Method for predicting mechanical drilling speed based on micro-inclination characteristics

Abstract

The application discloses a mechanical drilling speed prediction method based on micro-inclination characteristics, comprising the following steps: acquiring real drilling wellbore logging while drilling and logging data as a first data set; preprocessing the first data set and extracting the best feature and the micro-inclination feature as input features; the preprocessing comprises data cleaning and correlation analysis; training a mechanical drilling speed prediction model by using the input features; and predicting the mechanical drilling speed of a target well by using the trained mechanical drilling speed prediction model. The micro-inclination feature introduced in the application can significantly improve the ROP prediction accuracy of the model. By introducing the micro-inclination feature in the MLP neural network, the model weight is optimized, and the prediction accuracy of the ROP is improved. By introducing the micro-inclination feature in the SVR model and selecting the optimal hyperparameter, the prediction effect is significantly improved. The method provides a new idea for improving drilling parameter prediction and control by using machine learning, and can be applied to the optimization of drilling speed and efficiency.

Description

Technical Field

[0001] This invention relates to the field of oil and gas reservoir development technology, and in particular to a method for predicting mechanical rotation speed based on micro-well inclination characteristics. Background Technology

[0002] As oil and gas exploration and development continue, more and more deep and complex formations are encountered, posing many new challenges to drilling technology. Especially in the drilling of ultra-deep wells, the depth and formation complexity significantly increase the difficulty of drilling, leading to high extraction costs. To ensure high returns and efficiency in oil and gas well development, remote monitoring of the drilling process and optimization of drilling parameters are essential.

[0003] Rate of penetration (ROP) is a key parameter for evaluating drilling efficiency. It refers to the speed at which the drill bit breaks through rock or sediment, usually expressed in meters per hour or feet per hour. It reflects drilling efficiency and formation characteristics. Improving ROP can save drilling time and costs, while also reducing drill string wear and damage. Therefore, improving the prediction and control of ROP is crucial for optimizing drilling. Traditional empirical models struggle to accurately describe the nonlinear relationships between various complex factors affecting ROP. However, the development of machine learning technology offers new insights into using data to train models for ROP prediction.

[0004] Compared to traditional models, machine learning models can achieve faster and more accurate nonlinear predictions. However, existing methods for predicting ROP using machine learning still have room for improvement in accuracy, as they do not yet consider the impact of slight wellbore inclination above the drill bit (micro-inclination) on ROP. When the drill bit encounters micro-inclination, it will vibrate, which in turn affects ROP. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a mechanical drilling rate prediction method that introduces novel features reflecting micro-wellbore deviation information above the drill bit into a machine learning model to improve the prediction accuracy of ROP. Specifically, it includes the following technical solutions:

[0006] Mechanical drilling rate prediction methods based on micro-well deviation characteristics include:

[0007] Obtain logging-while-drilling and well logging data from the actual drilled wellbore as the first dataset;

[0008] The first dataset is preprocessed, and the best features and micro-well deviation features are extracted as input features; the preprocessing includes data cleaning and correlation analysis.

[0009] The input features are used to train the mechanical drilling rate prediction model;

[0010] The mechanical drilling rate of the target well is predicted using a trained mechanical drilling rate prediction model.

[0011] In some preferred embodiments, the preprocessing method includes:

[0012] Delete the rows containing missing values ​​in the first dataset and process them using a smoothing filter;

[0013] Pearson correlation coefficient quantitative analysis was performed on the first dataset after smoothing and filtering to remove features with low absolute values ​​of Pearson correlation coefficient.

[0014] Using an exhaustive search method, the remaining features are trained and evaluated through a multilayer perceptron model, and the best features are selected based on the evaluation results.

[0015] In some preferred embodiments, the optimal features are drilling pressure, torque, rotary drilling speed, inlet temperature, inlet conductivity, outlet conductivity, mud density, DC index, and wellbore curvature.

[0016] In some preferred embodiments, the micro-inclination feature characterizes the trend of change in the curvature of a certain section of the wellbore above the drill bit.

[0017] In some preferred embodiments, the method for extracting the micro-well deviation features includes:

[0018] Calculate the rate of change of borehole curvature at each well depth, and superimpose the rate of change of borehole curvature within a specified distance from the drill bit to above the drill bit to form the micro-inclination characteristics within that distance range.

[0019] In some preferred embodiments, the mechanical drilling rate prediction model includes a multilayer perceptron with initial stage weight parameters and bias parameters optimized by a genetic algorithm.

[0020] In some preferred embodiments, the mechanical drilling rate prediction model includes a support vector regression model with optimized hyperparameter combinations optimized by a network search algorithm.

[0021] Beneficial effects

[0022] The present invention proposes an approach that introduces micro-wellbore deviation features to significantly improve the ROP prediction accuracy of the model. By introducing micro-wellbore deviation features into the MLP neural network, the model weights are optimized, thus improving the accuracy of ROP prediction. Furthermore, by introducing micro-wellbore deviation features into the SVR model and selecting the optimal hyperparameters, the prediction performance is significantly enhanced. The proposed method provides a new approach to improving drilling parameter prediction and control using machine learning, and can be extended to the optimization of drilling speed and efficiency. Attached Figure Description

[0023] Figure 1This is a schematic diagram of the mechanical drilling rate prediction method based on micro-well deviation characteristics in a preferred embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of a method for preprocessing a first dataset in a preferred embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the micro-well inclination feature in a preferred embodiment of the present invention;

[0026] Figure 4 This is a graph showing the MSE variation of the MLP-GA model under different micro-well deviation characteristics in the experimental examples of this invention;

[0027] Figure 5 This is a scatter plot of the predicted values ​​and field measurements of the MLP-GA model under different micro-well deviation characteristics in the experimental examples of this invention.

[0028] Figure 6 This is a scatter plot of the predicted values ​​and field measurements of the SVR-GS model under different micro-well deviation characteristics in the experimental examples of this invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings. In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention.

[0030] Example 1

[0031] like Figure 1 As shown, this embodiment provides a mechanical drilling rate prediction method based on micro-well deviation characteristics, including:

[0032] S1. Obtain logging-while-drilling (LOD) and well logging data from the actual drilled well as the first dataset. During drilling, ROP is easily affected by many features, such as drilling pressure, torque, rotational speed, drill bit structure, flow rate, and formation drillability index (DC). In some preferred embodiments, 16 features are selected during the sample set preparation stage, with ROP as the target feature and the others as pre-input features. To build a predictive model, a comprehensive logging dataset of the drilling process needs to be collected as a sample. In specific implementations, logging data from a certain section of the actual well (e.g., above 1000 meters) can be selected. This dataset should cover various states during the drilling process and include key parameters affecting ROP. Simultaneously, it is necessary to ensure that the selected well section uses the same type of PDC drill bit to reduce the impact of drill bit structure variations. To obtain sample data diversity, the sample size needs to be in the thousands, with a target of at least over 2000 samples.

[0033] S2. Preprocess the first dataset and extract the best features and micro-well deviation features as input features; the preprocessing includes data cleaning and correlation analysis.

[0034] S3. Train the mechanical drilling rate prediction model using the input features; select a machine learning model suitable for handling complex mapping relationships to obtain more accurate ROP predictions. Optimize the model's structure and parameters using methods such as grid search and genetic algorithms. Train the ROP prediction model using a preprocessed sample dataset containing multiple features.

[0035] S4. Predict the mechanical rate of drilling (ROP) of the target well using the trained mechanical rate of drilling prediction model. Use the trained model to predict the drilling rate based on the logging data of the new well, and output the drilling speed results. Compared to traditional empirical models, this model can achieve more accurate and efficient monitoring of the drilling process and ROP prediction, providing important guidance for drilling parameter optimization.

[0036] Example 2

[0037] like Figure 2 As shown, this embodiment is based on Embodiment 1 above. Considering that the logging-while-drilling and well logging data obtained from actual drilled wells contain a large amount of noise, including the deletion of rows with missing values ​​and outliers, and that the original first dataset is outlier data, this type of data has a significant impact on the performance of the neural model in subsequent steps. Therefore, this embodiment improves the robustness of the prediction process to noise through data processing and model tuning, enabling the final model to maintain good prediction accuracy in complex noise environments. Thus, this embodiment provides a specific method for preprocessing the first dataset, including:

[0038] S201. Rows containing missing values ​​in the first dataset were deleted, and a smoothing filter was applied. To address noise in the first dataset, two techniques were used: data cleaning and smoothing filtering. Data cleaning initially removed some outliers by deleting rows with missing values. However, the remaining outliers in the dataset could significantly negatively impact subsequent modeling. To further improve data quality, a Savitzky-Golay (SG) filter was introduced for smoothing.

[0039] The working principle of an SG filter is to smooth local segments of a signal using a polynomial fitting method. Typically, a second-order or higher polynomial is chosen to eliminate high-frequency noise while maintaining the overall shape of the signal. Compared to methods such as simple moving averages, the advantage of SG filtering is that it can better preserve the details of signal characteristics, such as edges and peaks, while smoothing the signal.

[0040] To achieve optimal smoothing, SG filtering requires selecting appropriate filter window width and polynomial order. Based on experience and testing, when processing well logging signal data, choosing a 5th-order polynomial and 17 sample points as a fitting window can significantly reduce noise while maintaining signal integrity. The first dataset processed by the SG filter shows a significantly improved signal-to-noise ratio, is stable and reliable, and is more suitable for subsequent modeling and analysis.

[0041] The SG filtering approach can also be extended to other similar data preprocessing scenarios, maximizing the preservation of effective information in the sample data while reducing noise. In summary, SG filtering provides a simple and effective technique for improving data quality, making the model trained on the dataset more robust and reliable.

[0042] S202. Perform Pearson correlation coefficient quantitative analysis on the first dataset after smoothing and filtering, and remove features with low absolute values ​​of Pearson correlation coefficients. After smoothing and filtering the first dataset, feature selection is required to determine which features have a stronger correlation with the target value, Rate of Exploration (ROP), and use them as input features for the model. The Pearson correlation coefficient can be used to quantitatively analyze the degree of correlation between each feature and ROP.

[0043] The Pearson correlation coefficient ranges from -1 to 1, with a larger absolute value indicating a stronger linear correlation between the two variables. Calculate the Pearson coefficient between each candidate feature and the ROP, and take the absolute value. This allows for a direct comparison of the correlation strength between different features and the ROP.

[0044] Generally, a coefficient absolute value above 0.5 indicates a moderate to high correlation. Therefore, features with weak correlation to ROP can be filtered out and removed, retaining only the features with strong correlation.

[0045] Furthermore, different correlation coefficient thresholds can be set, different numbers of features can be selected, and the performance of the models corresponding to these feature combinations can be compared to find the optimal subset of input features. Pearson correlation analysis provides a simple and effective method for feature selection, making the model input more focused and thus improving the performance of ROP prediction.

[0046] S203. Using an exhaustive search method, the remaining features are trained and evaluated through a multilayer perceptron model, and the best features are selected based on the evaluation results. After obtaining the preprocessed feature set with strong correlations, it is necessary to further select the feature subset that best predicts the target value's mechanical drilling rate to prevent excessive redundant features from negatively impacting the model.

[0047] Here, an exhaustive method is used for feature selection. First, assume that there are n features in the feature set. Then, iterate through all possible combinations containing 1 to n features, construct a multilayer perceptron model for each feature combination, and evaluate its prediction performance on the validation set.

[0048] Typical evaluation metrics include mean squared error, mean absolute error, and R². The evaluation results for each feature combination are recorded. Finally, the subset of features with the best prediction performance (i.e., the smallest evaluation metric value) is selected from all combinations as the final input features of the model.

[0049] Compared to simple step-by-step selection methods, exhaustive search can systematically test different feature combinations, avoiding the omission of feature combinations that would otherwise perform well. Furthermore, multiple high-performing feature sets can be selected based on the exhaustive search results to build model ensembles and further improve predictive performance.

[0050] In summary, exhaustive search provides a systematic and effective feature selection method that can be used to obtain the optimal feature subset for a given problem, thereby maximizing the predictive performance of machine learning models.

[0051] It should be understood that standardizing the preprocessing of all data can preserve the distribution of the original data while eliminating the influence of different units on data training, so that the algorithm can effectively identify and learn each parameter.

[0052] Example 3

[0053] This embodiment is based on Embodiment 2 described above. It should be understood that there are many features reflecting drilling conditions; directly using all features to train the model would lead to redundancy and overfitting. Therefore, this embodiment selected 16 features as pre-input features during the sample set preparation stage. After the correlation analysis using the Pearson correlation coefficient from Embodiment 2, the three features with the weakest correlation to ROP were removed: outlet temperature, riser pressure, and outlet flow rate.

[0054] From the remaining 12 features, a specified number of features were selected and trained using a multilayer perceptron model, which was then evaluated on a test set. Based on the evaluation results, nine features were selected as the optimal input features: drilling pressure, torque, rotary table speed, inlet temperature, inlet conductivity, outlet conductivity, mud density, DC index, and wellbore curvature.

[0055] Example 4

[0056] This embodiment is based on the above embodiment 1.

[0057] Those skilled in the art will know that in oil and gas drilling, variations in formation hardness can cause numerous minute inclinations in the wellbore above the drill bit, known as micro-inclination. This is because in transitional regions with significant differences in formation hardness, the drill bit's direction becomes difficult to control, resulting in slight deflections. The presence of these micro-inclinations can cause significant lateral vibrations in the drill bit during drilling. In severe cases, this can lead to dynamic problems and a decrease in drilling speed. However, traditional mechanical drilling speed prediction models rarely consider this influencing factor. Therefore, this embodiment incorporates micro-inclination features into the machine learning model to improve the model's ability to predict mechanical drilling speed and to explore the influence of micro-inclination features on mechanical drilling speed.

[0058] like Figure 3 As shown, in a certain well depth d ( Figure 2 When the wellbore trajectory is a straight line, the wellbore curvature C is 0; when the wellbore trajectory is a smooth curve, the wellbore curvature is a constant A; when there are a large number of micro-inclinations, the wellbore curvature fluctuates, and at this time the wellbore curvature C is a function of the well depth d. Therefore, this embodiment considers using micro-inclination characteristics to characterize the changing trend of wellbore curvature in a certain section above the drill bit.

[0059] In some preferred embodiments, a preferred method for extracting micro-well deviation features is provided, including:

[0060] Calculate the rate of change of borehole curvature at each well depth, and superimpose the rate of change of borehole curvature within a specified distance from the drill bit to above the drill bit to form the micro-inclination characteristics within that distance range.

[0061] Taking a certain well depth d as an example, in the feature data preparation stage, the micro-inclination feature at each well depth is first represented by the rate of change of the wellbore curvature. Then, the micro-inclination features at 50m, 100m, 150m, 200m and 400m above the drill bit are superimposed as input features for the machine learning model.

[0062] Example 5

[0063] This embodiment is based on Embodiment 1 above. This embodiment provides a preferred choice for a mechanical drilling rate prediction model. The mechanical drilling rate prediction model includes a multi-layer perceptron (MLP) model whose initial stage weight parameters and bias parameters are optimized using a genetic algorithm (GA), i.e., the MLP-GA model.

[0064] Those skilled in the art will know that the MLP is a commonly used artificial neural network, consisting of an input layer, an output layer, and several hidden layers, and can be used for tasks such as classification, regression, and clustering. GA is an optimization algorithm that simulates the natural evolutionary process. Through operations such as selection, crossover, and mutation, it continuously updates a set of candidate solutions (called a population) to find the optimal or near-optimal solution. The main idea of ​​the MLP-GA model in this embodiment is to use the weights and biases of the MLP as chromosome encoding, and the prediction error or accuracy of the MLP as a fitness function. GA is then used to search for the optimal or near-optimal chromosome, thereby obtaining the optimal or near-optimal MLP parameters.

[0065] The advantages of the MLP-GA model are that it can avoid getting trapped in local optima, can adaptively adjust the network structure, and can handle nonlinear and high-dimensional problems. The disadvantages of the MLP-GA model are that it has a large computational cost, a slow convergence speed, and requires setting multiple parameters (such as population size, crossover probability, mutation probability, etc.).

[0066] In this embodiment, a genetic algorithm is used to optimize the weights w and biases b of the neural network in the initial stage during the training process of the MLP model, thereby reducing the influence of local extrema and improving the convergence speed of the neural network.

[0067] Example 6

[0068] This embodiment is based on Embodiment 1 above. This embodiment presents a better choice of another mechanical drilling rate prediction model. The mechanical drilling rate prediction model includes a Support Vector Regression (SVR) model with optimal hyperparameter combination optimized by a Grid Search (GS) algorithm, namely the SVR-GS model.

[0069] SVR is a regression analysis method based on Support Vector Machines. It uses kernel functions to map the input space to a high-dimensional feature space and uses a linear model to achieve nonlinear regression. GS is a method that iterates through different combinations of control elements to adjust the model. Its advantages are that it can avoid getting trapped in local optima, can adaptively adjust the network structure, and can handle nonlinear and high-dimensional problems.

[0070] Experimental Example

[0071] This experiment uses logging-while-drilling (LOD) and well logging data from a field-drilled well in Sichuan, China, from 1023m to 3761m, as the test set. The sample size is 2739, and the same PDC drill bit was used for this well section. The SVR-GS model and MLP-GA model were used as mechanical drilling rate prediction models, respectively. Nine features were selected as optimal input features: drilling pressure, torque, rotary table speed, inlet temperature, inlet conductivity, outlet conductivity, mud density, DC index, and wellbore curvature. The effects of introducing micro-inclination features at different distances on the model performance before and after the introduction of micro-inclination features were investigated.

[0072] like Figure 4 As shown, for MLP-GA, the MSE decreases by 0.52 after introducing micro-well deviation features, with a decrease rate of 20.97%.

[0073] Furthermore, the measured ROP value obtained on-site was compared with the predicted value obtained by the method of this invention, and the results are as follows: Figure 5 As shown, after incorporating micro-wellbore deviation features at 0m, 50m, 100m, 150m, 200m, and 400m above the drill bit, and completing 2000 iterations, the R² values ​​of the MLP-GA models on the test set were 0.891, 0.903, 0.917, 0.910, 0.902, and 0.899, respectively. The data shows that MLP-GA itself has high accuracy; after incorporating the micro-wellbore deviation feature at 100m above the drill bit, the coefficient of determination further increased from 0.891 to 0.917, an improvement of 2.6 percentage points.

[0074] The results of introducing micro-well deviation features during the training of the MLP-GA model show that micro-well deviation has a significant impact on ROP, and the micro-well deviation features within 100m above the drill bit have the greatest impact on ROP.

[0075] like Figure 6 As shown, for the SVR-GS model, after introducing the micro-inclination feature 100m above the drill bit, the MSE of the SVR-GS model on the test set decreased from 2.01 to 1.45, a decrease of 0.56, or 27.86%; the coefficient of determination R2 on the test set increased from 0.909 to 0.941, an increase of 3.2 percentage points, indicating a significant improvement in the model's predictive ability.

[0076] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting mechanical drilling rate based on micro-well deviation characteristics, characterized in that, include: Obtain logging-while-drilling and well logging data from the actual drilled wellbore as the first dataset; The first dataset is preprocessed, and the best features and micro-well deviation features are extracted as input features; The preprocessing includes data cleaning and correlation analysis; The input features are used to train the mechanical drilling rate prediction model; The mechanical drilling rate of the target well is predicted using a trained mechanical drilling rate prediction model. The micro-well deviation feature characterizes the changing trend of the wellbore curvature in a certain section above the drill bit; The method for extracting the micro-well deviation features includes: Calculate the rate of change of borehole curvature at each well depth, and superimpose the rate of change of borehole curvature within a specified distance from the drill bit to above the drill bit to form the micro-inclination characteristics within that distance range.

2. The mechanical drilling rate prediction method based on micro-well deviation characteristics as described in claim 1, characterized in that: The preprocessing method includes: Delete the rows containing missing values ​​in the first dataset and process them using a smoothing filter; Pearson correlation coefficient quantitative analysis was performed on the first dataset after smoothing and filtering to remove features with low absolute values ​​of Pearson correlation coefficient. Using an exhaustive search method, the remaining features are trained and evaluated through a multilayer perceptron model, and the best features are selected based on the evaluation results.

3. The mechanical drilling rate prediction method based on micro-well deviation characteristics as described in claim 2, characterized in that: The optimal characteristics are drilling pressure, torque, rotary table speed, inlet temperature, inlet conductivity, outlet conductivity, mud density, DC index, and wellbore curvature.

4. The mechanical drilling rate prediction method based on micro-well deviation characteristics as described in claim 1, characterized in that: The mechanical drilling rate prediction model includes a multilayer perceptron with initial stage weight parameters and bias parameters optimized by a genetic algorithm.

5. The mechanical drilling rate prediction method based on micro-well deviation characteristics as described in claim 1, characterized in that: The mechanical drilling rate prediction model includes a support vector regression model with optimized hyperparameter combinations through a network search algorithm.

Images