An overflow risk key feature screening method based on data mining

Abstract

The application provides a kind of overflow risk key feature screening method based on data mining, relating to the technical field of oil drilling engineering, comprising: using weighted moving average filter to carry out noise reduction processing to mud logging sampling data, to obtain overflow data set;The correlation analysis is carried out to the overflow data set, to obtain multiple groups of high correlation variable set;For multiple groups of high correlation variable set, the overflow risk key feature screening is carried out through feature screening model, to obtain the overflow risk key feature set.The overflow risk key feature screening method provided by the application screens out important features, enhances data interpretability, thereby more deeply understands the correlation between each feature and overflow risk, improves the accuracy of overflow risk early warning model, and has important reference value for the diagnosis and early warning of overflow risk.

Description

Technical Field

[0001] This invention relates to the field of oil drilling engineering technology, and more specifically, to a method for screening key features of overflow risk based on data mining. Background Technology

[0002] Drilling engineering is a crucial link in the exploration and development of oil, natural gas, and geothermal energy. It is a complex underground engineering project involving multiple disciplines, trades, stages, and procedures. Due to the complexity of the downhole environment and the uncertainty of geological conditions, drilling accidents occur frequently. Blowouts, as one of the most common and dangerous drilling accidents, pose a significant threat to the safe conduct of drilling operations. With the rapid development of emerging technologies such as big data and artificial intelligence, applying these technologies to drilling engineering is beneficial to the intelligent and information-based development of the petroleum industry.

[0003] Currently, many drilling engineering experts have applied artificial intelligence technology to the diagnosis and early warning of spill risks. However, due to the large volume, high dimensionality, and inconsistent data quality of data obtained from integrated logging tools, many spill warning models suffer from low accuracy, slow operation, high computational costs, and poor spill risk identification. Therefore, improving the quality of drilling data, resolving data redundancy issues, and extracting key features from drilling data are important research questions that are of great significance for improving the performance of spill warning models in the future.

[0004] To address the problems of existing technologies, this invention provides a data mining-based method for screening key features of overflow risk. Summary of the Invention

[0005] To address the problems of existing technologies, this invention provides a data mining-based method for screening key features of overflow risk, the method comprising:

[0006] A weighted moving average filter was used to reduce noise in the logging sampling data to obtain the overflow dataset;

[0007] Correlation analysis was performed on the overflow dataset to obtain multiple sets of highly correlated variables;

[0008] For the set of highly correlated variables, the key features of overflow risk are screened using a feature screening model to obtain the set of key features of overflow risk.

[0009] According to an embodiment of the present invention, the overflow dataset is obtained through the following steps:

[0010] a. Based on the sampling frequency of the logging data, set a suitable moving time window for the weighted moving average filter;

[0011] b. Divide the continuous logging sampling data into queues of fixed length according to the moving time window, assign different weights to the data within the moving time window, and calculate the weighted arithmetic mean of the data in the current queue as the output of the current queue;

[0012] c. Remove the first data of the current queue, shift the remaining data forward in sequence, and incorporate the new sampled data as the tail of the new queue. Perform a weighted calculation on the new queue using the calculation method in step b to obtain the output of the new queue.

[0013] d. Repeat step c until the moving time window moves to the end of the logging sampling data;

[0014] e. Combine all the outputs of all queues into a dataset, which will be used as the overflow dataset after noise reduction.

[0015] According to one embodiment of the present invention, the weighted moving average filter comprises:

[0016]

[0017] Where: y t The output of the weighted moving average filter; n is the number of samples within a moving time window; w i x is the weight of the i-th sample; i The i-th data in this queue.

[0018] According to an embodiment of the present invention, multiple sets of highly correlated variables are obtained through the following steps:

[0019] For the overflow dataset, Spearman correlation analysis was used to calculate the Spearman correlation coefficient between the variables to measure the monotonic relationship between the two variables.

[0020] Variables with Spearman correlation coefficients higher than a threshold are grouped together to obtain multiple sets of highly correlated variables.

[0021] According to an embodiment of the present invention, the feature selection model includes a nonlinear support vector machine model and a feature elimination model, and the set of key features for overflow risk is obtained through the following steps:

[0022] In the overflow dataset, the overflow condition dataset and the safe drilling condition dataset are randomly shuffled and divided into k samples on average, which serve as the cross-validation sample set.

[0023] Construct the nonlinear support vector machine model based on the Gaussian kernel function, and adjust the hyperparameters of the nonlinear support vector machine model;

[0024] The cross-validation sample set is input into the nonlinear support vector machine model to perform k-fold cross-validation. Accuracy is selected as the model evaluation index, and the average accuracy of k training iterations is taken as the final model accuracy.

[0025] The feature elimination model is used to screen key overflow features. The nonlinear support vector machine model is repeatedly trained with different samples. Weak features are eliminated according to the evaluation index of the nonlinear support vector machine model until the key overflow features that meet the accuracy threshold are obtained, forming the set of key overflow risk features.

[0026] According to an embodiment of the present invention, the optimization objective of the nonlinear support vector machine model is:

[0027]

[0028] sty i (w×φ(x i )+b)≥1-ζ i i = 1, 2, ..., n

[0029] ζ i ≥0, i=1,2,...,n

[0030] Where: x i Let φ(x) be the feature vector of the i-th sample in the input space. i ) is to transform the feature vector x i The function that maps to a higher-dimensional feature space; w is the weight vector that determines the direction of the hyperplane; ||w|| represents the square of the L2 norm of w; b is the bias term used to adjust the distance between the hyperplane and the origin; ζ i Let be the slack variable for the i-th sample; C is the penalty factor; y i is the class label of the i-th sample; n is the number of samples;

[0031] The Gaussian kernel function is:

[0032] K(x i ,x j )=φ(x i )·φ(x j )=exp(-γ||x i -x j || 2 )

[0033] Where K(x) i ,x j ) is the Gaussian kernel function; φ(x) i ) is to transform the feature vector x i A function that maps to a higher-dimensional feature space; φ(xj) is the function that maps the feature vector x to a higher-dimensional feature space.j A function mapped to a higher-dimensional feature space; γ is the parameter in the Gaussian kernel function that determines the function width; x i x j Let x be two sample points in the dataset; i -x j || represents x i and x j The Euclidean distance between them.

[0034] According to one embodiment of the present invention, key overflow characteristics are screened through the following steps:

[0035] A. By using a decision function, the features that contribute the least to establishing the decision boundary of the model are removed from the multiple sets of highly correlated variables.

[0036] B. Using the dataset after feature removal as new input samples, retrain the nonlinear support vector machine model to obtain the accuracy of the current model;

[0037] C. Compare the accuracy of the current model with the accuracy of the model trained in the last time to determine whether the accuracy has decreased;

[0038] D. Based on the accuracy judgment results, add the features that are highly relevant but were mistakenly removed to the overflow risk key feature set;

[0039] E. Repeat steps A through D until all drilling features are traversed to obtain a specific number of key features for overflow risk, thus obtaining the final set of key features for overflow risk.

[0040] According to an embodiment of the present invention, the accuracy judgment result is obtained through the following steps:

[0041] If the model accuracy decreases, the features that are removed are those that are highly relevant but were mistakenly removed.

[0042] If the model accuracy increases, the features that are deleted are those with low relevance and should be removed.

[0043] According to another aspect of the invention, a storage medium is also provided, which includes instructions for performing the methods described in any of the preceding claims.

[0044] According to another aspect of the present invention, a data mining-based overflow risk key feature screening apparatus is also provided, which performs the method as described in any of the preceding claims, the apparatus comprising:

[0045] The data denoising module uses a weighted moving average filter to denoise the logging sampling data to obtain the overflow dataset;

[0046] The feature correlation analysis module performs correlation analysis on the overflow dataset to obtain multiple sets of highly correlated variables.

[0047] The feature selection module selects key features of overflow risk for multiple sets of highly correlated variables through a feature selection model, thereby obtaining a set of key features of overflow risk.

[0048] This invention provides a data mining-based method for screening key features of overflow risk, which has the following advantages compared with existing technologies:

[0049] 1) This invention uses a weighted moving average filter to reduce the noise impact of integrated logging instrument data. By setting appropriate weights, it achieves a better signal smoothing effect, effectively eliminating the impact of short-term fluctuations and greatly helping to mine the long-term trend of data.

[0050] 2) This invention uses the SVM-RFE model to screen key overflow features. Through iterative training, weak features in the model are eliminated. Combined with the Spearman correlation analysis method, the Spearman correlation coefficient is calculated to obtain a set of highly correlated feature variables, thus avoiding the accidental elimination of highly correlated features and causing a decline in model performance.

[0051] 3) The overflow risk key feature screening method proposed in this invention enhances data interpretability by screening out important features, thereby gaining a deeper understanding of the correlation between each feature and overflow risk, improving the accuracy of the overflow risk early warning model, and providing important reference value for the diagnosis and early warning of overflow risk.

[0052] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 A flowchart illustrating the steps of a data mining-based method for screening key features of overflow risk according to an embodiment of the present invention is shown.

[0055] Figure 2 A technical roadmap according to an embodiment of the present invention is shown;

[0056] Figure 3 A flowchart of a feature selection model according to an embodiment of the present invention is shown.

[0057] In the accompanying drawings, the same parts use the same reference numerals. Also, the drawings are not drawn to scale. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0059] Drilling engineering is a crucial link in the exploration and development of oil, natural gas, and geothermal energy. It is a complex underground engineering project involving multiple disciplines, trades, stages, and procedures. Due to the complexity of the downhole environment and the uncertainty of geological conditions, drilling accidents occur frequently. Blowouts, as one of the most common and dangerous drilling accidents, pose a significant threat to the safe conduct of drilling operations. With the rapid development of emerging technologies such as big data and artificial intelligence, applying these technologies to drilling engineering is beneficial to the intelligent and information-based development of the petroleum industry.

[0060] Currently, many drilling engineering experts have applied artificial intelligence technology to the diagnosis and early warning of spill risks. However, due to the large volume, high dimensionality, and inconsistent data quality of data obtained from integrated logging tools, many spill warning models suffer from low accuracy, slow operation, high computational costs, and poor spill risk identification. Therefore, improving the quality of drilling data, resolving data redundancy issues, and extracting key features from drilling data are important research questions that are of great significance for improving the performance of spill warning models in the future.

[0061] To address the problems of existing technologies, this invention provides a data mining-based method for screening key features of spill risk. First, a weighted moving average filter is used to reduce noise in the integrated logging data, minimizing fluctuations caused by construction environment, measuring instruments, and human factors, thus highlighting the essential patterns in the data. Second, an SVM-RFE model is built, employing a nonlinear SVM model based on the linearly inseparable nature of the data to screen key features of spill risk. Furthermore, Spearman correlation analysis is combined to obtain multiple sets of highly correlated feature variables, helping to capture erroneously deleted key features. This results in the optimal set of key features for spill risk that maximizes model performance, providing a strong basis for spill risk diagnosis and early warning.

[0062] Figure 1 A flowchart illustrating the steps of a data mining-based method for screening key features of overflow risk according to an embodiment of the present invention is shown.

[0063] like Figure 1As shown, in step S1, a weighted moving average filter is used to denoise the logging sample data to obtain the overflow dataset. Specifically, this invention uses a weighted moving average filter to denoise the integrated logging instrument data (logging sample data), improving data quality and enhancing data smoothing.

[0064] In one embodiment, the overflow dataset is obtained through the following steps S11-S15.

[0065] Step S11: Based on the sampling frequency of the logging data, set a suitable moving time window for the weighted moving average filter. Specifically, based on the frequency of the data acquired by the integrated logging instrument, set a suitable moving time window for the filter.

[0066] Step S12: Divide the continuous logging sampling data into queues of fixed length according to the moving time window, assign different weights to the data within the moving time window, and calculate the weighted arithmetic mean of the data in the current queue as the output of the current queue. Specifically, divide the continuous sampling data into queues of fixed length according to the moving time window, assign different weights to the data within the window based on the degree of influence of data at different times within the same moving segment on the output value of the current queue, and calculate the weighted arithmetic mean of the data in the current queue as the output of the current queue. The specific formula is as follows:

[0067]

[0068] Where: y t The output of the weighted moving average filter; n is the number of samples within a moving time window; w i x is the weight of the i-th sample; i The i-th data in this queue.

[0069] Step S13: Remove the first data of the current queue, shift the remaining data forward sequentially, and incorporate the new sampled data as the tail of the new queue. Perform a weighted calculation on the new queue using the calculation method in step S12 to obtain the output of the new queue. Specifically, remove the first data of the above queue, shift the remaining n-1 data forward sequentially, and incorporate the new sampled data as the tail of the new queue. Perform a weighted calculation on the new queue using the same method as in step S12 to obtain the output of the new queue.

[0070] Step S14: Repeat step S13 until the moving time window reaches the end of the logging sampling data. Specifically, continuously repeat this process (step S13) until the time window reaches the end of the sampling dataset.

[0071] Step S15: Combine all the outputs of all queues into a single dataset, which serves as the denoised overflow dataset. Specifically, reassemble all the outputs of the filters mentioned above into a new dataset, which is the denoised drilling dataset (overflow dataset).

[0072] This invention employs a weighted moving average filter to reduce the noise impact of integrated logging instrument data. By setting appropriate weights, it achieves a better signal smoothing effect, effectively eliminating the impact of short-term fluctuations and greatly helping to uncover long-term data trends.

[0073] like Figure 1 As shown, in step S2, correlation analysis is performed on the overflow dataset to obtain multiple sets of highly correlated variables.

[0074] In one embodiment, in step S2, multiple sets of highly correlated variables are obtained through the following steps: For the overflow dataset, Spearman correlation analysis is used to calculate the Spearman correlation coefficient between variables to measure the monotonic relationship between two variables; variables with Spearman correlation coefficients higher than a threshold are grouped together to obtain multiple sets of highly correlated variables.

[0075] Specifically, correlation analysis was performed on the logging dataset using the Spearman correlation analysis method. The Spearman correlation coefficient was calculated to measure the monotonic relationship between two variables. Variables with high Spearman correlation coefficients were grouped together, indicating a high degree of association between them, providing a basis for subsequent variable selection. Through Spearman correlation analysis, multiple sets of highly correlated variables were obtained.

[0076] In one embodiment, the Spearman correlation coefficient is calculated using the following formula:

[0077]

[0078] Where ρ is the Spearman correlation coefficient between the two variables; n is the sample size; d i Let be the rank difference between two variables in the i-th sample.

[0079] like Figure 1 As shown, in step S3, for multiple sets of highly correlated variables, the key features of overflow risk are screened through the feature screening model to obtain the key feature set of overflow risk.

[0080] In one embodiment, the feature filtering model (SVM-RFE model) includes a nonlinear support vector machine model and a feature elimination model. In step S3, the key feature set of overflow risk is obtained through the following steps S31-S34.

[0081] Step S31: In the overflow dataset, randomly shuffle the overflow condition dataset and the safe drilling condition dataset, and divide them into k samples on average as the cross-validation sample set. Specifically, randomly shuffle the overflow condition dataset and the safe drilling condition dataset, and divide them into k samples on average. This sample set contains all overflow risk features. k is generally a positive integer between 3 and 10. This step provides the sample basis for subsequent k-fold cross-validation.

[0082] Step S32: Construct a nonlinear support vector machine (SVM) model based on a Gaussian kernel function and adjust its hyperparameters. Specifically, construct a nonlinear support vector machine (SVM) model based on a Gaussian kernel function. The basic idea of ​​the SVM algorithm is to find an optimal decision boundary that maximizes the margin between samples of different classes to achieve the best sample classification effect. That is, the SVM algorithm transforms the sample classification problem into an optimization problem, and achieves sample classification by solving for the maximum margin, which is the distance from the nearest sample point to the decision boundary.

[0083] In one embodiment, the optimization objective of the nonlinear support vector machine model is:

[0084]

[0085] sty i (w×φ(x i )+b)≥1-ζ i i = 1, 2, ..., n

[0086] ζ i ≥0, i=1,2,...,n

[0087] Where: x i Let φ(x) be the feature vector of the i-th sample in the input space. i ) is to transform the feature vector x i The function that maps to a higher-dimensional feature space; w is the weight vector that determines the direction of the hyperplane; ||w|| represents the square of the L2 norm of w; b is the bias term used to adjust the distance between the hyperplane and the origin; ζ i Let be the slack variable for the i-th sample; C is the penalty factor; y i is the class label of the i-th sample; n is the number of samples;

[0088] Because integrated logging data is characterized by high dimensionality and large volume, linear classification is difficult to achieve. Therefore, this invention introduces a nonlinear support vector machine (SVM) model and uses a kernel function to map the data to a higher-dimensional feature space. The kernel function can directly calculate the inner product of mapping functions with high dimensionality without performing actual high-dimensional mapping, thus avoiding complex calculations in high-dimensional space. This invention cleverly transforms the nonlinear problem in the original space into a linear problem in a high-dimensional space, improving the classification performance of the SVM model.

[0089] Specifically, the nonlinear SVM model employs a Gaussian kernel function. The Gaussian kernel function implicitly maps sample data to a higher-dimensional feature space, making originally linearly inseparable data linearly separable, thus effectively handling nonlinear relationships between features and improving classification performance. The specific formula for the Gaussian kernel function is as follows:

[0090] K(x i ,x j )=φ(x i )·φ(x j )=exp(-γ||x i -x j || 2 )

[0091] Where K(x) i ,x j ) is the Gaussian kernel function; φ(x) i ) is to transform the feature vector x i A function that maps to a higher-dimensional feature space; φ(xj) is the function that maps the feature vector x to a higher-dimensional feature space. j A function mapped to a higher-dimensional feature space; γ is the parameter in the Gaussian kernel function that determines the function width; x i x j Let x be two sample points in the dataset; i -x j || represents x i and x j The Euclidean distance between them.

[0092] In one embodiment, in step S32, the hyperparameters of the nonlinear SVM model, such as regularization parameters, Gaussian kernel width parameters, and penalty factors, are adjusted to optimize the model's classification performance.

[0093] Step S33: Input the cross-validation sample set into the nonlinear support vector machine model, perform k-fold cross-validation, select accuracy as the model evaluation metric, and take the average accuracy of k training iterations as the final model accuracy. Specifically, input the samples containing the overflow risk feature set into a nonlinear support vector machine (SVM) model based on a Gaussian kernel function, perform k-fold cross-validation, select accuracy as the model evaluation metric, and take the average accuracy of k training iterations as the final model accuracy.

[0094] In one embodiment, the specific formula for calculating model accuracy is as follows:

[0095]

[0096] Where ACC is the accuracy; TP represents the number of samples that were actually positive but were predicted as positive; FP represents the number of samples that were actually negative but were predicted as positive; and FN represents the number of samples that were actually positive but were predicted as negative.

[0097] Step S34: Employ a feature elimination model to screen key overflow features. This involves repeatedly training a nonlinear support vector machine (SVM) model using different samples, and eliminating weak features based on the SVM model's evaluation metrics until a set of key overflow features meeting an accuracy threshold is obtained. Specifically, the recursive feature elimination (RFE) method is used as the feature elimination model to screen key overflow features. This involves repeatedly training the model using different samples, eliminating weak features based on the model's evaluation metrics, and continuing until a specific number of key overflow features meeting a certain accuracy threshold are obtained.

[0098] This invention uses the SVM-RFE model to screen key overflow features. Through iterative training, weak features in the model are eliminated. Combined with the Spearman correlation analysis method, the Spearman correlation coefficient is calculated to obtain a set of highly correlated feature variables, thus avoiding the accidental removal of highly correlated features and the resulting degradation of model performance.

[0099] Figure 2 A technology roadmap according to an embodiment of the present invention is shown.

[0100] like Figure 2 As shown, a weighted moving average filter is used to reduce noise in the integrated logging instrument data, thereby improving data quality and smoothing effect, and yielding an overflow dataset.

[0101] like Figure 2As shown, correlation analysis was performed on the logging dataset using the Spearman correlation analysis method to calculate the Spearman correlation coefficient between parameters. The Spearman correlation coefficient measures the monotonic relationship between two variables. Variables with high Spearman correlation coefficient values ​​were grouped together, indicating a high degree of association between variables in that group, providing a basis for subsequent variable selection. Through Spearman correlation analysis, multiple sets of highly correlated variables were obtained.

[0102] like Figure 2 As shown, a nonlinear support vector machine (SVM) model based on the Gaussian kernel function is constructed. Recursive Feature Elimination (RFE) is used to screen overflow key features. The model is repeatedly trained using different samples, and weak features are eliminated based on model evaluation metrics until a specific number of overflow key features that meet a certain threshold accuracy are obtained.

[0103] The overflow risk key feature screening method proposed in this invention enhances data interpretability by screening out important features, thereby gaining a deeper understanding of the correlation between each feature and overflow risk, improving the accuracy of the overflow risk early warning model, and providing important reference value for the diagnosis and early warning of overflow risk.

[0104] Figure 3 A flowchart of a feature selection model according to an embodiment of the present invention is shown.

[0105] like Figure 3 As shown, key overflow characteristics are screened through the following steps A-E.

[0106] Step A: Using a decision function, features that contribute the least to establishing the model's decision boundary are removed from the set of highly correlated variables. Specifically, the support vectors and their corresponding coefficients are analyzed; the support vectors are the key sample data constituting the decision boundary. Features that contribute the least to establishing the model's decision boundary are removed from the spillover risk feature set. Specifically, the decision function is expressed as:

[0107]

[0108] Where n is the number of support vectors; α i For Lagrange multipliers; x i Let y be the i-th support vector; i Let x be the label of the i-th support vector; x be the data point to be classified; and b be the bias term.

[0109] Step B: Using the dataset after feature removal as new input samples, retrain the nonlinear support vector machine model to obtain the accuracy of the current model. Specifically, using the dataset after feature removal as new input samples, retrain the SVM model according to step S33 to obtain the model's accuracy.

[0110] Step C: Compare the accuracy of the current model with the accuracy of the model trained in the last time to determine whether the accuracy has decreased.

[0111] In one embodiment, the accuracy judgment result is obtained through the following steps: if the model accuracy decreases, the features to be removed are features with high relevance but which were mistakenly removed; if the model accuracy increases, the features to be deleted are features with low relevance that should be deleted.

[0112] Specifically, compare the accuracy of the current model with the accuracy of the model trained previously to determine if the model accuracy has decreased. If the model accuracy has decreased, it indicates that the removed features have a significant impact on classification performance and should be reinstated into the overflow risk key feature set. If the model accuracy has increased, it indicates that the removed features are not key features for overflow risk and should be removed from the overflow risk key feature set.

[0113] Step D: Based on the accuracy assessment, add highly relevant but mistakenly removed features to the overflow risk key feature set. Specifically, re-add highly relevant but mistakenly removed features to the overflow risk feature set. Furthermore, to avoid discarding all highly relevant features and degrading model performance, after each round of feature selection, for all removed feature sets, select the most important feature. Based on the multiple sets of highly relevant variables obtained in step S2, find the feature variables highly correlated with this feature. If these feature variables have all been removed, add the feature back to the overflow risk feature set and train again.

[0114] Step E: Repeat steps A-D until all drilling features have been traversed, obtaining a specific number of key features for spill risk, and thus obtaining the final set of key features for spill risk. Specifically, continuously repeat steps A-D until all drilling features have been traversed, obtaining a specific number of key features for spill risk, and thus obtaining the final set of key features for spill risk.

[0115] The data mining-based method for screening key features of overflow risk provided by this invention can also be used in conjunction with a computer-readable storage medium. The storage medium stores a computer program, which is executed to run the data mining-based method for screening key features of overflow risk. The computer program can execute computer instructions, which include computer program code. The computer program code can be in the form of source code, object code, executable file, or some intermediate form.

[0116] Computer-readable storage media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0117] It should be noted that the contents of computer-readable storage media may be appropriately added to or subtracted from the contents according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media may not include electrical carrier signals and telecommunication signals.

[0118] According to another aspect of the present invention, a data mining-based overflow risk key feature screening device is also provided, which performs a data mining-based overflow risk key feature screening method. The device includes: a data denoising module, a feature correlation analysis module, and a feature screening module.

[0119] In one embodiment, the data denoising module uses a weighted moving average filter to denoise the logging sampling data to obtain an overflow dataset; the feature correlation analysis module performs correlation analysis on the overflow dataset to obtain multiple sets of highly correlated variables; and the feature selection module uses a feature selection model to select key features of overflow risk for the multiple sets of highly correlated variables to obtain a set of key features of overflow risk.

[0120] In summary, this invention provides a data mining-based method for screening key features of overflow risk, which has the following advantages compared with existing technologies:

[0121] 1) This invention uses a weighted moving average filter to reduce the noise impact of integrated logging instrument data. By setting appropriate weights, it achieves a better signal smoothing effect, effectively eliminating the impact of short-term fluctuations and greatly helping to mine the long-term trend of data.

[0122] 2) This invention uses the SVM-RFE model to screen key overflow features. Through iterative training, weak features in the model are eliminated. Combined with the Spearman correlation analysis method, the Spearman correlation coefficient is calculated to obtain a set of highly correlated feature variables, thus avoiding the accidental elimination of highly correlated features and causing a decline in model performance.

[0123] 3) The overflow risk key feature screening method proposed in this invention enhances data interpretability by screening out important features, thereby gaining a deeper understanding of the correlation between each feature and overflow risk, improving the accuracy of the overflow risk early warning model, and providing important reference value for the diagnosis and early warning of overflow risk.

[0124] It should be understood that the embodiments disclosed herein are not limited to the specific structures, processing steps, or materials disclosed herein, but should be extended to equivalent substitutions of these features as understood by those skilled in the art. It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting.

[0125] In the description of this invention, unless otherwise stated, "a plurality of" means two or more; the terms "upper," "lower," "left," "right," "inner," "outer," "front end," "rear end," "head," "tail," 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. Furthermore, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0126] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0127] Certain terms are used throughout this application to refer to specific system components. As those skilled in the art will recognize, the same components may often be referred to by different names, and therefore this application is not intended to distinguish those components that differ only in name and not in function. In this application, the terms “comprise,” “include,” and “have” are used in an open-ended manner and should therefore be interpreted as meaning “including, but not limited to…”. Furthermore, the terms “substantially,” “materially,” or “approximately” as used herein refer to industry-accepted tolerances for the corresponding terms. The term “coupling,” as may be used herein, includes direct coupling and indirect coupling via additional components, elements, circuits, or modules, wherein, for indirect coupling, the intermediate component, element, circuit, or module does not alter the information of the signal but may adjust its current level, voltage level, and / or power level. Inferred coupling (e.g., one element is inferredly coupled to another element) includes direct and indirect coupling between two elements in the same manner as “coupling.”

[0128] The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0129] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

[0130] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for screening key features of spillover risk based on data mining, characterized in that, The method includes: A weighted moving average filter was used to reduce noise in the logging sampling data to obtain the overflow dataset; Correlation analysis was performed on the overflow dataset to obtain multiple sets of highly correlated variables; For the set of highly correlated variables, the key features of overflow risk are screened using a feature screening model to obtain the set of key features of overflow risk.

2. The method for screening key features of overflow risk based on data mining as described in claim 1, characterized in that, The overflow dataset is obtained through the following steps: a. Based on the sampling frequency of the logging data, set a suitable moving time window for the weighted moving average filter; b. Divide the continuous logging sampling data into queues of fixed length according to the moving time window, assign different weights to the data within the moving time window, and calculate the weighted arithmetic mean of the data in the current queue as the output of the current queue; c. Remove the first data of the current queue, shift the remaining data forward in sequence, and incorporate the new sampled data as the tail of the new queue. Perform a weighted calculation on the new queue using the calculation method in step b to obtain the output of the new queue. d. Repeat step c until the moving time window moves to the end of the logging sampling data; e. Combine all the outputs of all queues into a dataset, which will be used as the overflow dataset after noise reduction.

3. The method for screening key features of overflow risk based on data mining as described in claim 2, characterized in that, The weighted moving average filter includes: Where: y t The output of the weighted moving average filter; n is the number of samples within a moving time window; w i x is the weight of the i-th sample; i The i-th data in this queue.

4. A method for screening key features of overflow risk based on data mining as described in any one of claims 1-3, characterized in that, The following steps are used to obtain multiple sets of highly correlated variables: For the overflow dataset, Spearman correlation analysis was used to calculate the Spearman correlation coefficient between the variables to measure the monotonic relationship between the two variables. Variables with Spearman correlation coefficients higher than a threshold are grouped together to obtain multiple sets of highly correlated variables.

5. A method for screening key features of overflow risk based on data mining as described in any one of claims 1-4, characterized in that, The feature selection model includes a nonlinear support vector machine model and a feature elimination model. The key feature set of overflow risk is obtained through the following steps: In the overflow dataset, the overflow condition dataset and the safe drilling condition dataset are randomly shuffled and divided into k samples on average, which serve as the cross-validation sample set. Construct the nonlinear support vector machine model based on the Gaussian kernel function, and adjust the hyperparameters of the nonlinear support vector machine model; The cross-validation sample set is input into the nonlinear support vector machine model to perform k-fold cross-validation. Accuracy is selected as the model evaluation index, and the average accuracy of k training iterations is taken as the final model accuracy. The feature elimination model is used to screen key overflow features. The nonlinear support vector machine model is repeatedly trained with different samples. Weak features are eliminated according to the evaluation index of the nonlinear support vector machine model until the key overflow features that meet the accuracy threshold are obtained, forming the set of key overflow risk features.

6. The method for screening key features of overflow risk based on data mining as described in claim 5, characterized in that, The optimization objective of the nonlinear support vector machine model is: Where: x i Let φ(x) be the feature vector of the i-th sample in the input space. i ) is to transform the feature vector x i The function that maps to a higher-dimensional feature space; w is the weight vector that determines the direction of the hyperplane; ||w|| represents the square of the L2 norm of w; b is the bias term used to adjust the distance between the hyperplane and the origin; ζ i Let be the slack variable for the i-th sample; C is the penalty factor; y i is the class label of the i-th sample; n is the number of samples; The Gaussian kernel function is: K(x i ,x j )=φ(x i )·φ(x j )=exp(-γ||x i -x j || 2 ) Where K(x) i ,x j ) is the Gaussian kernel function; φ(x) i ) is to transform the feature vector x i A function that maps to a higher-dimensional feature space; φ(xj) is the function that maps the feature vector x to a higher-dimensional feature space. j A function mapped to a higher-dimensional feature space; γ is the parameter in the Gaussian kernel function that determines the function width; x i x j Let x be two sample points in the dataset; i -x j || represents x i and x j The Euclidean distance between them.

7. A method for screening key features of overflow risk based on data mining as described in claim 5 or 6, characterized in that, Filter key overflow characteristics using the following steps: A. By using a decision function, the features that contribute the least to establishing the decision boundary of the model are removed from the multiple sets of highly correlated variables. B. Using the dataset after feature removal as new input samples, retrain the nonlinear support vector machine model to obtain the accuracy of the current model; C. Compare the accuracy of the current model with the accuracy of the model trained in the last time to determine whether the accuracy has decreased; D. Based on the accuracy judgment results, add the features that are highly relevant but were mistakenly removed to the overflow risk key feature set; E. Repeat steps A through D until all drilling features are traversed to obtain a specific number of key features for overflow risk, thus obtaining the final set of key features for overflow risk.

8. The method for screening key features of overflow risk based on data mining as described in claim 7, characterized in that, The accuracy assessment result is obtained through the following steps: If the model accuracy decreases, the features that are removed are those that are highly relevant but were mistakenly removed. If the model accuracy increases, the features that are deleted are those with low relevance and should be removed.

9. A storage medium, characterized in that, It contains instructions for performing the method as described in any one of claims 1-8.

10. A data mining-based overflow risk key feature screening device, characterized in that, The apparatus for performing the method as described in any one of claims 1-8 comprises: The data denoising module uses a weighted moving average filter to denoise the logging sampling data to obtain the overflow dataset; The feature correlation analysis module performs correlation analysis on the overflow dataset to obtain multiple sets of highly correlated variables. The feature selection module selects key features of overflow risk for multiple sets of highly correlated variables through a feature selection model, thereby obtaining a set of key features of overflow risk.

Images