Drilling rate prediction method and device based on artificial intelligence
By adopting an artificial intelligence-based drilling rate prediction method in offshore oil and gas mining, and using GAN enhanced training data and integrated learning algorithms to fusion the model, the accuracy of drilling rate prediction is solved and more efficient and economical drilling operations are achieved.
Patent Information
- Application Number
- CN202411951228.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-05-13
AI Technical Summary
In offshore oil and gas mining, accurate prediction of drilling rate is of great significance, but the existing technology is difficult to effectively solve this problem, resulting in increased drilling costs and extended operating time.
Using an artificial intelligence-based drilling rate prediction method, drilling samples are enhanced by generating adversarial networks (GANs), training data sets are augmented, and drilling penetration prediction models are constructed using classic machine learning algorithms. The method also includes using an integrated learning algorithm for model fusion and local interpretability analysis to improve the transparency and prediction accuracy of the model.
Through this method, the accuracy and robustness of drilling rate prediction can be significantly improved, drilling costs, non-production time, and drilling safety can be improved.
Smart Images
Figure CN119990399A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of energy extraction technology, and in particular, to a drilling rate prediction method and device based on artificial intelligence. Background Art
[0002] As global onshore oil and natural gas resources gradually decrease, the field of energy extraction has paid more attention to offshore areas. The share of offshore extraction in the total oil and gas extraction continues to increase. In order to meet the demand for offshore oil and gas extraction, the construction of drilling platforms has become inevitable. However, the space of drilling platforms is relatively limited, which brings many challenges.
[0003] In this case, cluster well drilling technology is widely used. This technology can connect multiple different reservoirs by drilling in different directions, effectively reducing the frequency of repositioning the platform due to the exploitation of different reservoirs, and minimizing the impact on the surrounding environment during the exploitation process. In offshore oil and gas development operations, ROP is always a core concern. This is because the offshore operating area is remote, far away from land infrastructure and material supply points, which greatly increases the cost of offshore drilling. Once the drilling speed is not properly controlled, it will not only extend the operation time, but also cause the cost to rise sharply.
[0004] Therefore, accurate prediction of ROP is extremely important, as it is a key link in ensuring efficient and economical offshore oil and gas development. Summary of the invention
[0005] In order to solve the problem of how to accurately predict ROP, the present application provides a drilling rate prediction method and device based on artificial intelligence.
[0006] The embodiment of the present application is implemented as follows:
[0007] In a first aspect, the present application provides a drilling rate prediction method based on artificial intelligence, comprising:
[0008] The drilling samples are enhanced by generative adversarial network method to expand the training data set;
[0009] Based on the training data set, generating an optimal drilling penetration rate prediction model;
[0010] The optimal borehole penetration prediction model is analyzed for interpretability using a locally interpretable model-independent explanation algorithm to improve the transparency and understandability of the model.
[0011] In a possible implementation, the method of enhancing the drilling samples by using a generative adversarial network method to expand the training data set further includes:
[0012] Based on daily drilling reports, calculate parameters covering multiple aspects, including logging parameters, drilling parameters, well trajectory parameters, and drill bit parameters;
[0013] Generative adversarial network technology is used to generate additional training samples. Generative adversarial network technology can generate new and diverse data samples by learning the distribution characteristics of existing data, thereby effectively expanding and enhancing the training set.
[0014] In a possible implementation, generating an optimal drilling penetration rate prediction model based on the training data set further includes:
[0015] Based on the training data set, a drilling penetration rate prediction model is constructed using a variety of classic machine learning algorithms;
[0016] 0%, 1.7%, and 5.1% noise data were added to the training set, and the prediction accuracy and robustness under different noise levels were compared to evaluate the performance of these models under different noise conditions.
[0017] Screen out models with high prediction accuracy and strong robustness;
[0018] Use ensemble learning algorithms to fuse these models and create multiple fusion models;
[0019] A plurality of the fusion models are evaluated to determine an optimal drilling penetration rate prediction model.
[0020] In a possible implementation, the using of an ensemble learning algorithm to fuse these models to create multiple fusion models further includes:
[0021] Set the selected model as the base learner;
[0022] The LR model is selected as the meta-learner;
[0023] In this way, multiple fusion models are constructed.
[0024] In one possible implementation, multiple classic machine learning methods include decision tree supervised learning algorithm, random forest ensemble learning algorithm, lightweight gradient boosting algorithm, extreme gradient boosting algorithm, support vector regression algorithm and back propagation neural network.
[0025] In a possible implementation, the generative adversarial network includes a generator and a discriminator;
[0026] The goal of the generator is to generate realistic data samples from random noise, and the goal of the discriminator is to distinguish between the generated samples and the real samples;
[0027] The two compete with each other during the training process. The generator strives to generate more realistic samples to deceive the discriminator, while the discriminator strives to improve its ability to distinguish true and false samples. Through this adversarial mechanism, the generator is eventually able to generate high-quality data samples.
[0028] In a possible implementation, the drilling penetration rate is the depth of drilling by the drill bit per unit time, expressed in meters per hour (m / h), which is a key indicator for measuring drilling speed and affects the efficiency and cost of drilling operations;
[0029] The drilling penetration rate is affected by many factors, including drill bit type, drilling pressure, rotation speed, drilling fluid properties, and formation characteristics;
[0030] During the drilling process, optimizing the borehole penetration rate can increase the drilling speed, reduce the drilling cost, reduce the non-productive time and improve the drilling safety.
[0031] In one possible implementation, the ensemble learning method improves the overall prediction performance by combining the prediction results of multiple different models, which includes two main levels:
[0032] The first layer is the base learner, which is a machine learning algorithm, including decision trees, support vector machines, and neural networks;
[0033] The second layer is the meta-learner, which is a model, such as logistic regression, that learns how to most effectively combine the predictions of the base learners.
[0034] The ensemble learning method divides the training set multiple times through cross-validation. Each base learner trains and predicts the divided data, and then uses these prediction results as new features for the meta-learner to make the final prediction.
[0035] In a second aspect, the present application provides a drilling rate prediction device based on artificial intelligence, comprising:
[0036] Training set enhancement module: used to enhance drilling samples through the generative adversarial network method to expand the training data set;
[0037] Optimal model generation module: used to generate an optimal drilling penetration rate prediction model based on the training data set;
[0038] Optimal model analysis module: It is used to perform interpretability analysis on the optimal drilling penetration prediction model using a locally interpretable model-independent interpretation algorithm to improve the transparency and understandability of the model.
[0039] In a possible implementation, the model generation module further includes:
[0040] Model building module: used to build a drilling penetration rate prediction model based on the training data set using a variety of classic machine learning algorithms;
[0041] Noise addition module: used to add 0%, 1.7% and 5.1% noise data to the training set, and compare its prediction accuracy and robustness under different noise levels to evaluate the performance of these models under different noise conditions;
[0042] Model screening module: screen out models with high prediction accuracy and strong robustness;
[0043] Model fusion module: Use ensemble learning algorithms to fuse these models and create multiple fusion models;
[0044] Model evaluation module: evaluates the multiple fusion models to determine the optimal drilling penetration rate prediction model.
[0045] The technical solution provided by this application can at least achieve the following beneficial effects:
[0046] The present application provides a drilling rate prediction method and device based on artificial intelligence, which enhances drilling samples through the GAN method to expand the training data set. Based on these data, a ROP prediction model was constructed using a classic machine learning algorithm. In order to evaluate the performance of these models under different noise conditions, 0%, 1.7% and 5.1% noise data were added to the training set, and their prediction accuracy and robustness under different noise levels were compared. Then, models with high prediction accuracy and strong robustness were screened out, and these models were fused using the Stacking algorithm to create multiple fusion models. Subsequently, these fusion models were evaluated to determine the best ROP prediction model. Finally, the LIME algorithm was used to perform an interpretability analysis on the optimal ROP prediction model to improve the transparency and understandability of the model, thereby improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0048] Figure 1 It is a flowchart of a drilling rate prediction method based on artificial intelligence shown in an exemplary embodiment of the present application;
[0049] Figure 2It is a schematic diagram of a specific implementation process of a prediction method shown in an exemplary embodiment of the present application;
[0050] Figure 3 It is a structural schematic diagram of a drilling rate prediction device based on artificial intelligence shown in an exemplary embodiment of the present application.
[0051] Reference numerals:
[0052] 1. Training set enhancement module; 2. Optimal model generation module; 3. Optimal model analysis module. DETAILED DESCRIPTION
[0053] In order to make the purpose, implementation mode and advantages of the present application clearer, the exemplary implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, not all of the embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.
[0054] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their ordinary and common meanings.
[0055] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.
[0056] The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.
[0057] To facilitate the technical solution of the application, some concepts involved in the application are first explained below.
[0058] GAN: GAN (Generative Adversarial Network), or Generative Adversarial Network, is a deep learning model composed of two neural networks: Generator and Discriminator. The goal of the generator is to generate realistic data samples from random noise, while the goal of the discriminator is to distinguish between generated samples and real samples. The two networks compete with each other during training: the generator strives to generate more realistic samples to deceive the discriminator, while the discriminator strives to improve its ability to distinguish between real and fake samples. Through this adversarial mechanism, the generator is eventually able to generate high-quality data samples.
[0059] ROP: ROP (Rate of Penetration), or borehole penetration rate, is an important parameter in drilling engineering. It refers to the depth of the drill bit in a unit of time, usually expressed in meters per hour (m / h). ROP is a key indicator to measure drilling speed, which directly affects the efficiency and cost of drilling operations. The level of ROP is affected by many factors, including drill bit type, drilling pressure, rotation speed, drilling fluid performance, formation characteristics, etc. In the drilling process, optimizing ROP is of great significance to increasing drilling speed, reducing drilling costs, reducing non-productive time, and improving drilling safety. Therefore, accurate prediction of ROP is crucial for drilling project management and decision-making.
[0060] Decision Tree: Decision Tree (DT) is a commonly used supervised learning algorithm for classification and regression tasks. It builds a tree-like model by learning patterns in the training data set, which can be used to predict the category or continuous value of new data. Each internal node of the decision tree represents a test on a feature, each branch represents the result of the test, and each leaf node represents the final decision or classification label. The learning process of the decision tree includes feature selection, tree generation and pruning, aiming to find the best features and split points to maximize the distinction between categories. Its advantages are that the model is highly interpretable, the output results are easy to understand, and the data preparation requirements are relatively low. However, decision trees also have some disadvantages, such as easy overfitting, so in practical applications, it is often necessary to use methods such as pruning to improve the generalization ability of the model.
[0061] Random Forest: Random Forest (RF) is an ensemble learning algorithm consisting of multiple decision trees for classification and regression tasks. Its core idea is to improve the accuracy and robustness of predictions by building multiple decision trees and aggregating their results. Each decision tree of the random forest is trained on a different subset of the data set, which is obtained by sampling with replacement from the original data set. When building each decision tree, the random forest also randomly selects a part of the features for splitting, increasing the diversity of the model. The final prediction result is determined by the prediction results of all decision trees by voting or averaging. This method can effectively reduce the risk of overfitting and improve the generalization ability of the model to new data. The advantages of random forests include high accuracy, ability to handle a large number of input variables, evaluation of variable importance, and good tolerance to outliers and noise. In addition, random forests have good resistance to overfitting because it combines the results of multiple decision trees by averaging or voting, thereby reducing the variance of the model.
[0062] Light Gradient Boosting Machine: Light Gradient Boosting Machine (LGBM) is an efficient algorithm for machine learning that makes predictions by combining multiple simple decision trees. LGBM is very fast and flexible, can handle large amounts of data, and has relatively low requirements for computing resources. It is often used for classification and prediction tasks and has performed well in many data science competitions. In short, LGBM is a fast and accurate prediction model.
[0063] Extreme Gradient Boosting: eXtreme Gradient Boosting (XGB) is an efficient gradient boosting algorithm proposed by Tianqi Chen in 2014. It is based on decision tree learning, and improves the accuracy of prediction by building multiple weak prediction models (decision trees) and combining them. XGBoost is optimized on the basis of the original GBDT (gradient boosted decision tree), adding regularization terms to avoid overfitting and improving computational efficiency. It supports parallel processing and distributed computing, making it more efficient when processing large-scale data sets. XGBoost also features the ability to handle missing values, as well as the ability to customize optimization objectives and evaluation criteria. Due to its excellent performance and flexibility, XGBoost has been widely used in the fields of data science and machine learning.
[0064] Support Vector Regression: Support Vector Regression (SVR) is a regression algorithm based on Support Vector Machine (SVM) for solving regression problems. It works by finding an optimal hyperplane in a high-dimensional space so that the data has the best linear regression features in this space. The core idea of SVR is to minimize a loss function that includes an ε-insensitive loss term and a regularization term. The ε-insensitive loss term allows the model to not penalize the difference between the predicted value and the true value within a certain error range, while the regularization term is used to prevent the model from overfitting. By using kernel functions, SVR can handle nonlinear relationships and map data to high-dimensional space, thereby finding a better regression hyperplane. This algorithm has applications in many fields, such as financial market analysis, time series forecasting, etc.
[0065] Back-propagation Neural Network: Back-propagation Neural Network (BPNN) is a multi-layer feed-forward neural network that is trained using a back-propagation algorithm. This network structure usually includes an input layer, one or more hidden layers, and an output layer. The main feature of BPNN is that its learning process involves two stages: forward propagation and back-propagation. In the forward propagation stage, the input data is passed through each layer of neurons and processed to produce an output. If the prediction at the output layer differs from the expected result, in the back-propagation stage, the error is passed back to the network and reduced by adjusting the connection weights between neurons. BPNN is able to learn complex nonlinear relationships between input and output, so it has a wide range of applications in pattern recognition, classification, prediction, and other fields. However, the training of BPNN may take a long time and there is a risk of falling into a local optimal solution. Despite this, BPNN is still a very popular tool in the field of machine learning and artificial intelligence due to its flexibility and effectiveness.
[0066] Linear Regression: Linear Regression (LR) is a basic statistical method used to model and analyze the relationship between two variables: a dependent variable and one or more independent variables. The purpose of linear regression is to find a straight line (in two-dimensional space) or a hyperplane (in multidimensional space) that can be used to predict the output value. The linear regression model assumes that there is a linear relationship between the input features and the output values. The model determines the best position of this line by learning the weights (coefficients) of the features so that the difference between the predicted value and the actual value (usually expressed as squared error) is minimized. Linear regression can be simple linear regression (one independent variable) or multiple linear regression (multiple independent variables). It is widely used in fields such as forecasting, trend analysis, and causal relationship research. Although the linear regression model is relatively simple, it is very effective in many practical applications and is the basis of many more complex machine learning algorithms.
[0067] Stacking algorithm: Stacking algorithm is an integrated learning method that improves the overall prediction performance by combining the prediction results of multiple different models. This method contains two main levels: the first layer is the base learner, which can be any machine learning algorithm, such as decision tree, support vector machine, neural network, etc.; the second layer is the meta-learner, which is usually a simple model, such as logistic regression, and its task is to learn how to most effectively combine the prediction results of the base learners. Stacking divides the training set multiple times through cross-validation. Each base learner trains and predicts the divided data, and then uses these prediction results as new features, and the meta-learner makes the final prediction. The advantage of this method is that it can fully utilize the advantages of each base learner and improve the generalization ability of the model.
[0068] Local Interpretable Model-agnostic Explanations: Local Interpretable Model-agnostic Explanations (LIME) is a method for explaining the prediction results of complex machine learning models. It approximates the predictions of complex models using a simple interpretable model (such as linear regression) in the local neighborhood of a specific sample. The core idea of LIME is to generate perturbation samples around the sample to be explained, assign weights to these samples according to their similarity to the original sample, and then train a local interpretable model to explain the predictions of the black box model. This method can reveal the contribution of each input feature to the model's prediction results, thereby providing insights into the model's decision-making process. The characteristic of LIME is that it does not depend on a specific model and can be applied to any type of predictive model, including text, images, and tabular data. Its application can help users understand the model's predictions and increase their trust in the model.
[0069] Before explaining the artificial intelligence-based drilling rate prediction method provided in the embodiment of the present application, the application scenario and implementation environment of the embodiment of the present application are first introduced.
[0070] As global onshore oil and natural gas resources gradually decrease, the field of energy extraction has paid more attention to offshore areas. The share of offshore extraction in the total oil and gas extraction continues to increase. In order to meet the demand for offshore oil and gas extraction, the construction of drilling platforms has become inevitable. However, the space of drilling platforms is relatively limited, which brings many challenges.
[0071] In this case, cluster well drilling technology is widely used. This technology can connect multiple different reservoirs by drilling in different directions, effectively reducing the frequency of platform repositioning due to exploitation of different reservoirs, and minimizing the impact on the surrounding environment during the mining process.
[0072] In offshore oil and gas development operations, ROP is always the core concern. This is because the offshore operation area is remote, far away from land infrastructure and material supply points, which greatly increases the cost of offshore drilling. Once the drilling speed is not properly controlled, it will not only extend the operation time, but also cause the cost to rise sharply. Therefore, accurate prediction of ROP is extremely critical. It is a key link to ensure efficient and economical offshore oil and gas development.
[0073] In the field of ROP prediction, traditional methods mainly rely on theoretical models and physical experiments to carry out related work. However, with the continuous advancement of artificial intelligence technology, machine learning has shown extraordinary potential in solving complex engineering problems. In particular, in recent years, machine learning has been regarded as a promising tool in the field of ROP prediction due to its powerful data processing capabilities. Many studies have tried to apply classic machine learning algorithms to ROP prediction in order to improve the accuracy of prediction.
[0074] At present, the technologies for ROP prediction include theoretical models, physical experiments, and machine learning, but their shortcomings are as follows:
[0075] (1) Theoretical models and physical experiments: ROP prediction is extremely complex because it involves many independent parameters, such as geotechnical properties, geological conditions, operating settings, and machine characteristics. This complexity makes it challenging to accurately present ROP in the form of a mathematical function of certain variables. On the one hand, the construction of theoretical equations usually requires supplementary data and empirical coefficients, and these data and coefficients will change with different drilling formations. On the other hand, traditional ROP theoretical models involve complex mathematical calculations, which often make researchers focus only on parameters that have a significant impact on ROP. Furthermore, dynamic factors such as fluid pressure changes, formation changes, and energy transmission losses make the actual drilling process full of variables, and traditional theoretical models and physical experiments often find it difficult to accurately capture these changes, resulting in the prediction results being consistent with the overall trend but unable to reflect real-time changes.
[0076] (2) Machine learning: First, the selection of its input parameters is often not comprehensive, which hinders the further improvement of model accuracy. ROP is affected by many factors, including drilling parameters, logging parameters, and wellbore trajectory. However, most studies only consider specific categories or key influencing factors, which makes the model performance unsatisfactory. Secondly, previous studies have paid less attention to the robustness of the model. Studies have shown that noisy data will weaken the prediction performance of the ROP model. Due to the complex drilling environment and defects in various sensors, data loss and distortion are difficult to avoid during data collection, and noise cannot be completely eliminated. In this case, the robustness of the prediction model is particularly critical. After all, the prediction accuracy of the model depends on the data quality. Poor robustness will limit the generalization ability of the model and may even cause overfitting problems.
[0077] Based on this, the present application provides a drilling rate prediction method and device based on artificial intelligence.
[0078] Next, the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems will be described in detail through embodiments and in combination with the accompanying drawings. The embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of them.
[0079] Figure 1 It is a flowchart of a drilling rate prediction method based on artificial intelligence shown in an exemplary embodiment of the present application.
[0080] In an exemplary embodiment, Figure 1 As shown, a drilling rate prediction method based on artificial intelligence is provided. In this embodiment, the method may include the following steps:
[0081] Step 100: Enhance the drilling samples by generating adversarial network methods to expand the training data set.
[0082] Step 200: Generate an optimal drilling penetration rate prediction model based on the training data set.
[0083] Step 300: Perform interpretability analysis on the optimal borehole penetration prediction model using a locally interpretable model-independent interpretation algorithm to improve the transparency and understandability of the model.
[0084] Figure 2 It is a schematic diagram of a specific implementation process of a prediction method shown in an exemplary embodiment of the present application.
[0085] In one possible implementation, Figure 2 As shown, the implementation process of its prediction method is as follows:
[0086] First, the drilling samples are enhanced by GAN method to expand the training data set.
[0087] Based on these data, a ROP prediction model was constructed using a classic machine learning algorithm.
[0088] To evaluate the performance of these models under different noise conditions, 0%, 1.7%, and 5.1% noise data were added to the training set, and their prediction accuracy and robustness under different noise levels were compared.
[0089] Then, models with high prediction accuracy and strong robustness were screened out, and these models were fused using the Stacking algorithm to create multiple fusion models.
[0090] Subsequently, these fusion models were evaluated to determine the best ROP prediction model.
[0091] Finally, the LIME algorithm was used to perform interpretability analysis on the optimal ROP prediction model to improve the transparency and understandability of the model.
[0092] In one possible implementation, Figure 2 As shown, the specific implementation of this method is:
[0093] (1) Enhanced training set
[0094] First, based on the daily drilling reports, parameters covering multiple aspects are calculated, including logging parameters (gamma ray, interval transmission time, microsphere focused resistivity, deep resistivity), drilling parameters (hook weight, drill bit weight, standpipe pressure, drill bit revolutions per minute, torque, flow rate, drill bit travel distance, drill bit travel time), well trajectory parameters (depth, angle and azimuth), and drill bit parameters (drill bit size).
[0095] Then GAN technology is used to generate additional training samples. By learning the distribution characteristics of existing data, GAN can generate new and diverse data samples, thereby effectively expanding and enhancing the training set.
[0096] This method not only improves the diversity of data, but also provides a richer sample basis for model training.
[0097] (2) Generate a model with excellent prediction accuracy and strong robustness
[0098] Based on the processed drilling data, six ROP prediction models were constructed using a variety of classic machine learning algorithms, including DT, RF, LGBM, XGB, SVR, and BPNN.
[0099] Next, different proportions of noise data (0%, 1.7%, 5.1%) were introduced into the training set to compare the prediction accuracy and robustness of these models under different noise levels.
[0100] Then, after comparative analysis, the models with outstanding performance in prediction accuracy and robustness were selected. Next, the Stacking algorithm was used to fuse these selected models.
[0101] The specific operation is to set these selected models as base learners and select the LR model as the meta-learner to construct multiple fusion models.
[0102] Finally, a comprehensive evaluation of the prediction accuracy and robustness of these fusion models was performed to determine the optimal ROP prediction model.
[0103] (3) Explain the optimal model
[0104] After determining the optimal ROP prediction model, the LIME algorithm was used to conduct an in-depth interpretability analysis.
[0105] This step aims to reveal the logic behind the model prediction results, especially to explore how key input factors such as gamma rays, hook weight, riser pressure, depth, etc. jointly affect ROP prediction.
[0106] The LIME algorithm can quantify the specific contribution of each feature to the model output, providing clear insights into the model's decision-making process.
[0107] This analysis not only enhances the transparency of the model, but also helps identify key data points that may affect model performance, providing direction for further optimization of the model.
[0108] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the instructions, these steps are not necessarily executed in sequence according to the order of the instructions. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above-mentioned embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.
[0109] Corresponding to the aforementioned embodiment of the drilling rate prediction method based on artificial intelligence, adopting the same technical concept, the present application also provides an embodiment of the drilling rate prediction device based on artificial intelligence.
[0110] Figure 3 It is a structural schematic diagram of a drilling rate prediction device based on artificial intelligence shown in an exemplary embodiment of the present application.
[0111] In an exemplary embodiment, Figure 3 As shown, the drilling rate prediction device based on artificial intelligence includes:
[0112] Training set enhancement module 1: used to enhance drilling samples through the generative adversarial network method to expand the training data set;
[0113] Optimal model generation module 2: used to generate an optimal drilling penetration rate prediction model based on the training data set;
[0114] Optimal model analysis module 3: It is used to perform interpretability analysis on the optimal drilling penetration prediction model using a locally interpretable model-independent interpretation algorithm to improve the transparency and understandability of the model.
[0115] In a possible implementation, the model generation module further includes:
[0116] Model building module: used to build a drilling penetration rate prediction model based on the training data set using a variety of classic machine learning algorithms;
[0117] Noise addition module: used to add 0%, 1.7% and 5.1% noise data to the training set, and compare its prediction accuracy and robustness under different noise levels to evaluate the performance of these models under different noise conditions;
[0118] Model screening module: screen out models with high prediction accuracy and strong robustness;
[0119] Model fusion module: Use ensemble learning algorithms to fuse these models and create multiple fusion models;
[0120] Model evaluation module: evaluates the multiple fusion models to determine the optimal drilling penetration rate prediction model.
[0121] For the specific definition of the drilling rate prediction device based on artificial intelligence, please refer to the definition of the drilling rate prediction method based on artificial intelligence above, which will not be repeated here. Each module in the above-mentioned drilling rate prediction device based on artificial intelligence can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0122] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0123] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. A drilling rate prediction method based on artificial intelligence, characterized in that: include: The drilling samples are enhanced by generative adversarial network method to expand the training data set; Based on the training data set, generating an optimal drilling penetration rate prediction model; The optimal borehole penetration prediction model is analyzed for interpretability using a locally interpretable model-independent explanation algorithm to improve the transparency and understandability of the model.
2. The drilling rate prediction method based on artificial intelligence according to claim 1, characterized in that: The method of enhancing the drilling samples by generating an adversarial network method to expand the training data set further includes: Based on daily drilling reports, calculate parameters covering multiple aspects, including logging parameters, drilling parameters, well trajectory parameters, and drill bit parameters; Generative adversarial network technology is used to generate additional training samples. Generative adversarial network technology can generate new and diverse data samples by learning the distribution characteristics of existing data, thereby effectively expanding and enhancing the training set.
3. The drilling rate prediction method based on artificial intelligence according to claim 1, characterized in that: The step of generating an optimal drilling penetration rate prediction model based on the training data set further comprises: Based on the training data set, a drilling penetration rate prediction model is constructed using a variety of classic machine learning algorithms; 0%, 1.7%, and 5.1% noise data were added to the training set, and the prediction accuracy and robustness under different noise levels were compared to evaluate the performance of these models under different noise conditions. Screen out models with high prediction accuracy and strong robustness; Use ensemble learning algorithms to fuse these models and create multiple fusion models; A plurality of the fusion models are evaluated to determine an optimal drilling penetration rate prediction model.
4. The drilling rate prediction method based on artificial intelligence according to claim 3, characterized in that: The method of fusing these models using an integrated learning algorithm to create multiple fusion models further includes: Set the selected model as the base learner; The LR model is selected as the meta-learner; In this way, multiple fusion models are constructed.
5. The drilling rate prediction method based on artificial intelligence according to claim 1, characterized in that: Various classic machine learning methods include decision tree supervised learning algorithm, random forest ensemble learning algorithm, lightweight gradient boosting algorithm, extreme gradient boosting algorithm, support vector regression algorithm and back propagation neural network.
6. The drilling rate prediction method based on artificial intelligence according to claim 5, characterized in that: The generative adversarial network includes a generator and a discriminator; The goal of the generator is to generate realistic data samples from random noise, and the goal of the discriminator is to distinguish between the generated samples and the real samples; The two compete with each other during the training process. The generator strives to generate more realistic samples to deceive the discriminator, while the discriminator strives to improve its ability to distinguish true and false samples. Through this adversarial mechanism, the generator is eventually able to generate high-quality data samples.
7. The drilling rate prediction method based on artificial intelligence according to claim 1, characterized in that: The drilling penetration rate is the depth of drilling by the drill bit per unit time, expressed in meters per hour (m / h). It is a key indicator for measuring drilling speed and affects the efficiency and cost of drilling operations. The drilling penetration rate is affected by many factors, including drill bit type, drilling pressure, rotation speed, drilling fluid properties, and formation characteristics; During the drilling process, optimizing the borehole penetration rate can increase the drilling speed, reduce the drilling cost, reduce the non-productive time and improve the drilling safety.
8. The drilling rate prediction method based on artificial intelligence according to claim 1, characterized in that: The ensemble learning method improves the overall prediction performance by combining the prediction results of multiple different models. It consists of two main levels: The first layer is the base learner, which is a machine learning algorithm, including decision trees, support vector machines, and neural networks; The second layer is the meta-learner, which is a model, such as logistic regression, that learns how to most effectively combine the predictions of the base learners. The ensemble learning method divides the training set multiple times through cross-validation. Each base learner trains and predicts the divided data, and then uses these prediction results as new features for the meta-learner to make the final prediction.
9. A drilling rate prediction device based on artificial intelligence, characterized in that: include: Training set enhancement module: used to enhance drilling samples through the generative adversarial network method to expand the training data set; Optimal model generation module: used to generate an optimal drilling penetration rate prediction model based on the training data set; Optimal Model Analysis Module: Use a locally interpretable model-independent interpretation algorithm to perform interpretability analysis on the optimal borehole penetration prediction model to improve the transparency and understandability of the model.
10. The drilling rate prediction device based on artificial intelligence according to claim 9, characterized in that: The model generation module further comprises: Model building module: used to build a drilling penetration rate prediction model based on the training data set using a variety of classic machine learning algorithms; Noise addition module: used to add 0%, 1.7% and 5.1% noise data to the training set, and compare its prediction accuracy and robustness under different noise levels to evaluate the performance of these models under different noise conditions; Model screening module: screen out models with high prediction accuracy and strong robustness; Model fusion module: Use ensemble learning algorithms to fuse these models and create multiple fusion models; Model evaluation module: evaluates the multiple fusion models to determine the optimal drilling penetration rate prediction model.