A method and system for predicting the probability of differential pressure sticking based on a Bayesian belief network

Abstract

The application discloses a differential pressure sticking probability prediction method and system based on a Bayesian belief network, and comprises the following steps: collecting historical drilling data of a target oil reservoir block, preprocessing sample data, and creating a sample training set and a test set; determining characteristic input variables and characteristic output variables from the sample data; using the sample training set to establish a differential pressure sticking training model based on the Bayesian belief network; and verifying the differential pressure sticking probability result after calculation by using the verification set. The differential pressure sticking probability prediction model based on the Bayesian belief network is used to input the fact drilling characteristic input parameter value of the target oil reservoir development area into the model for prediction analysis, so that the differential pressure sticking probability of instant drilling is obtained, drilling guidance is provided for field engineering and technical personnel, a decision scheme is timely adjusted, the differential pressure sticking probability is reduced, and unnecessary operation time and economic loss are reduced.

Description

Technical Field

[0001] This invention relates to the field of drilling sticking and leakage prediction, specifically to a method and system for predicting differential pressure sticking probability based on Bayesian belief networks. Background Technology

[0002] In the oil industry, where drilling is the primary development method, downhole stuck pipe accidents, caused by both subjective and objective factors, have become a key factor affecting well completion success and drilling costs. Severe stuck pipe accidents can even lead to significant casualties on-site, making it difficult to guarantee on-site operational safety. After differential pressure stuck pipe occurs, methods such as lifting, hammering, and shock are typically used to release the stuck pipe. However, these processes often induce secondary accidents such as pipe collapse and drill string breakage, further impacting drilling production time. Numerous factors influence differential pressure stuck pipe, including drilling time, torque, pump pressure, drilling pressure, drilling fluid density, and fluid loss. Due to current drilling technology and equipment conditions, it is difficult to eliminate differential pressure stuck pipe at its source. Therefore, predicting the probability of differential pressure stuck pipe occurrences from field data of these numerous influencing factors and promptly eliminating such accidents when a sticking tendency appears is of great significance for improving drilling efficiency and on-site safety.

[0003] Due to the complexity and high uncertainty of downhole conditions, the entire drilling process is essentially a black box, making it difficult for traditional linear mathematical models to clearly define this process. The application of artificial intelligence (AI) technology in drilling operations is gradually becoming a hot research area in the oil and gas exploration industry. Current AI-based stuck pipe prediction methods include: multivariate statistical analysis-based early warning methods using big data; and neural network stuck pipe prediction models that improve initial weights and thresholds through algorithm optimization. AI-based machine learning and data analysis theories are used to perform AI-driven data cleaning on large amounts of data collected at the drilling site, constructing an effective early warning system for stuck pipe accidents. This system issues alarms when the risk of stuck pipe is too high, providing more comprehensive and reasonable decision support for drilling operations. It helps on-site personnel to promptly detect and take effective measures to avoid differential pressure stuck pipe, reduce drilling input costs, and minimize unnecessary production time.

[0004] However, existing early warning analysis methods suffer from low data processing efficiency due to the massive amount of data they handle, lack of sensitivity to data gaps, and insufficient real-time performance and accuracy. Summary of the Invention

[0005] The technical problem this invention aims to solve is to overcome the shortcomings and limitations of existing differential pressure stuck drill probability prediction methods, such as high false alarm rates and poor real-time performance. These limitations prevent on-site engineers from promptly detecting differential pressure stuck drill accidents and thus hinder effective measures to eliminate the risk. Furthermore, to enrich differential pressure stuck drill risk monitoring and early warning technologies based on real-time logging data, this invention provides a differential pressure stuck drill probability prediction method and system based on Bayesian belief networks. This method predicts the probability of differential pressure stuck drill occurrences from on-site data of numerous influencing factors, enabling timely elimination of stuck drill accidents when a stuck drill tendency appears. This is of great significance for improving drilling efficiency and on-site safety.

[0006] This invention is achieved through the following technical solution:

[0007] A method for predicting drilling rig efficiency based on Bayesian belief networks includes the following steps:

[0008] S1. Collect historical drilling data of the target oil reservoir block, preprocess it to obtain sample data, and create a sample training set and test set based on the sample data;

[0009] S2. Determine the feature input variables and feature output variables from the sample data;

[0010] The feature input training tuple is represented by a vector X, denoted as X = (x1, x2, x3, x4, x5, x6); where x1 represents the riser pump pressure, x2 represents the formation lithology, x3 represents the drilling fluid density, x4 represents the rotary table rotation speed, x5 represents the rotary table torque, and x6 represents the drilling fluid filtrate loss.

[0011] Let Y represent the feature output category tuple, denoted as Y = (R1, R2); where R1 represents the normal drilling state and R2 represents the tendency to stick due to differential pressure.

[0012] The requirements for setting the feature input variables corresponding to the drilling data to be trained are closely related to the risk of differential pressure stuck drill pipe. Through the study of differential pressure stuck drill pipe mechanism and review of relevant literature, the specific parameters determined by this invention include: rotary table speed, riser pump pressure, rotary table torque, drilling fluid density, drilling fluid filtrate loss, and formation lithology.

[0013] S3. Using the sample training set, establish a differential pressure stuck drill training model based on a Bayesian belief network.

[0014] S4. The calculated differential pressure rig rate results are validated using a validation set.

[0015] The method provided by this invention uses real-time logging parameters as feature inputs and the probability of differential pressure stuck drill pipe calculated by the established Bayesian belief network model as feature outputs to determine the probability of stuck drill pipe. The model has fast training speed and high prediction accuracy, and has good practical guiding significance in predicting differential pressure stuck drill pipe.

[0016] Alternatively, in step S1, the historical drilling data includes historical drilling data and stuck pipe cases.

[0017] Further, optionally, in step S1, the preprocessing means include data cleaning, integration, and / or transformation.

[0018] Alternatively, in step S1, the sample data is divided into a training set and a test set by random sampling according to a preset ratio.

[0019] Alternatively, in step S2, the feature output adopts an "approximation" output representation, classifying the probability of the feature output variable as an approximation problem.

[0020] Further, optionally, the "approximation" output representation includes the following steps: using 0.5 as a boundary,

[0021] When the output probability value of R2 is in (0.5,1], it indicates a tendency for differential pressure stuck pipe, and the closer the output value is to 1, the more obvious the tendency for differential pressure stuck pipe. When the output value is in [0,0.5), it indicates that the possibility of drilling work being in a normal drilling state is higher than the possibility of a stuck pipe accident, and the closer the output value is to 0, the smaller the probability of differential pressure stuck pipe.

[0022] Further, optionally, step S3 includes the following steps:

[0023] Includes the following steps:

[0024] S3-1. Determine the topological relationships between the feature input variables to form a directed acyclic graph;

[0025] Based on the samples in the training set, the Naive Bayes algorithm is used to statistically calculate the prior probability P(X=x) of each feature input variable value. i ), (i = 1, 2, ..., 6) and the prior probability P(Y = R) of the output class. k (k = 1, 2), the prior probability of the output class is obtained using the following formula:

[0026]

[0027] Where |Y| represents the total number of training classes for the feature output, and |R| represents the total number of training classes. k ,Y| represents category R in Y k The number of training samples;

[0028] S3-2: Based on historical drilling data of the target reservoir block, construct a conditional probability table and calculate the conditional probability P(X=x). i |

[0029] Y = R k );

[0030] S3-3: Based on the conditional probability table and conditional probabilities, calculate the probability of differential pressure jamming under the joint probability distribution of the characteristic input variables:

[0031]

[0032] Where P(Y|X) represents the joint posterior probability of differential pressure stuck drill bit, P(Y=R k ) indicates the output category R k The prior probability, P(X), represents the joint feature probability of the feature input variables, P(X|Y=R). k ) represents the feature output category R k The conditional probability of X, P(R) k |net) represents the feature input variable x i In the case of feature output category R k The conditional probability, P(net), is the feature input variable x in the DAG graph. i The probability under the joint of predecessor nodes.

[0033] To determine the topological relationships between feature input variables and to clearly express the dependencies between attributes, a directed acyclic graph (DAG) is formed through continuous iteration and improvement. The data for the conditional probability table (CPT) is obtained by statistical calculation from the reservoir well history database of the target development area.

[0034] Alternatively, since there are dependencies among the multiple feature input variables, the joint probability distribution of P(net) can be further calculated using the given conditional probability table, as follows:

[0035]

[0036] Among them, parent(x) i ) represents the feature input variable x i The predecessor node.

[0037] Further, optionally, step S5, the actual real-time application of the Bayesian belief network differential pressure stuck drill prediction model, is also included:

[0038] Input real-time data of the target drilling characteristics and variables, and the model calculates and determines the probability of differential pressure stuck drill bit.

[0039] A differential pressure rig utilization prediction system based on Bayesian belief networks, used in the aforementioned differential pressure rig utilization prediction method based on Bayesian belief networks, includes:

[0040] Data acquisition and processing module: used to collect historical drilling data of the target oil reservoir block, preprocess it to obtain sample data, and create sample training and test sets based on the sample data;

[0041] Feature Input and Output Determination Module: Used to determine feature input variables and feature output variables from sample data; the feature input training tuple is represented by a vector X, denoted as X = (x1, x2, x3, x4, x5, x6); where x1 represents the riser pump pressure, x2 represents the formation lithology, x3 represents the drilling fluid density, x4 represents the rotary table rotation speed, x5 represents the rotary table torque, and x6 represents the drilling fluid filtrate loss.

[0042] Let Y represent the feature output category tuple, denoted as Y = (R1, R2); where R1 represents the normal drilling state and R2 represents the tendency to stick due to differential pressure.

[0043] Model building module: used to build a differential pressure stuck drill training model based on Bayesian belief network using the sample training set;

[0044] Model Validation Module: Used to validate the calculated differential pressure rig rate results using a validation set.

[0045] The present invention has the following advantages and beneficial effects:

[0046] The Bayesian Belief Network method of this invention is an important approach in machine learning and data mining. Because it relaxes the requirement for feature attributes to be conditionally independent in the Naive Bayes algorithm, it allows for dependencies between variables, thus better reflecting the real-world effects of drilling data. This method offers high efficiency in processing large amounts of data generated during drilling, is insensitive to missing data, exhibits good real-time performance, and high accuracy. Therefore, it shows promising application potential in predicting the probability of differential pressure stuck pipe.

[0047] This invention utilizes a Bayesian belief network differential pressure stuck drill probability prediction model. It inputs actual drilling characteristics of the target oil reservoir development area into the model for predictive analysis. Based on artificial intelligence machine learning and data analysis theories, it performs AI-based data cleaning on a large amount of data collected from the drilling site, constructing an effective early warning system for stuck drill incidents. This system obtains the real-time differential pressure stuck drill probability and issues an alarm when the risk of stuck drilling is too high. It provides more comprehensive and reasonable decision support for drilling operations, allowing for timely adjustments to decision-making plans, reducing the differential pressure stuck drill probability, and minimizing unnecessary operating time and economic losses. This helps on-site personnel promptly identify and take effective measures to avoid differential pressure stuck drilling, reducing drilling input costs and unnecessary production time. Attached Figure Description

[0048] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0049] Figure 1 This is a flowchart illustrating the preprocessing of historical drilling data for the standard oil reservoir block in this invention.

[0050] Figure 2 This is the directed acyclic graph (DAG) of the present invention.

[0051] Figure 3 This is the Conditional Probability Table (CPT) diagram of the present invention; where (a) represents the output category probability table and (b) represents the output feature input parameter probability table.

[0052] Figure 4 This is a flowchart of the differential pressure drilling rig rate prediction method based on Bayesian belief networks of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.

[0054] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other embodiments, well-known structures, circuits, materials, or methods have not been specifically described in order to avoid obscuring the invention.

[0055] Throughout this specification, references to "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "an embodiment," "an example," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0056] Example 1

[0057] This embodiment provides a differential pressure rig utilization prediction method based on Bayesian belief networks. The overall process is as follows: Figure 4 As shown below:

[0058] Step 1: Collect and preprocess historical drilling data of the target reservoir block.

[0059] The collection and organization of historical drilling data and stuck pipe incidents for the target oil reservoir block, followed by data preprocessing, such as... Figure 1 As shown, the raw data stored in the well logging software database often suffers from errors, missing data, duplication, or contradictions due to human error, equipment malfunction, and other factors. Before using the Bayesian belief network prediction model for data mining and learning, the raw data needs to be preprocessed. This includes data cleaning (filling in missing values ​​and smoothing noise), data integration (resolving outliers, redundancy, and duplication), and data conversion (unifying parameters with units of measurement). Specific preprocessing methods are as follows:

[0060] (11) Data cleaning: missing values ​​are filled with the central tendency measure of the attribute. Noisy data is smoothed by binning. The ordered data values ​​are smoothed by examining the data’s “nearest neighbor”.

[0061] (12) Data integration uses metadata analysis to determine the name, meaning, type, and attribute range of abnormal data to avoid errors in data integration. It also uses correlation analysis to detect redundant data and deletes duplicate data objects or data with similarity greater than the threshold that exist in the same dataset.

[0062] (13) Data transformation adopts the minimum-maximum normalization method, and the specific formula is as follows:

[0063]

[0064] Wherein, the range of values ​​y max =1, y min =0, y is the normalized result, x is the data of a certain attribute to be normalized, x min x is the minimum value of a certain attribute. max This represents the maximum value of a certain attribute.

[0065] (14) In this embodiment, in order to avoid the additional error caused by data partitioning from affecting the accuracy of the final prediction result, and in order to ensure the representativeness of the data, a simple random sampling method is used to partition the preprocessed sample data into datasets, ensuring that each sample has an equal probability of being drawn and only one sample is drawn each time, and the training set and test set are divided according to a preset ratio of 7:3.

[0066] Step 2: Determine the feature input variables and feature output variables from the sample data.

[0067] (21) To ensure that the feature input variable data is closely related to the differential pressure sticking, the preprocessed drilling data includes six feature input variables: rotary table speed, riser pump pressure, rotary table torque, drilling fluid density, drilling fluid loss, and formation lithology. Among them, riser pump pressure is x1, formation lithology is x2, drilling fluid density is x3, rotary table speed is x4, rotary table torque is x5, and drilling fluid loss is x6. The feature input variables are denoted as vector X = (x1, x2, x3, x4, x5, x6).

[0068] (22) Determine the feature output of the Bayesian belief network stuck pipe prediction model. Let Y represent the feature output category tuple, denoted as Y = (R1, R2), where R1 represents the normal drilling state and R2 represents the tendency to stick pipe due to differential pressure. The "approximation" output representation method is adopted, with 0.5 as the boundary. When the output probability value of R2 is in (0.5, 1], it indicates that there is a tendency to stick pipe due to differential pressure. The closer the output value is to 1, the more obvious the tendency to stick pipe due to differential pressure. When the output value is in [0, 0.5), it indicates that the probability of the drilling operation being in a normal drilling state is higher than the probability of a stuck pipe accident. The closer the output value is to 0, the lower the probability of a stuck pipe accident due to differential pressure.

[0069] Step 3: Use the training set for supervised learning training to establish a differential pressure stuck drill training model based on a Bayesian belief network.

[0070] (31) Determine the topological relationships between the feature input variables, and qualitatively form a directed acyclic graph (DAG) of interrelated and dependent variables, such as... Figure 2As shown, each node represents a feature input random variable, and the arc represents the relationship between two random variables, indicating that the pointing node affects the pointed-to node. The directed acyclic graph qualitatively gives the conditional dependency between feature input variables.

[0071] (32) Based on the samples in the training set, the Naive Bayes algorithm is used to statistically calculate the prior probability P(X=x) of each feature attribute value. i ), (i = 1, 2, ..., 6) and the prior probability P(Y = R) of the output class. k The prior probability of the output class (k=1,2) is obtained using the following formula:

[0072]

[0073] Where |Y| represents the total number of training classes for the feature output, and |Rk,Y| represents the number of classes R in Y. k The number of training samples.

[0074] (33) Based on historical drilling data of the target reservoir block, construct a conditional probability table (CPT) and calculate the conditional probability P(X=x). i |Y=R k) Construct a conditional probability table (CPT) as follows: Figure 3 As shown, the horizontal axis represents the feature input variables, and the vertical axis represents the feature output condition variables. The conditional probability table quantitatively represents the joint conditional probability between non-conditionally independent variables. Furthermore, to ensure the confidentiality of well history data, the specific probability values ​​involved in the conditional probability table in this embodiment are all T... i (i = 1, 2, ..., 14) are replaced.

[0075] 3-4) Based on the conditional probabilities in the conditional probability table, calculate the probability of differential pressure jamming under the joint probability distribution of the feature input variables:

[0076]

[0077] Where P(Y|X) represents the joint posterior probability of differential pressure stuck drill bit, P(Y=R k) Indicates the feature output category R k The prior probability, P(X), represents the joint feature probability of the feature input variables, P(X|Y=R). k ) represents the feature output category R k The conditional probability of X, P(R) k |net) represents the feature input variable x i In the case of feature output category R k The conditional probability, P(net), is the feature input variable x in the DAG graph. i The probability under the joint of predecessor nodes.

[0078] (35) Furthermore, since the multiple feature input variables are not independent of each other, as shown in the figure Figure 2 As shown, the joint probability distribution of P(net) above is further calculated from the given CPT, and the calculation formula is:

[0079]

[0080] Among them, parent(x) i ) represents the feature input variable x i The predecessor node.

[0081] Step 4: Verify the calculated differential pressure rig rate using a partitioned validation set.

[0082] Step 5: Real-time application of the Bayesian belief network differential pressure stuck drill prediction model.

[0083] By inputting real-time data of the target drilling characteristics into the input variables, the model calculates and determines the probability of differential pressure stuck pipe. This embodiment provides drilling site personnel with more accurate and effective predictions of differential pressure stuck pipe, enabling real-time monitoring and alerts for such incidents. This helps technicians take timely and effective measures to prevent differential pressure stuck pipe from occurring, thereby improving the level of drilling operation risk management.

[0084] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting drilling rig efficiency based on Bayesian belief networks, characterized in that, Includes the following steps: S1. Collect historical drilling data of the target oil reservoir block, preprocess it to obtain sample data, and create a sample training set and test set based on the sample data; S2. Determine the feature input variables and feature output variables from the sample data; The feature input training tuple is represented by a vector X, denoted as X = (x1, x2, x3, x4, x5, x6); where x1 represents the riser pump pressure, x2 represents the formation lithology, x3 represents the drilling fluid density, x4 represents the rotary table rotation speed, x5 represents the rotary table torque, and x6 represents the drilling fluid filtrate loss. Let Y represent the feature output category tuple, denoted as Y = (R1, R2); where R1 represents the normal drilling state and R2 represents the tendency to stick due to differential pressure. S3. Using the sample training set, establish a differential pressure stuck drill training model based on a Bayesian belief network. S4. The calculated differential pressure rig rate results were validated using a validation set; Step S3 includes the following steps: S3-1. Determine the topological relationships between the feature input variables to form a directed acyclic graph; Based on the samples in the training set, the Naive Bayes algorithm is used to statistically calculate the prior probability P(X=x) of each feature input variable value. i ), (i = 1, 2, ..., 6) and the prior probability P(Y = R) of the output class. k The prior probability of the output class (k=1,2) is obtained using the following formula: ； Where |Y| represents the total number of training classes for the feature output, and |R| represents the total number of training classes. k ,Y| represents the number of training samples of category Rk in Y; S3-2: Based on historical drilling data of the target reservoir block, construct a conditional probability table and calculate the conditional probability P(X=x). i |Y=R k ); S3-3: Based on the conditional probability table and conditional probabilities, calculate the probability of differential pressure jamming under the joint probability distribution of the characteristic input variables: ； Where P(Y|X) represents the joint posterior probability of differential pressure stuck drill bit, P(Y=R k ) indicates the output category R k The prior probability, P(X) represents the joint feature probability of the feature input variables, and P(X|Y=Rk) represents the feature output category R. k The conditional probability of X, P(R) k |net) represents the feature input variable x i In the case of feature output category R k The conditional probability, P(net), is the feature input variable x in the DAG graph. i The probability under the joint of predecessor nodes; Since there are dependencies among the multiple feature input variables, the joint probability distribution of P(net) is further calculated from the given conditional probability table, and the formula is: ； Among them, parent(x) i ) represents the feature input variable x i The predecessor node.

2. The differential pressure drilling rig probability prediction method based on Bayesian belief networks according to claim 1, characterized in that, In step S1, the historical drilling data includes historical drilling data and stuck pipe cases.

3. The differential pressure drilling rig probability prediction method based on Bayesian belief networks according to claim 1, characterized in that, In step S1, preprocessing methods include data cleaning, integration, and / or transformation.

4. The differential pressure drilling rig probability prediction method based on Bayesian belief networks according to claim 1, characterized in that, In step S1, the sample data is divided into training set and test set by random sampling according to a preset ratio.

5. The differential pressure drilling rig rate prediction method based on Bayesian belief networks according to claim 1, characterized in that, In step S2, the feature output adopts the "approximation" output representation method, which classifies the probability of the feature output variable as an approximation problem.

6. The differential pressure drilling rig probability prediction method based on Bayesian belief networks according to claim 5, characterized in that, The "approximation" output representation method includes the following steps: taking 0.5 as the boundary, when the output probability value of R2 is located in (0.5, 1], it indicates that there is a tendency for differential pressure to jam the drill, and the closer the output value is to 1, the more obvious the tendency for differential pressure to jam the drill is. When the output value is in [0, 0.5), it indicates that the drilling operation is more likely to be in a normal drilling state than to be stuck. The closer the output value is to 0, the lower the probability of differential pressure stuck.

7. The differential pressure drilling rig probability prediction method based on Bayesian belief networks according to claim 1, characterized in that, It also includes step S5. The practical and real-time application of the Bayesian belief network differential pressure stuck drill prediction model: Input real-time data of the target drilling characteristics and variables, and the model calculates and determines the probability of differential pressure stuck drill bit.

8. A differential pressure rig rate prediction system based on Bayesian belief networks, used to implement the differential pressure rig rate prediction method based on Bayesian belief networks as described in any one of claims 1 to 7, characterized in that, include: Data acquisition and processing module: used to collect historical drilling data of the target oil reservoir block, preprocess it to obtain sample data, and create sample training and test sets based on the sample data; Feature input and output determination module: used to determine feature input variables and feature output variables from sample data; Model building module: used to build a differential pressure stuck drill training model based on Bayesian belief network using the sample training set; Model Validation Module: Used to validate the calculated differential pressure rig rate results using a validation set.

Images