An ultra-deep well downhole vibration monitoring and control method

Abstract

The application discloses a downhole vibration monitoring and control method for ultra-deep well, comprising the following steps: collecting vibration data near a drill bit which can represent drill string vibration intensity by using a measurement-while-drilling device from downhole, and obtaining drilling parameters which can represent drill string operation from uphole; after the drilling work in the specific time period is completed, the vibration data measured from downhole is obtained by tripping out, vibration data is classified and machine learning is performed, and meanwhile, the drilling parameters obtained from uphole are classified and machine learning is performed; in the drilling process in the subsequent time period, the vibration intensity of the drill string from downhole is judged by the drilling parameters from uphole. The method is to predict downhole operation conditions from ground drilling parameters, so as to prevent future premature equipment damage, reduce drilling risks and potential non-production time. The drilling operation can be guided by machine learning, and premature failure of downhole tools and wellbore instability can be avoided.

Description

Technical Field

[0001] This invention relates to the field of petroleum and geological drilling technology, specifically to a method for monitoring and controlling downhole vibration in ultra-deep wells. Background Technology

[0002] Drill string vibrations occur during downhole operation, which can lead to premature failure of downhole tools and wellbore instability, potentially causing serious downhole accidents. Real-time, high-frequency downhole vibration data acquisition and transmission are prohibitively expensive for widespread application. Relevant patents and literature in the existing technology are as follows:

[0003] Chinese patent ZL2015104156437, entitled "Downhole Vibration Monitoring Method," provides a method for visually and quantitatively determining the intensity of downhole vibration. Based on a mechanical model established using wellbore trajectory parameters, drill string assembly parameters, friction coefficient, mud performance parameters, and mechanical property parameters, the method calculates the drill string resonant frequency using a transfer matrix. It then generates the vibration intensity of the drill bit's stick-slip vibration using the wellhead rotation speed, wellhead torque fluctuation, and the drill string transfer matrix during stick-slip vibration. Finally, it generates the vibration intensity index of the drill bit's axial vibration using the drill string axial vibration resonant frequency, real-time drilling pressure, wellhead weight fluctuation, and the axial vibration drill string transfer matrix. This invention can determine the vibration state of the bottom hole drill string even without downhole measurement and information transmission equipment. However, this invention requires knowledge of a series of parameters, including wellbore trajectory parameters, drill string assembly parameters, friction coefficient, mud performance parameters, drill string mechanical property parameters, real-time wellhead rotation speed, real-time wellhead torque fluctuation, real-time drilling pressure, and real-time wellhead weight fluctuation. The transfer matrix of the entire drill string is obtained by using Newton's equations of motion and transfer matrices, and then the vibration intensity at the drill bit is determined. This requires collecting a lot of information and performing a lot of calculations, which has certain limitations in terms of drill string length.

[0004] Chinese patent ZL2018212458368, entitled "A Storage-Type Downhole Vibration Monitoring Device," provides a storage-type downhole vibration monitoring device, mainly comprising a body, vibration sensors, and a sealing cap. The body is a drill bit or a test section between the drill bit and the screw drill string. Sensor holes are opened radially along the body, and vibration sensors are tightly installed at the bottom. Battery holes, the same number as the sensors, are opened radially along the body. After being filled with high-temperature adhesive, the device is sealed with the sealing cap. This invention provides a device that can be installed between the drill bit and the screw drill string, eliminating the influence of the drill string between the monitoring device and the drill bit on the accuracy of vibration information, and accurately collecting vibration information from the drill bit. The data is stored downhole, avoiding the influence of communication technology on obtaining vibration data close to the drill bit. However, the method provided in this invention does not involve high-frequency downhole vibration acquisition and large-capacity storage. Furthermore, this method cannot process vibration information to predict downhole vibration, and the excessive number of holes in the device body affects the overall strength of the drill string.

[0005] In summary, existing technologies either utilize a large number of parameters to establish a continuous mechanical model along the drill string to monitor downhole vibration, or store vibration data by installing vibration measurement subs in the drill string assembly. These methods complicate downhole vibration monitoring systems in different ways.

[0006] Therefore, how to provide a method for monitoring and controlling downhole vibration in ultra-deep wells that can avoid the above-mentioned technical problems has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] This invention provides a method for monitoring and controlling downhole vibration in ultra-deep wells, comprising the following steps:

[0008] During a specific time period in the drilling process, vibration data that characterizes the drill string vibration intensity is collected from downhole using measurement-while-drilling equipment near the drill bit, and drilling parameters that characterize the operation of the drill string during the drilling process within the specific time period are obtained from the surface.

[0009] After the drilling work for this specific period is completed, the drilling is pulled out and the vibration data measured downhole is obtained. The vibration data is then subjected to classification machine learning, and the drilling parameters obtained above ground are also subjected to classification machine learning.

[0010] During subsequent drilling periods, the vibration intensity of the drill string downhole is determined by the drilling parameters on the surface.

[0011] Preferably, the vibration data includes the radial acceleration, tangential acceleration, and axial acceleration of the drill string, and the drilling parameters include the drill string's pressure on the drill string, torque, and rotational speed.

[0012] Preferably, the classification and machine learning of the vibration data includes the following steps:

[0013] The radial acceleration, tangential acceleration, and axial acceleration are preprocessed, and the vibration intensity is measured using the root mean square amplitude of the lateral acceleration and axial acceleration, as follows:

[0014]

[0015] The lateral acceleration is derived from the root mean square of the radial and tangential accelerations.

[0016] The preprocessed accelerations were classified using machine learning, employing a Gaussian Naive Bayes classifier.

[0017] The process of using the Gaussian Naive Bayes classifier for machine learning is as follows:

[0018] Gaussian Naive Bayes Classifier

[0019]

[0020] Simplify the equation

[0021]

[0022] It is a pre-classification estimated from the training data.

[0023] It is a probability distribution used to model data.

[0024] It is a constant in all instances.

[0025]

[0026] In this classification process, the extracted feature values ​​include five statistical measures: variance, mean, root mean square, peak value, and kurtosis. A time window of a certain length is selected on the time axis of the acquired signal to calculate the feature value at each time point. The five feature values ​​at each time point form a feature vector, which is used as the feature for each pattern in subsequent classification. Prior probabilities are manually defined in the acquired data beforehand during the classification process. The classification model is trained by identifying the n types of patterns and the proportion of each pattern in all data events, which serve as the prior probability of each pattern. The feature vectors are then categorized according to these prior probabilities. The feature matrix formed by the feature vectors of each pattern corresponds to the prior probability of that pattern. This prior knowledge is used to train the classification model, thus establishing its structure. In subsequent classifications, only comparisons are needed. The size of the probability can be used to make classification decisions. The largest one is selected as the output posterior probability. If the output probability of a certain pattern is the largest, then that pattern is classified. By repeating this process, all events in the collected data can be classified according to the manually determined patterns.

[0027] variance:

[0028] Mean:

[0029] Root mean square value:

[0030] Peak value:

[0031] kurtosis:

[0032] In the above formula This is the pre-processed acceleration value in the patent.

[0033] As a preferred approach, the process of classifying and machine learning the drilling parameters obtained from the well site is as follows:

[0034] After the ground parameters are collected, the five characteristic values ​​of the collected drilling pressure, torque, and rotation speed are calculated once: variance, mean, root mean square, peak value, and kurtosis. The five characteristic values ​​of these three drilling parameters are then arranged into a feature vector with fifteen elements.

[0035] The elements in the feature vector are:

[0036] [Drilling pressure (variance, mean, root mean square, peak value, kurtosis), torque (variance, mean, root mean square, peak value, kurtosis), rotational speed (variance, mean, root mean square, peak value, kurtosis)];

[0037] The method for obtaining the five feature values ​​(variance, mean, root mean square, peak value, and kurtosis) from the collected drilling pressure, torque, and rotational speed is the corresponding five formulas mentioned above. After obtaining the feature vector, the machine learning method is the same as the machine learning method for downhole parameters mentioned above, that is, to clarify the prior knowledge and establish a classification model.

[0038] Preferably, during subsequent drilling periods, the process of determining the vibration intensity of the downhole drill string using the drilling parameters above the well is as follows:

[0039] In the time domain, downhole acquired parameters, through classification analysis, clarify the occurrence of different vibration modes in the time domain; after extracting feature values, surface parameters can be arranged into a time series diagram in the time domain. The degree of overlap between the vibration modes expressed by downhole parameters and those expressed by surface parameters determines whether surface parameters can be used to determine downhole vibration. If the degree of overlap reaches a preset value, the vibration intensity of the drill string downhole can be determined by the drilling parameters above ground. If the degree of overlap does not reach the preset value, downhole data acquisition needs to continue, and vibration judgment is made using downhole data, while continuously correcting the above ground data.

[0040] Preferably, the vibration intensity is divided into levels i, ii, iii and iv, wherein in level i, the vibration intensity is less than or equal to 1.5; in level ii, the vibration intensity is greater than 1.5 and less than or equal to 2; in level iii, the vibration intensity is greater than 2 and less than or equal to 4; and in level iv, the vibration intensity is greater than 4.

[0041] The ultra-deep well downhole vibration monitoring and control method provided by this invention has the following technical effects:

[0042] This method predicts downhole operating conditions from surface drilling parameters. Specifically, it acquires downhole vibration data and surface drilling parameters within the same time period and performs classification and machine learning on them simultaneously. This method can help operators predict high-risk downhole vibrations and develop corresponding plans or proactive controls to prevent premature equipment damage, reduce drilling risks, and minimize potential non-productive time. Machine learning can guide drilling operations, avoiding premature downhole tool failure and wellbore instability.

[0043] As a preferred method, the Gaussian Naive Bayes classifier is used for machine learning, as it is simple and easy to implement.

[0044] Preferably, if the overlap reaches a preset value, the vibration intensity of the downhole drill string can be determined using the drilling parameters from the surface well. If the overlap does not reach the preset value, downhole data acquisition needs to continue, and vibration assessment is performed using the downhole data, continuously correcting the surface data. This method employs continuous machine learning until the overlap reaches the preset value, at which point the vibration intensity of the downhole drill string can be determined using the drilling parameters from the surface well, ensuring the accuracy of the assessment. Attached Figure Description

[0045] Figure 1 This is a schematic flowchart of the ultra-deep well downhole vibration monitoring and control method provided by the present invention. Detailed Implementation

[0046] like Figure 1 As shown, this invention provides a method for monitoring and controlling downhole vibration in ultra-deep wells, comprising the following steps:

[0047] During a specific time period in the drilling process, vibration data that characterizes the drill string vibration intensity is collected from downhole using measurement-while-drilling equipment near the drill bit, and drilling parameters that characterize the operation of the drill string during the drilling process within the specific time period are obtained from the surface.

[0048] After the drilling work for this specific period is completed, the drilling is pulled out and the vibration data measured downhole is obtained. The vibration data is then subjected to classification machine learning, and the drilling parameters obtained above ground are also subjected to classification machine learning.

[0049] During subsequent drilling periods, the vibration intensity of the drill string downhole is determined by the drilling parameters on the surface.

[0050] This method predicts downhole operating conditions from surface drilling parameters. Specifically, it acquires downhole vibration data and surface drilling parameters within the same time period and performs classification and machine learning on them simultaneously. This method can help operators predict high-risk downhole vibrations and develop corresponding plans or proactive controls to prevent premature equipment damage, reduce drilling risks, and minimize potential non-productive time. Machine learning can guide drilling operations, avoiding premature downhole tool failure and wellbore instability.

[0051] In one specific embodiment, the vibration data includes the radial acceleration, tangential acceleration, and axial acceleration of the drill string, and the drilling parameters include the drill string's drilling pressure, torque, and rotational speed.

[0052] In one specific embodiment, classifying the vibration data using machine learning includes the following steps:

[0053] The radial acceleration, tangential acceleration, and axial acceleration are preprocessed, and the vibration intensity is measured using the root mean square amplitude of the lateral acceleration and axial acceleration, as follows:

[0054]

[0055] The lateral acceleration is derived from the root mean square of the radial and tangential accelerations.

[0056] The preprocessed accelerations were classified using machine learning, employing a Gaussian Naive Bayes classifier.

[0057] The process of using the Gaussian Naive Bayes classifier for machine learning is as follows:

[0058] Gaussian Naive Bayes Classifier

[0059]

[0060] Simplify the equation

[0061]

[0062] It is a pre-classification estimated from the training data.

[0063] It is a probability distribution used to model data.

[0064] It is a constant in all instances.

[0065]

[0066] In this classification process, the extracted feature values ​​include five statistical measures: variance, mean, root mean square, peak value, and kurtosis. A time window of a certain length is selected on the time axis of the acquired signal to calculate the feature value at each time point. The five feature values ​​at each time point form a feature vector, which is used as the feature for each pattern in subsequent classification. Prior probabilities are manually defined in the acquired data beforehand during the classification process. The classification model is trained by identifying the n types of patterns and the proportion of each pattern in all data events, which serve as the prior probability of each pattern. The feature vectors are then categorized according to these prior probabilities. The feature matrix formed by the feature vectors of each pattern corresponds to the prior probability of that pattern. This prior knowledge is used to train the classification model, thus establishing its structure. In subsequent classifications, only comparisons are needed. The size of the probability can be used to make classification decisions. The largest one is selected as the output posterior probability. If the output probability of a certain pattern is the largest, then that pattern is classified. By repeating this process, all events in the collected data can be classified according to the manually determined patterns.

[0067] variance:

[0068] Mean:

[0069] Root mean square value:

[0070] Peak value:

[0071] kurtosis:

[0072] In the above formula This is the pre-processed acceleration value in the patent.

[0073] In this specific implementation, the process of classifying and machine learning the drilling parameters obtained from the well site is as follows:

[0074] After the ground parameters are collected, the five characteristic values ​​of the collected drilling pressure, torque, and rotation speed are calculated once: variance, mean, root mean square, peak value, and kurtosis. The five characteristic values ​​of these three drilling parameters are then arranged into a feature vector with fifteen elements.

[0075] The elements in the feature vector are:

[0076] [Drilling pressure (variance, mean, root mean square, peak value, kurtosis), torque (variance, mean, root mean square, peak value, kurtosis), rotational speed (variance, mean, root mean square, peak value, kurtosis)];

[0077] The method for obtaining the five feature values ​​(variance, mean, root mean square, peak value, and kurtosis) from the collected drilling pressure, torque, and rotational speed is to use the corresponding five formulas mentioned above to obtain the feature vector. After obtaining the feature vector, the machine learning method is the same as the machine learning method for downhole parameters mentioned above, that is, to clarify the prior knowledge and establish a classification model.

[0078] Using the Gaussian Naive Bayes classifier for machine learning is simple and easy.

[0079] Furthermore, during subsequent drilling operations, the process of determining the vibration intensity of the drill string downhole using the drilling parameters from the surface is as follows:

[0080] In the time domain, downhole acquired parameters, through classification analysis, clarify the occurrence of different vibration modes in the time domain; after extracting feature values, surface parameters can be arranged into a time series diagram in the time domain. The degree of overlap between the vibration modes expressed by downhole parameters and those expressed by surface parameters determines whether surface parameters can be used to determine downhole vibration. If the degree of overlap reaches a preset value, the vibration intensity of the drill string downhole can be determined by the drilling parameters above ground. If the degree of overlap does not reach the preset value, downhole data acquisition needs to continue, and vibration judgment is made using downhole data, while continuously correcting the above ground data.

[0081] If the overlap reaches a preset value, the vibration intensity of the drill string downhole can be determined using the drilling parameters from the surface well. If the overlap does not reach the preset value, downhole data acquisition needs to continue, and vibration assessment is performed using the downhole data, continuously correcting the surface well data. This method employs continuous machine learning until the overlap reaches the preset value, at which point the vibration intensity of the drill string downhole can be determined using the drilling parameters from the surface well, ensuring the accuracy of the assessment.

[0082] In one specific embodiment, the vibration intensity is divided into levels i, ii, iii, and iv, wherein in level i, the vibration intensity is less than or equal to 1.5; in level ii, the vibration intensity is greater than 1.5 and less than or equal to 2; in level iii, the vibration intensity is greater than 2 and less than or equal to 4; and in level iv, the vibration intensity is greater than 4.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for downhole vibration monitoring and control in ultra-deep wells, characterized in that, Includes the following steps: During a specific time period in the drilling process, vibration data that characterizes the drill string vibration intensity is collected from downhole using measurement-while-drilling equipment near the drill bit, and drilling parameters that characterize the operation of the drill string during the drilling process within the specific time period are obtained from the surface. After the drilling work for this specific period is completed, the drilling is pulled out and the vibration data measured downhole is obtained. The vibration data is then subjected to classification machine learning, and the drilling parameters obtained above ground are also subjected to classification machine learning. During subsequent drilling operations, the drilling parameters above the well are used to determine the vibration intensity of the drill string downhole. The vibration data includes the radial acceleration, tangential acceleration, and axial acceleration of the drill string, and the drilling parameters include the drill string's pressure on the drill string, torque, and rotational speed. The classification and machine learning of the vibration data includes the following steps: The radial acceleration, tangential acceleration, and axial acceleration are preprocessed, and the vibration intensity is measured using the root mean square amplitude of the lateral acceleration and axial acceleration, as follows: The lateral acceleration is derived from the root mean square of the radial and tangential accelerations. The preprocessed accelerations were classified using machine learning, employing a Gaussian Naive Bayes classifier. The process of using the Gaussian Naive Bayes classifier for machine learning is as follows: Gaussian Naive Bayes Classifier Simplify the equation It is a pre-classification estimated from the training data. It is a probability distribution used to model data. It is a constant in all instances. In this classification process, the extracted feature values ​​include five statistical measures: variance, mean, root mean square, peak value, and kurtosis. A time window of a certain length is selected on the time axis of the acquired signal to calculate the feature value at each time point. The five feature values ​​at each time point form a feature vector, which is used as the feature for each pattern in subsequent classification. Prior probabilities are manually defined in the acquired data beforehand during the classification process. The classification model is trained by identifying the n types of patterns and the proportion of each pattern in all data events, which serve as the prior probability of each pattern. The feature vectors are then categorized according to these prior probabilities. The feature matrix formed by the feature vectors of each pattern corresponds to the prior probability of that pattern. This prior knowledge is used to train the classification model, thus establishing its structure. In subsequent classifications, only comparisons are needed. The size of the probability can be used to make classification decisions. The largest one is selected as the posterior probability of the output. If the output probability of a certain pattern is the largest, then that pattern is classified. By repeating this process, all events in the collected data can be classified according to the manually determined patterns. variance: Mean: Root mean square value: Peak value: kurtosis: In the above formula This refers to the pre-processed acceleration value in the patent. During subsequent drilling periods, the process of determining the vibration intensity of the drill string downhole using the drilling parameters from the surface is as follows: In the time domain, downhole acquired parameters, through classification analysis, clarify the occurrence of different vibration modes in the time domain; after extracting feature values, surface parameters can be arranged into a time series diagram in the time domain. The degree of overlap between the vibration modes expressed by downhole parameters and those expressed by surface parameters determines whether surface parameters can be used to determine downhole vibration. If the degree of overlap reaches a preset value, the vibration intensity of the drill string downhole can be determined by the drilling parameters above ground. If the degree of overlap does not reach the preset value, downhole data acquisition needs to continue, and vibration judgment is made using downhole data, while continuously correcting the above ground data.

2. The method for monitoring and controlling downhole vibration in ultra-deep wells according to claim 1, characterized in that, The process of classifying drilling parameters obtained from the well site using machine learning is as follows: After collecting surface parameters, the five feature values ​​(variance, mean, root mean square, peak value, and kurtosis) of the collected drilling pressure, torque, and rotational speed are calculated once. These five feature values ​​are then arranged into a feature vector with fifteen elements. The elements in the feature vector are: variance, mean, root mean square, peak value, and kurtosis of drilling pressure; variance, mean, root mean square, peak value, and kurtosis of torque; and variance, mean, root mean square, peak value, and kurtosis of rotational speed. The method for calculating these five feature values—variance, mean, root mean square, peak value, and kurtosis—of the collected drilling pressure, torque, and rotational speed is to use the corresponding five formulas in claim 1. After obtaining the feature vector, the machine learning method is the same as the machine learning method for downhole parameters in claim 1, which is to clarify the prior knowledge and establish a classification model.

3. The method for monitoring and controlling downhole vibration in ultra-deep wells according to claim 1, characterized in that, The vibration intensity is divided into levels i, ii, iii and iv. In level i, the vibration intensity is less than or equal to 1.5; in level ii, the vibration intensity is greater than 1.5 and less than or equal to 2; in level iii, the vibration intensity is greater than 2 and less than or equal to 4; and in level iv, the vibration intensity is greater than 4.

Images