling rate.

[0043] Furthermore, step S4, the calculation of the mechanical drilling rate prediction model including cross-well residual calibration, is specifically as follows: In step S4, the mechanical drilling rate prediction model is first trained. This model can employ gradient boosting trees, random forests, neural networks, CatBoost, XGBoost, LightGBM, or a physical fusion model. The model input consists of valid random drilling scenarios filtered by relevant sampling and drilling engineering constraints, and the output is the corresponding predicted mechanical drilling rate.

[0044]

[0045] In equation (15), For parameters Mechanical drilling rate prediction model, For the first Group of effective random drilling scenarios, This is the predicted mechanical drilling rate for this scenario, in m / h.

[0046] Mechanical specific energy can be introduced into the model input features. As a priori physical feature. To avoid unknown factors in the pre-drilling prediction stage. This method will address the issue of target variable leakage caused by feature computation. The historical well training phase and the target well pre-drilling prediction phase were processed separately: the historical well training samples were based on actual measurements. calculate When predicting the target well before drilling, do not directly use unknowns. calculate Instead, the statistical method using adjacent wells in the same layer is adopted. Initial experience Or the previous model prediction Perform approximate calculations, or only count adjacent wells. As a priori physical input for the layer segment.

[0047] The The target leakage avoidance process includes: during the model training phase, Based solely on historical well measurements The results are obtained from calculations based on the corresponding drilling parameters; during the pre-drilling prediction stage of the target well, Do not use the real to be predicted According to the adjacent well in the same layer Quantiles, designed mechanical drilling rate, empirical mechanical drilling rate, or model iteration predictions are generated. Approximate value. If using model iterative predictions, first use values ​​without... The feature model obtains the initial Estimate, then calculate Approximate value and input containing The prediction model is iterated to two consecutive iterations. The estimated difference is less than the preset threshold or reaches the preset number of iterations.

[0048]

[0049]

[0050] In the formula, Drilling pressure, unit: kN; The area of ​​rock broken at the bottom of the well is expressed in m². Rotational speed, in r / min; The torque of the drill bit is expressed in kN·m. The measured mechanical drilling rate for historical wells is expressed in m / h. The initial mechanical drilling rate estimate is obtained from statistical values, empirical values, or model iterations of adjacent wells in the same layer during the pre-drilling stage, in m / h. The unit is kPa. It is used to characterize the energy required to break a unit volume of rock and can reflect the combined effect of drill bit, lithology, and drilling parameters.

[0051] For the target well, target interval, and a given candidate scheme, generate repeatedly A set of random drilling scenarios are input into the model to obtain... Sample set:

[0052] In equation (17), For a target well at a certain depth and corresponding candidate scheme Prediction sample set, The number of valid samples in Monte Carlo. For the first The calibrated mechanical drilling rate corresponding to a group of effective random drilling scenarios. The optimal value is 5000 to 10000, but it can also be adjusted based on computing resources and risk resolution.

[0053] To further quantify the uncertainty of data cognition and model generalization, this invention employs a cross-well validation method, either leaving one well in each well or leaving one block in each region, to establish a model residual sample library. Specifically, each time, one well or one block is used as the validation object, and the remaining wells or blocks are used as training objects, obtaining the predicted and measured values ​​of the validation object. The residual between; when It is non-negative and the error varies with When the horizontal changes, it is preferable to calculate the residuals in logarithmic space and form a cross-well residual distribution.

[0054] In the target well pre-drilling prediction stage, effective random drilling scenarios are input into the mechanical drilling rate prediction model to obtain the initial... After sampling, residuals are resampled from the cross-well residual sample library and superimposed onto the prediction results; if residual calibration is performed in log space, then first... The predicted values ​​are logarithmed, then the logarithmic residuals from resampling are summed, and finally, the calibrated values ​​are obtained through an inverse exponential transform. Sample. Thus, the final probability profile for pre-drilling prediction of reliable mechanical drilling rate simultaneously includes the results of input parameter uncertainty propagation and model cross-well generalization error.

[0055] Furthermore, step S5, outputting the pre-drilling prediction result of the mechanical drilling rate with confidence level, specifically involves: For each well depth location of the target well, steps S2 to S4 are performed to generate multiple sets of valid random drilling scenarios and calculate the corresponding mechanical drilling rate. The obtained results for each well depth location... The predicted sample set, and the data from each well depth location. and The resulting probability profile of pre-drilling prediction of mechanical drilling speed with credibility constitutes the main prediction result of this invention. Based on this, the quantiles, interval widths, and risk probabilities at each well depth can be further calculated, and these can be summarized to form target-level evaluation indicators for risk assessment and scheme selection.

[0056] like Figure 2 As shown, the probability profile of pre-drilling prediction of mechanical drilling rate with confidence level is plotted with well depth as the vertical axis and mechanical drilling rate as the horizontal axis. The x-axis is... and The lower and upper boundaries of the prediction results are respectively used as the lower and upper boundaries, and the shaded area between them represents... Confidentiality zone. This profile directly displays the variation in the range of predicted mechanical drilling rate at different well depths, thus reflecting the un-drilled section of the target well. The reliability level of the prediction results.

[0057] This invention also discloses the application of any of the above-described methods for predicting pre-drilling mechanical drilling speed with credibility, considering the uncertainty of multi-source information, in the field of oil and gas drilling engineering technology.

[0058] It can be used to evaluate the stability of pre-drilling plans and screen the optimal plan among multiple candidate plans. The specific steps are as follows: The target segment corresponding to the predicted scheme is discretized into multiple well depth locations according to a preset well depth step size. Based on the above-mentioned pre-drilling prediction method for mechanical drilling speed with credibility, which considers the uncertainty of multi-source information, the well depth locations are obtained. Predicted sample set; As in Figure 2 Based on the pre-drilling probability profile of the mechanical drilling rate with credibility shown, the quantiles, interval widths and risk probabilities of the predicted samples at each well depth are further calculated to obtain quantitative indicators for scheme evaluation. The specific calculations are as follows.

[0059]

[0060] The P05-P95 confidence interval band width Defined as:

[0061] In equations (18) to (22), The number of valid prediction samples at a certain well depth. For the first A sample of mechanical drilling rate predictions and These are the sample mean and sample standard deviation, respectively. ( )for sample set Quantiles For quantile levels, -P95 is a confidence interval composed of the 5th percentile and the 95th percentile. The width of the confidence interval.

[0062] Let the minimum economic drilling rate threshold be... Drilling speed threshold is The threshold for successful speed-up is The probability of low mechanical drilling speed risk, the probability of not achieving the target, the probability of achieving the target, and the probability of successfully increasing drilling speed are defined as follows: The minimum economic drilling rate threshold, target drilling rate threshold, and successful speed-up threshold can be determined as follows: when the block or project already has drilling cycle design indicators, the mechanical drilling rate corresponding to the design indicators is used as the threshold; when there is statistical data on adjacent wells in the same layer, the adjacent wells in the same layer are used. The thresholds are the quantile, the lower limit of the economic drilling rate, or the mechanical drilling rate calculated from the average drilling cycle. When contractual or field operational constraints exist, the mechanical drilling rate corresponding to the contractual indicators, well section construction window, or speed-up target is used as the threshold. These thresholds are kept consistent throughout the evaluation of the same candidate scheme to ensure comparability between schemes.

[0063] The three thresholds are usually satisfied ≤ When a project has only one success threshold set, Can be with Take the same value. This indicates the probability of drilling below the minimum economic drilling rate. This indicates the probability that the target drilling speed has not been reached. This indicates the probability of achieving the target drilling speed. This indicates the probability of successfully reaching the speed-up threshold.

[0064]

[0065] In equations (23) to (26), (·) indicates the number of samples that satisfy the conditions within the parentheses; , , and These represent the probability of low mechanical drilling speed, the probability of not reaching the target, the probability of achieving the target, and the probability of successfully increasing the drilling speed. For the same... , + =1.

[0066] The instability index of the scheme can be determined by the confidence interval bandwidth of P05-P95. Together with the coefficient of variation, it characterizes:

[0067] In equation (27), The bandwidth of the confidence interval from P05 to P95 is [missing information]. The average mechanical drilling speed, The standard deviation of mechanical drilling rate, and These are non-negative weighting coefficients, and + =1.

[0068] When performing target-level scheme evaluation, the indicators at each well depth are first summarized into the target layer. For the first... Each target segment will be used to create a set of well depth locations within the segment boundaries. and with the first The depth step represented by each well depth position (Unit: m) represents the weight, for , , , , and A weighted average is performed to obtain the corresponding layer-level index. When the well depths are evenly spaced, the weighted average is equivalent to the arithmetic average.

[0069] The stability of the proposed scheme is evaluated using the segment-level scheme instability index. A higher value indicates a poorer stability of the pre-drilling scheme.

[0070] Different candidate schemes use the same layer boundaries, well depth grids and weights to ensure consistent comparison results.

[0071] For multiple candidate solutions, a comprehensive scoring function can be constructed:

[0072] In equation (28), Number the candidate schemes. For the first The overall score of each candidate solution (·) is the normalization function; , , , and All are the first Each candidate solution corresponds to a segment-level indicator of the target segment; to These are non-negative weight coefficients, and the sum of the weights is 1.

[0073] The methods for determining weighting coefficients include one or more of the following: expert weighting, analytic hierarchy process (AHP), entropy weighting, or historical scheme effect inversion. If the project focuses more on a conservative achievable drilling rate, then the weighting coefficients should be increased. Corresponding weight; if the project focuses more on the speed-up effect, then increase the weight. and Corresponding weight; if the project focuses more on risk control, then increase the weight. and Corresponding weights. When there is a lack of explicit preferences, a pre-defined weight reorganization can be used, for example... =0.25、 =0.25、 =0.20、 =0.15、 =0.15, and the weighting remains unchanged in the candidate scheme evaluation report.

[0074] The final recommended option is the one with the highest overall score:

[0075] In equation (29), Let be the ID of the final recommended solution, and arg max represent the number of solutions selected from all candidates. The scheme number corresponding to the maximum value.

[0076] Table 2 presents the evaluation criteria based on the risk probability of low mechanical drilling rate.

[0077] Table 2 Risk Assessment Standards for Pre-Drilling Mechanical Drilling Rate

[0078] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This invention categorizes the multi-source uncertainties in pre-drilling mechanical drilling rate prediction into five types, including geological and rock drillability uncertainties, formation pressure and wellbore pressure environment uncertainties, drill bit and tool assembly and speed-up tools uncertainties, drilling engineering parameters and construction execution uncertainties, and data cognition and model generalization uncertainties, thereby more comprehensively reflecting the main sources of uncertainty in pre-drilling mechanical drilling rate prediction.

[0079] 2. This invention can establish probability distribution models for the above-mentioned types of uncertainties and display them separately, making the sources of uncertainty in pre-drilling mechanical drilling speed prediction clearer and more interpretable.

[0080] 3. This invention uses Monte Carlo joint sampling, parameter correlation constraints, and drilling engineering constraint filtering to propagate the uncertainty of multi-source inputs to the mechanical drilling rate prediction results, ensuring that the random drilling scenario meets the field executable conditions such as drill bit-lithology adaptation, drilling parameter window, wellbore pressure safety window, and tool efficiency range, thus realizing the transformation from traditional single-value prediction to a probability profile of mechanical drilling rate prediction before drilling with credibility.

[0081] 4. This invention is achieved through... Targeted leakage avoidance and cross-well residual calibration prevent the direct use of unknowns during the pre-drilling prediction stage. Calculate physical prior features and incorporate model generalization error. , and The confidence interval band statistical results are used to characterize the confidence range of the mechanical drilling rate prediction results at different well depths.

[0082] 5. This invention can evaluate the stability of pre-drilling schemes based on the probability of low mechanical drilling speed risk, the probability of achieving the target drilling speed, and the width of the confidence interval, providing quantitative basis for drill bit selection, drilling parameter window design, speed-up tool selection, and engineering scheme comparison. Attached Figure Description

[0083] Figure 1 A schematic diagram of a reliable mechanical drilling rate prediction method considering the uncertainty of multi-source information; Figure 2 This is a probability profile of pre-drilling prediction of mechanical drilling speed with credibility. Figure 3a This is a chart showing the percentage of different lithological types. Figure 3b This is a diagram showing the pressure distribution in the riser. Figure 3c This is a diagram showing the drilling pressure distribution. Figure 3d This is a diagram showing the rotational speed distribution. Figure 3e This is a displacement distribution diagram; Figure 3f This is a chart showing the percentage of different drill bit types. Figure 3g For cross-well verification Residual distribution diagram.

[0084] Figure 4a This is a traditional single-value mechanical drilling rate prediction diagram for target well A in the example embodiment; Figure 4b This is a probability profile of the pre-drilling prediction of the mechanical drilling rate for target well A in the example. Detailed Implementation

[0085] The following embodiments are used to further illustrate the present invention, but do not constitute a limitation on the scope of protection of the present invention. Given that the present invention relates to multiple aspects such as multi-source data processing, probabilistic modeling, machine learning prediction, and drilling engineering constraints, the following focuses on implementation methods directly related to the technical points of the present invention; parts not described in detail can employ existing data processing, machine learning modeling, and drilling engineering calculation methods in the art.

[0086] Example 1: This embodiment selects historical well data from six wells (A, B, C, D, E, and F) in the same area to verify the pre-drilling prediction method with reliable mechanical drilling rate. Hereinafter, A to F are used as anonymous well codes. In the well-by-well verification, one well is selected each time as the target well, and the remaining five wells are used as training wells.

[0087] The method for predicting the pre-drilling rate of mechanical drilling with confidence, considering the uncertainty of multi-source information, includes the following steps: S1. Multi-source data integration and definition of influencing factors Based on historical well data from six wells (A to F), we compiled stratigraphic layers, lithological descriptions, logging parameters, logging-while-drilling parameters, bit parameters, drill string assembly parameters, and measured mechanical drilling rates. Using depth as the primary index, we performed depth matching on data from different sources to obtain standardized samples. Specifically, each bit record in the well's basic information was converted into a bit usage range, and each record was associated with the bit type, bit size, and bit number according to the well code and logging depth.

[0088] S2. Establishment of Probability Distribution Model In step S2, the logging CSV files, formation stratification files, and well basic information files of six historical wells (A to F) are read. The logging fields are quantified and filtered according to valid drilling conditions, resulting in 794,024 valid records. Further parsing from the well basic information file yields 62 drill bit information records; each record includes the entry depth and exit depth, with 50 records representing drill bit usage intervals where the exit depth is greater than the entry depth. Matching the drill bit usage intervals with the logging depth results in 739,622 depth-by-depth samples containing drill bit type and size. To reduce the impact of sampling frequency differences between wells on the fitting results, 30,000 records are extracted from each well, forming 180,000 balanced samples. Figures 3a to 3g The parameter distribution, drill bit type ratio, and cross-well verification residuals obtained based on actual data are presented.

[0089] Table 3. Examples of Uncertain Parameter Distribution in the Target Interval of Well A

[0090] Figures 3a to 3f This is a parameter distribution and drill bit type percentage chart generated based on actual data from the implementation examples. Figure 3g For cross-well verification Residual distribution map. The percentage of lithology categories is statistically analyzed based on the depth matching results between the formation stratification file and the effective drilling samples; standpipe pressure and drilling pressure are preferably normally distributed; rotation speed and displacement are preferably log-normally distributed; the percentage of drill bit types is statistically analyzed based on 62 drill bit records in the well basic information file, of which PDC bits account for 79.0%, insert roller cone bits account for 12.9%, milling roller cone bits account for 4.8%, and roller cone bits account for 3.2%. The well basic information file also includes drill bit number, drill bit size, depth of entry, depth of exit, entry time, exit time, footage, pure drilling time, IADC wear assessment, and drill string assembly description. For the first... Drill bit records, defining their usage intervals as follows: ≤ < ,in and These are the depth of entry into the well and the depth of exit from the well. For well logging depth measurement; for the last valid endpoint, it can be included in the corresponding interval. Based on the anonymous well code and By performing interval matching, the drill bit type and drill bit size can be associated with the depth-by-depth logging samples.

[0091] Drill bit information plays different roles in the cross-well residual calibration and candidate scheme simulation stages. In the cross-well residual calibration stage, historical well drill bit types and sizes are stitched together according to actual usage depth intervals and used as model input features. In the target well candidate scheme simulation stage, drill bit type, drill bit size, and planned drill string assembly are pre-drilling scheme variables, which can be fixed values ​​according to candidate schemes or sampled according to the probabilities of alternative schemes, and filtered using drilling engineering constraints based on drill bit-lithology compatibility and tool allowable ranges. IADC wear assessment is information obtained post-drilling and is not directly used as input for known pre-drilling features of the target well, but it can be used for grouped statistical analysis of prior historical drill bit performance degradation.

[0092] S3, Generation of Related Random Drilling Scenarios Establish With torque, With torque, With pump pressure, lithology and The correlation matrix between them. After performing a positive definiteness check on the correlation matrix, relevant random samples are generated using formulas (11) to (13). Each group of random samples needs to be filtered by drilling engineering constraints; for example, when a drill bit is not suitable for the lithology obtained from the sampling, the sample is removed; when or When the sample exceeds the tool's allowed range, it is regenerated.

[0093] S4, Monte Carlo Simulation and Computation Set the number of Monte Carlo simulations =10000. Each simulation generates a set of valid random drilling scenarios that satisfy parameter correlation constraints and are filtered by drilling engineering constraints, and then calls the trained prediction model to calculate... During the Monte Carlo simulation phase of the candidate scheme, the model inputs include stratigraphy, lithology, depth sounding, logging characteristics, drilling pressure, rotational speed, torque, displacement, pump pressure, drill bit type, drill bit size, planned drill string assembly type, and tool efficiency, etc. The drill bit type and drill bit size are fixed values ​​or sampled according to the candidate scheme, and the wear evaluation obtained after drilling the target well is not used.

[0094] To quantify the uncertainty of data perception and model generalization, this embodiment uses depth measurement, drilling pressure, torque, rotational speed, displacement, drilling fluid density, riser pressure, and drill bit type and size obtained by splicing together the operating depth ranges as input features. Drill bit type is categorically coded. A ridge regression baseline model is established, and a cross-well residual sample library is formed. Mechanical drilling rate is also considered. It is a non-negative variable and the error magnitude varies with Horizontal changes; in this embodiment, the target value is adopted... Transformation, model fitting in logarithmic space and well-by-well leave-one-across-well verification, then through Inverse transformation yields Predicted values. One well from six wells (A to F) was used as the target well for verification each time, and the remaining five wells were used as training wells, for a total of six verification cycles. After summarizing 180,000 test samples, the mean absolute error of the mechanical drilling rate prediction residuals was 9.71 m / h, the root mean square error was 18.10 m / h, and the mean residual was 1.91 m / h. The residual distribution is shown in [see attached diagram]. Figure 3g The ridge regression model described is only used to illustrate the cross-well residual statistics and the method for constructing confidence intervals. In practical applications, it can be replaced by gradient boosting trees, random forests, neural networks, or physical fusion models.

[0095] S5. Prediction Results Output, Risk Assessment and Visualization like Figure 4a and Figure 4b As shown, traditional mechanical drilling rate prediction methods typically only provide a single-value prediction curve that varies along the well depth, making it difficult to reflect the impact of multiple uncertainties such as geology, pressure, drill bit and tools, engineering parameters, and model generalization on the pre-drilling mechanical drilling rate prediction results. This invention outputs a more accurate prediction result by using probabilistic modeling of multi-source uncertainty parameters, drilling engineering constraint filtering, and cross-well residual calibration. and Boundary curves and the area between them The invention establishes a reliable interval range and forms a probability profile for pre-drilling prediction of mechanical drilling rate with confidence level. Compared with traditional single-value prediction results, this invention can reflect the interval range and confidence level of mechanical drilling rate prediction results at different well depths, thereby providing a more reliable quantitative basis for pre-drilling drill bit selection, parameter design, and optimization of speed-up schemes.

[0096] Figure 4a and Figure 4b This is a comparison chart of the pre-drilling prediction probability profile with confidence-based mechanical drilling rate, generated based on actual data from the implementation examples. Well A was used as the target well for verification, and the other five wells were used as training wells. The drill bit types and sizes of the historical wells have been stitched together into the depth-by-depth logging samples according to the infeed depth to outfeed depth range. Figure 4a The traditional single-value curve in the figure represents the baseline predicted value of the mechanical drilling rate for each depth window of well A; Figure 4b The P05-P95 confidence interval band is obtained by resampling and propagating the predicted distribution within the well depth window and the logarithmic residuals obtained from leave-one-for-one verification with other historical wells, and then... The result is obtained after inverse transformation. The residuals of well A itself are not used for well A confidence interval calibration, thus simulating pre-drilling application conditions before the target well is encountered. This processing ensures that the pre-drilling prediction probability profile containing confidence-based mechanical drilling rate satisfies non-negativity constraints, avoiding negative results when using the original scale additive residuals. And the result is truncated to 0. Well A retains one unit to verify the root mean square error of 10.97 m / h. Only the designation A is used in the figure; the actual block name and well name are not shown.

[0097] For example, in the evaluation of candidate solutions, if candidate solution A... If the probability is less than 10% and the confidence interval between P05 and P95 is narrow, it can be evaluated as low risk; if candidate scheme B... If the risk is between 10% and 30%, it can be assessed as medium risk; if candidate option C... A candidate scheme with a confidence interval of not less than 50% or a P05-P95 range that is too wide can be evaluated as high risk. Then, the comprehensive score of each candidate scheme is calculated according to Equation (28), and the recommended scheme is determined according to Equation (29).

[0098] It can be seen that this method can not only provide statistical results of mechanical drilling rate at various well depths and target formations, but also provide... , It also includes the P05-P95 confidence interval between the two, and can further calculate the low mechanical drilling rate risk probability and the comprehensive score of candidate schemes, thereby improving the accuracy of the evaluation of the pre-drilling speed scheme and the scientific nature of the decision-making.

Claims

1. A method for pre-drilling prediction of mechanical drilling rate with credibility considering the uncertainty of multi-source information, characterized in that, Includes the following steps: S1. Definition of factors influencing multi-source data integration and mechanical drilling rate: The first The factors influencing the mechanical drilling rate of the nth sample are defined as feature vectors, and the nth sample is used as the feature vector. The target value for each sample is defined as , This refers to the mechanical drilling speed; The factors affecting mechanical drilling speed are classified into five categories according to their sources of uncertainty, namely, multi-source uncertainty parameters; S2. Establishment of a probability distribution model for multi-source uncertainty parameters: The multi-source uncertainty parameters are classified into continuous parameters, discrete parameters, and interval parameters according to their numerical characteristics, and corresponding probability distribution models are established for each. S3. Generation of relevant random drilling scenarios and filtering of drilling engineering constraints: A correlation matrix is ​​established for the aforementioned multi-source uncertainty parameters to generate a stochastic drilling scenario with specified correlations. Drilling engineering constraint filtering is then applied to obtain the final result. Group of random drilling scenarios; S4. Calculation of mechanical drilling rate prediction model including cross-well residual calibration: Training a mechanical drilling rate prediction model; A cross-well residual sample library was established by adopting either a one-well-by-well or a one-block-by-block cross-well verification method. The result obtained in step S3 The initial random drilling scenario is input into the mechanical drilling rate prediction model to obtain the initial... The sample set is obtained by resampling residuals from the cross-well residual sample library and superimposing them onto the prediction results to obtain the calibrated result. Predicted sample set; S5. Output the pre-drilling prediction results of mechanical drilling rate with confidence level: Repeat steps S2 to S4 for each well depth location of the target well to obtain the location of each well depth. Predict the sample set; calculate the values ​​at each well depth. and This forms a probability profile for pre-drilling prediction of mechanical drilling speed with credibility.

2. The method for pre-drilling prediction of mechanical drilling speed with credibility considering the uncertainty of multi-source information according to claim 1, characterized in that, Step S4 also includes The steps to avoid target leakage are as follows: Mechanical specific energy is incorporated into the model input features of the mechanical drilling rate prediction model. As a priori physical characteristic; (16) In equation (16), Drilling pressure, unit: kN; The area of ​​rock broken at the bottom of the well is expressed in m². Rotational speed, in r / min; The torque of the drill bit is expressed in kN·m. This refers to the mechanical drilling rate, measured in m / h; during the model training phase, Use historical well measured mechanical drilling rates; in the prediction phase, The initial mechanical drilling rate estimate is obtained from statistical values, empirical values, or model iterations of adjacent wells in the same layer during the pre-drilling stage; The unit is kPa.

3. The method for pre-drilling prediction of mechanical drilling speed with credibility considering the uncertainty of multi-source information according to claim 1, characterized in that, Step S1, the multi-source uncertainty parameters include: the first category is geological and rock drillability uncertainty parameters; the second category is formation pressure and wellbore pressure environment uncertainty parameters; the third category is drill bit, drill string assembly and speed-up tools uncertainty parameters; the fourth category is drilling engineering parameters and construction execution uncertainty parameters; and the fifth category is data cognition and model generalization uncertainty parameters.

4. The method for pre-drilling prediction of mechanical drilling speed with credibility considering the uncertainty of multi-source information according to claim 1, characterized in that, In step S2, For continuous parameters, the normal distribution, log-normal distribution, triangular distribution, uniform distribution, Beta distribution, or Gamma distribution can be used to describe them; For discrete parameters, a category probability distribution is used to describe them; For interval parameters, a uniform distribution or a triangular distribution is used for description.

5. The method for pre-drilling prediction of mechanical drilling speed with credibility considering the uncertainty of multi-source information according to claim 1, characterized in that, In step S3, the specific method for establishing a correlation matrix for the multi-source uncertainty parameters and generating a stochastic drilling scenario with a specified correlation is as follows: parameter and The correlation coefficient between them is defined as: In equation (9), ( , ) is a parameter and covariance, and These are the standard deviations of the two parameters. The correlation coefficient between the two parameters; The original correlation matrix is ​​composed of all correlation coefficients. ,like If the result is not positive definite, it is made positive definite through eigenvalue correction: In equations (10) and (11), The original correlation matrix, This is the positive definite correlation matrix. The eigenvector matrix, to for eigenvalues, For the number of parameters, Let T be the smallest positive number used to correct non-positive eigenvalues, and let T be the transpose of the matrix. Perform Cholesky decomposition on the positive definite correlation matrix: In equation (11), for The Cholesky decomposition of the lower triangular matrix; Generate independent standard normal random vectors Subsequently, the relevant standard normal random vector for: Then, random samples for each parameter are generated through probability integral transformation: In equations (12) and (13), For the first A group of independent standard normal random vectors, The correlation is the standard normal random vector obtained by the correlation matrix transformation. for The There are 1 component, Φ(·) is the standard normal distribution function. (·) is the first The inverse function of the marginal cumulative distribution function with one parameter. For the generated first The first scene in the group A random sample of parameters; The specific method for drilling engineering constraint filtering is as follows: Let the first Group of random scenarios Engineering inequality constraints are The equality constraint is Then the effective sample discriminant function is: In equation (14), ( ) is the inequality constraint discriminant function. ( ) is the equality constraint discrimination function; it takes the value 1 if all corresponding constraints are satisfied, otherwise it takes the value 0; For the first Effective sample discrimination results for a group of random scenarios; when When =0, the random scene is either discarded or regenerated; The drilling engineering constraint filtering includes drill bit type and permissible lithology matching constraints, upper limit constraints on drilling pressure, upper limit constraints on rotation speed, constraints on the correspondence between displacement and pump pressure, constraints on the applicable range of drilling fluid parameters, constraints on the safe window of wellbore pressure, and constraints on the range of tool efficiency.

6. The method for pre-drilling prediction of mechanical drilling speed with credibility considering the uncertainty of multi-source information according to claim 1, characterized in that, In step S4, The mechanical drilling rate prediction model can employ gradient boosting tree, random forest, neural network, CatBoost, XGBoost, LightGBM, or physical fusion model. The specific method for establishing a cross-well residual sample library using either a one-well-per-well or one-block-per-well verification approach is as follows: Each time, one well or block is used as the validation object, and the remaining wells or blocks are used as training objects. The measured values ​​of the validation object are then calculated. With prediction The residual between; when It is a non-negative variable and the error magnitude varies with When the horizontal changes, the residuals are calculated in the logarithmic space. Each well or block is used as a verification object to complete the calculation once, and the results are summarized to obtain a cross-well residual sample library.

7. The application of the reliable mechanical drilling rate prediction method considering the uncertainty of multi-source information as described in any one of claims 1 to 6 in the field of oil and gas drilling engineering technology.

8. The application of the pre-drilling prediction method for mechanical drilling speed with credibility considering the uncertainty of multi-source information as described in claim 7, characterized in that, The specific steps for evaluating the stability of the pre-drilling plan are as follows: The target segment corresponding to the scheme to be predicted is discretized into multiple well depth positions according to a preset well depth step size. Based on the pre-drilling prediction method for mechanical drilling speed with credibility that considers the uncertainty of multi-source information, the well depth positions are obtained. Predicted sample set; Based on the locations of each well depth Predict the sample set and calculate the interval width, average drilling rate, and standard deviation at each well depth. Based on the interval width, average drilling rate, and standard deviation, the scheme instability index at each well depth is calculated. : In equation (27), The bandwidth of the confidence interval from P05 to P95; This represents the average drilling speed. Standard deviation; , These are non-negative weighting coefficients, and + =1; Based on the well depth step represented by each well depth location, the scheme instability index of each well depth location within the target segment is weighted and summarized to obtain the segment-level scheme instability index of the target segment. The stability of the pre-drilling scheme is characterized by the segment-level scheme instability index.

9. The application of the pre-drilling prediction method for mechanical drilling speed with credibility considering the uncertainty of multi-source information as described in claim 7, characterized in that, This is used to select the optimal solution from multiple candidate solutions, specifically: Based on the aforementioned pre-drilling prediction method for mechanical drilling speed with credibility, which considers the uncertainty of multi-source information, the well depth locations for each prediction scheme are obtained. Predicted sample set; The target segment corresponding to the scheme to be predicted is discretized into multiple well depth positions according to a preset well depth step size. Based on each well depth position... Predict the sample set and calculate the location at each well depth. , , , and ; Locations at different well depths within the same target stratigraphic segment , , , and The indicators are weighted and averaged according to the corresponding well depth step size to obtain the layer level of each candidate scheme. Level Level Level and level ; Based on the aforementioned segment level Level Level Level and level Calculate the comprehensive scoring function for each candidate solution: In equation (28), Number the candidate schemes. For the first The overall score of each candidate solution (·) is the normalization function; , , , and All are the first Each candidate solution corresponds to a segment-level indicator of the target segment; to These are non-negative weight coefficients, and the sum of the weights is 1; Comprehensive scoring function The highest-scoring solution is the optimal solution.

10. The application of the pre-drilling prediction method for mechanical drilling speed with credibility considering the uncertainty of multi-source information as described in claim 9, characterized in that, In equation (28), in, For the minimum economic drilling rate threshold, The target drilling speed threshold.

Images