Virtual angle sensor online measurement method based on graph neural network

Abstract

The invention discloses a virtual angle sensor online measurement method based on a graph neural network. The method comprises the following steps: acquiring an environment magnetic field signal of a magnetometer and a motion attitude signal of an inertial sensor; training to obtain a graph neural network model capable of predicting the incidence relation between the environment magnetic field signal and the motion posture signal; when a drill bit is at a low speed in a drilling detection interval, inputting a collected motion attitude signal of an inertial sensor into the graph neural network model for reverse prediction, and performing compensation calibration on an environment magnetic field signal detected by a magnetometer; and when the drill bit is at a high speed in the drilling detection interval, measuring the drilling environment magnetic field of the drill bit by using the compensated and calibrated magnetometer as the input of the corrected and trained graph neural network model so as to predict the motion attitude of the output drill bit, and correcting the drift error of the inertial sensor. According to the invention, mutual correction of the magnetometer and the inertial sensor is realized, and the problem that errors exist in the magnetometer and the inertial sensor during measurement while drilling is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of measurement while drilling, and in particular to an online measurement method of a virtual angle sensor based on a graph neural network. Background Art

[0002] In the oil and gas drilling process, measurement while drilling (MWD) is a key technology that allows real-time measurement of wellbore parameters such as azimuth, well inclination, tool face, etc. during the drilling process.

[0003] Among the current measurement methods, most are magnetometer measurements. Magnetometers rely on the earth's magnetic field to measure the relative orientation of the borehole trajectory. They are usually used for continuous orientation measurements. However, when encountering magnetic materials during drilling or in an environment with large magnetic interference, the output of the magnetometer may produce errors due to changes in the magnetic properties of the formation, the magnetism of the drill bit, and other electromagnetic interference. Gyroscopes can be used to measure the true north orientation in the borehole trajectory. They do not rely on the earth's magnetic field, so the gyroscope's measurement will not be affected by changes in the magnetic properties of the formation. However, gyroscopes have drift errors that accumulate over time and are not suitable for long-term and dynamic measurements.

[0004] Furthermore, the rapid development of artificial intelligence has attracted widespread attention in various fields. Deep learning is a machine learning technology. The core idea of ​​deep learning is to transform the original data layer by layer and gradually abstract higher-level feature representations. During the training process, the model continuously adjusts the connection weights of each layer in the network by learning the inherent laws and characteristics of the data. Ultimately, the model can accurately predict or classify new input data. Common deep learning models include convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory networks (LSTM), graph neural networks (GNN), etc.

[0005] Therefore, how to reduce the error of the magnetometer output in the presence of magnetic materials or in an environment with large magnetic interference, reduce the drift error accumulated by the gyroscope over time, and improve the accuracy of dynamic measurement during drilling measurement has become a problem that needs further research and solution. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides an online measurement method for a virtual angle sensor based on a graph neural network, which utilizes an inertial sensor and a magnetometer, and realizes mutual correction of the magnetometer and the inertial sensor through a graph neural network model that can predict the correlation between the environmental magnetic field signal and the motion posture signal, thereby ensuring accurate prediction of the drill tool attitude angle under different magnetic field environments, and solving the problem of errors in the measurement of the magnetometer and the inertial sensor in the measurement while drilling.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A virtual angle sensor online measurement method based on graph neural network includes the following steps:

[0009] In the model training stage, magnetometers and inertial sensors are used to measure the ambient magnetic field and motion posture of the drill bit under different magnetic field strength environments, and the ambient magnetic field signals of the magnetometer and the motion posture signals of the inertial sensor are collected. The ambient magnetic field signals and motion posture signals collected under different magnetic field strength environments are used as the input of the graph neural network, and a graph neural network model that can predict the correlation between the ambient magnetic field signals and the motion posture signals is trained.

[0010] During the online measurement phase of drilling work, a drilling detection interval is preset, and in each drilling detection interval, the drill bit is controlled to first enter the detection speed rotation working phase, and then enter the super detection speed rotation working phase;

[0011] During the drilling detection interval, when the drill bit is in the detection speed rotation working stage, the magnetometer and inertial sensor are used to measure the environmental magnetic field of the drill bit and the motion posture of the drill bit respectively for while-drilling measurement, and the motion posture signal of the collected inertial sensor is input into the graph neural network model for reverse prediction, and the environmental magnetic field prediction value obtained by reverse prediction is used to compensate and calibrate the environmental magnetic field signal detected by the magnetometer;

[0012] When the drill bit is in the super-detection speed rotation working stage in the drilling detection interval, a compensated and calibrated magnetometer is used to measure the environmental magnetic field of the drill bit, and the environmental magnetic field signal of the magnetometer is collected as the input of the calibrated and trained graph neural network model to predict the output motion posture of the drill bit, which is recorded as the motion posture information for online while-drilling measurement of the drill bit in the super-detection speed rotation working stage; and, based on the motion posture of the drill bit predicted and output by the graph neural network model, the drift error of the inertial sensor is corrected to ensure the accuracy of the motion posture detection of the inertial sensor in the next drilling detection interval;

[0013] Thus, the online while-drilling measurement of the motion posture of the drill bit in different drilling detection intervals during the online measurement process is realized.

[0014] As a preferred solution, during the model training phase, the collected environmental magnetic field signals and motion posture signals are also preprocessed to remove invalid, erroneous and repeated signals, and then used as input for training of the graph neural network.

[0015] As a preferred solution, during the online measurement stage of drilling work, when the drill bit is in the detection speed rotation working stage, the working speed of the drill bit is within the detection speed requirement range of the inertial sensor; when the drill bit is in the super-detection speed rotation working stage, the working speed of the drill bit is higher than the upper limit value of the detection speed requirement range of the inertial sensor.

[0016] As a preferred solution, during the online measurement stage of drilling work, the specific method of using the environmental magnetic field prediction value obtained by reverse prediction to compensate and calibrate the environmental magnetic field signal detected by the magnetometer is:

[0017] The compensation calibration is completed by adjusting the value of the ambient magnetic field signal detected by the magnetometer to be equal to the ambient magnetic field prediction value obtained by reverse prediction.

[0018] As a preferred solution, during the online measurement stage of drilling work, the specific process of correcting the drift error of the inertial sensor based on the motion posture of the drill bit predicted and output by the graph neural network model includes:

[0019] S201, acquiring the motion posture of the inertial sensor under static conditions according to the motion posture signal of the inertial sensor collected, and solving to obtain the initial roll angle and initial pitch angle of the drill bit;

[0020] S202, determining a horizontal mathematical plane according to an initial roll angle and an initial pitch angle of the drill bit, acquiring a motion attitude signal of an inertial sensor under dynamic conditions, and calculating a motion attitude of the drill bit in a carrier coordinate system;

[0021] S203, obtaining an attitude conversion matrix according to a conversion relationship between a quaternion attitude differential equation and the motion attitude of the drill bit in a carrier coordinate system;

[0022] S204. Using the attitude conversion matrix, the motion attitude of the drill bit in the navigation coordinate system predicted and output by the graph neural network model is converted into the motion attitude of the drill bit in the carrier coordinate system, which is used as the error correction value of the motion attitude detection of the inertial sensor to complete the correction of the drift error of the inertial sensor.

[0023] As a preferred solution, in step S201, the solution formula for obtaining the initial roll angle and initial pitch angle of the drill bit is:

[0024]

[0025] Where β is the initial pitch angle, γ is the initial roll angle, and a x 、a y 、a z They are the measured values ​​of the accelerometer along the wingspan direction of the aircraft, perpendicular to the ground and in the vertical direction respectively, and g is the acceleration due to gravity.

[0026] As a preferred solution, in step S202, the specific process of calculating the motion posture of the drill bit in the carrier coordinate system includes:

[0027] S2021, controlling the drill motor to rotate to the 0° and 180° positions to stabilize, and collecting angular velocity data output by the inertial sensor under static conditions;

[0028] S2022. Based on the determined horizontal mathematical plane, the collected output data of the inertial sensor is converted into horizontal plane output data reflecting the posture of the drill bit relative to the horizontal plane, so as to calculate the motion posture of the drill bit in the carrier coordinate system.

[0029] As a preferred solution, the solution formula for the motion posture of the drill bit in the carrier coordinate system is:

[0030]

[0031] Where α is the azimuth of the drill bit in the carrier coordinate system, ω1 and ω2 are the angular velocities of the static output of the inertial sensor at the two positions of 0° and 180° respectively, and ω ie is the angular velocity of the earth's rotation, and L is the geographical latitude.

[0032] As a preferred solution, in step S203, the expression of the posture conversion matrix is:

[0033]

[0034] In the formula, is the attitude transformation matrix, They are the inclination angle, azimuth angle and tool face angle of the drill bit in the navigation coordinate system.

[0035] As a preferred solution, in step S204, the formula for converting the motion posture of the drill bit in the navigation coordinate system predicted and output by the graph neural network model into the motion posture of the drill bit in the carrier coordinate system is:

[0036]

[0037] In the formula, (x b ,y b , z b ) is the carrier coordinate system, (x n ,y n , z n ) is the motion posture of the drill bit in the navigation coordinate system predicted and output by the graph neural network model.

[0038] Compared with the prior art, the present invention has the following technical effects:

[0039] (1) The present invention divides the drilling detection interval into a detection speed rotation working stage and a super-detection speed rotation working stage. When the drill bit is in the detection speed rotation working stage, the motion posture signal of the inertial sensor is collected, and the predicted environmental magnetic field prediction value is obtained through reverse prediction of the graph neural network model to compensate and calibrate the environmental magnetic field signal detected by the magnetometer, thereby reducing the error of the magnetometer caused by formation magnetic changes, drilling tool magnetism and electromagnetic interference, improving the accuracy of the drill bit motion posture measurement, and increasing the robustness; when the drill bit is in the super-detection speed rotation working stage, the calibrated environmental magnetic field is used to predict the output of the drill bit's motion posture through the graph neural network model to correct the drift error of the inertial sensor, thereby ensuring the accuracy of measurement under high-speed rotation conditions, reducing the drift error accumulated due to long-term operation of the inertial sensor, and improving the reliability and accuracy of the drill bit posture measurement.

[0040] (2) Based on graph neural networks, the present invention effectively constructs the complex relationship between magnetometer and inertial sensor data, performs feature learning on sensor data through deep learning, and uses the learned features and relationships to achieve real-time measurement of the drill tool motion posture, thereby improving the reliability of the drill bit attitude angle. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0042] Figure 1 A flowchart of an online measurement method of a virtual angle sensor based on a graph neural network disclosed in the present invention;

[0043] Figure 2 It is a schematic diagram of the output of the magnetometer and gyroscope in polar coordinates when there is no interference in this embodiment;

[0044] Figure 3 It is a schematic diagram of the output of the magnetometer and gyroscope under interference in this embodiment;

[0045] Figure 4 This is the GNN network structure diagram of this embodiment;

[0046] Figure 5 This is a flowchart of the virtual angle sensor in this embodiment;

[0047] Figure 6 Schematic diagram of online measurement in this embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but only represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0049] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0050] like Figure 1 As shown, the present invention discloses an online measurement method of a virtual angle sensor based on a graph neural network, comprising the following steps:

[0051] In the model training stage, magnetometers and inertial sensors are used to measure the ambient magnetic field and motion posture of the drill bit under different magnetic field strength environments, and the ambient magnetic field signals of the magnetometer and the motion posture signals of the inertial sensor are collected. The ambient magnetic field signals and motion posture signals collected under different magnetic field strength environments are used as the input of the graph neural network, and a graph neural network model that can predict the correlation between the ambient magnetic field signals and the motion posture signals is trained.

[0052] During the online measurement phase of drilling work, a drilling detection interval is preset, and in each drilling detection interval, the drill bit is controlled to first enter the detection speed rotation working phase, and then enter the super detection speed rotation working phase;

[0053] During the drilling detection interval, when the drill bit is in the detection speed rotation working stage, the magnetometer and inertial sensor are used to measure the environmental magnetic field of the drill bit and the motion posture of the drill bit respectively for while-drilling measurement, and the motion posture signal of the collected inertial sensor is input into the graph neural network model for reverse prediction, and the environmental magnetic field prediction value obtained by reverse prediction is used to compensate and calibrate the environmental magnetic field signal detected by the magnetometer;

[0054] When the drill bit is in the super-detection speed rotation working stage in the drilling detection interval, a compensated and calibrated magnetometer is used to measure the environmental magnetic field of the drill bit, and the environmental magnetic field signal of the magnetometer is collected as the input of the calibrated and trained graph neural network model to predict the output motion posture of the drill bit, which is recorded as the motion posture information for online while-drilling measurement of the drill bit in the super-detection speed rotation working stage; and, based on the motion posture of the drill bit predicted and output by the graph neural network model, the drift error of the inertial sensor is corrected to ensure the accuracy of the motion posture detection of the inertial sensor in the next drilling detection interval;

[0055] Thus, the online while-drilling measurement of the motion posture of the drill bit in different drilling detection intervals during the online measurement process is realized.

[0056] Among the existing measurement while drilling methods, most are magnetometer measurements. However, the output of the magnetometer will produce errors due to changes in the magnetic properties of the formation, the magnetism of the drill bit, and other electromagnetic interference. Although the measurement of the gyroscope will not be affected by changes in the magnetic properties of the formation, the gyroscope has the problem of drift errors accumulated over time. In view of the above problems and shortcomings, the present invention proposes an online measurement method of a virtual angle sensor based on a graph neural network, which combines an inertial sensor and a magnetometer, predicts and outputs a predicted value of the environmental magnetic field through a graph neural network model, and compensates and calibrates the environmental magnetic field signal detected by the magnetometer; the predicted drill bit motion attitude angle during high-speed rotation is used to correct the drift error of the inertial sensor. Aiming at real-time working conditions, the present invention achieves the purpose of online measurement by designing a cyclic measurement of low-speed north seeking, model training, and high-speed prediction within the drilling interval, thereby solving the problem of errors in the measurement of magnetometers and gyroscopes in measurement while drilling.

[0057] This embodiment discloses an online measurement method of a virtual angle sensor based on a graph neural network. In the specific implementation, firstly, taking the horizontal and static state (Z axis is perpendicular to the horizontal plane) as an example, the magnetic field is stable in a non-magnetic interference environment, and the three-dimensional magnetic perception of the magnetometer is:

[0058]

[0059] In the formula, B b is the three-dimensional magnetic perception of the magnetometer, H i is the local geomagnetic intensity, is the azimuth;

[0060] The three-dimensional gyroscope perception is:

[0061]

[0062] In the formula, ω b is the angular velocity sensed by the three-dimensional gyroscope, ω ieis the angular velocity of the earth's rotation, L is the geographical latitude;

[0063] Analysis of the three-dimensional perception of the magnetometer and gyroscope shows that their perception of the geomagnetic field and the angular velocity of the earth's rotation corresponds to There is a sine and cosine law. If polar coordinates are used, it can be expressed as a standard circle, such as Figure 2 As shown in Figure 1, there is a certain corresponding relationship between the magnetometer output and the gyroscope output signal: B = f(ω).

[0064] However, in the case of magnetic interference, the output "circle" (magnetic field strength) will "deform" (change), the center of the circle will shift, and the major and minor axes will rotate, which will increase the error in the azimuth angle calculation and even change the image from a circle to a slant line, making it impossible to calculate. At the same time, the output curve of the gyroscope will also change similarly when it is vibrated or under conditions of time-temperature drift. The output of the magnetometer and gyroscope under interference is as follows: Figure 3 Although the magnet and gyroscope sense different physical quantities, the accelerometer is used to establish a horizontal platform, and both can calculate the azimuth angle through coordinate changes, and the calculation results of the two are consistent.

[0065] Based on the analysis of the similarity of the characteristic performance of the original output of the sensor output under the interference of error disturbance and the similarity of the solution results in one rotation, a virtual angle sensor online measurement method based on graph neural network is proposed. The method specifically includes:

[0066] S1, model training stage:

[0067] During the model training phase, magnetometers and inertial sensors are used to measure the ambient magnetic field of the drill bit and the motion posture of the drill bit in different magnetic field strength environments, and the ambient magnetic field signal of the magnetometer and the motion posture signal of the inertial sensor are collected; the inertial sensor includes a gyroscope and an accelerometer.

[0068] In a specific implementation, after collecting the environmental magnetic field signal of the magnetometer and the motion posture signal of the inertial sensor, the method further includes: preprocessing the collected environmental magnetic field signal and motion posture signal to remove invalid, erroneous and repeated signals. The processed signal data is divided into a training set, a verification set and a test set, and the processed output signal data is normalized and standardized.

[0069] Then, the ambient magnetic field signals and motion posture signals collected under different magnetic field strength environments are used as the input of the graph neural network, and a graph neural network model that can predict the correlation between the ambient magnetic field signals and motion posture signals is trained.

[0070] In the specific implementation, model training is performed based on the collected environmental magnetic field signals and motion posture signals to obtain a graph neural network model that can predict the correlation between environmental magnetic field signals and motion posture signals. The graph neural network is trained using environmental magnetic field signals and motion posture signals collected under different magnetic field strength environments, and the collected environmental magnetic field signals are used as the input of the graph neural network to predict the motion posture of the output drill bit, and the collected motion posture signals are used as prediction verification data to train a graph neural network model that can predict the correlation between environmental magnetic field signals and motion posture signals.

[0071] In the present invention, a training model of a graph neural network (GNN) is selected based on the advantage that the GNN can effectively capture the complex relationships and interdependencies between nodes. Figure 4 As shown in the figure, Graph Neural Network (GNN) is a deep learning model specifically designed to process graph structured data. Graph neural networks can effectively capture the complex relationships and interdependencies between nodes (vertices in the graph) by operating directly on the graph structure. The graph consists of nodes (representing the output of each sensor) and edges (representing the relationship between sensors, physical model). Each node can have a feature vector that describes the properties of the node. The state of the node is updated through the process of message passing. In this process, each node receives information from its neighboring nodes, and then updates its own feature representation based on this information and its own state.

[0072] In the present invention, each node of the GNN represents a sensor data type, such as an acceleration node, an angular velocity node, a magnetic field strength node, etc. Such a design allows the network to intuitively represent the interactions and dependencies between different data types. The edges between the nodes represent the physical connections or data dependencies between these sensor data. These connections can be defined based on physical formulas or known data correlations. For example, the sinusoidal relationship between the outputs of the magnetometer and gyroscope derived above, the relationship between acceleration and angle based on kinematic formulas, the relationship between magnetic field strength and angle, etc. For each node, a suitable feature vector needs to be defined to represent the sensor data. This may include raw data values, statistical characteristics of the data, or preprocessed features.

[0073] In the specific implementation, during the training process, steps such as hyperparameter adjustment and model validation are required to optimize the performance and generalization ability of the model. The performance of the model is evaluated through methods such as cross-validation and test set validation, and the model is adjusted and optimized to improve the accuracy and stability of the model in practical applications.

[0074] S2, the online measurement stage of drilling work,

[0075] During the online measurement phase of drilling work, a drilling detection interval is preset, and in each drilling detection interval, the drill bit is controlled to first enter the detection speed rotation working phase and then enter the super detection speed rotation working phase. When the drill bit is in the detection speed rotation working phase, the working speed of the drill bit is within the detection speed requirement range of the inertial sensor; when the drill bit is in the super detection speed rotation working phase, the working speed of the drill bit is higher than the upper limit of the detection speed requirement range of the inertial sensor.

[0076] During the drilling detection interval, when the drill bit is in the detection speed rotation working stage, the magnetometer and inertial sensor are used to measure the ambient magnetic field of the drill bit and the motion posture of the drill bit respectively for while-drilling measurement, and the motion posture signal of the collected inertial sensor is input into the graph neural network model for reverse prediction. The ambient magnetic field prediction value obtained by reverse prediction is used to compensate and calibrate the ambient magnetic field signal detected by the magnetometer, such as Figure 5 shown.

[0077] In the specific implementation, during the online measurement stage of drilling work, when the drill bit is in the detection speed rotation stage, the specific method of using the ambient magnetic field prediction value obtained by inverse prediction to compensate and calibrate the ambient magnetic field signal detected by the magnetometer is: adjust the value of the ambient magnetic field signal detected by the magnetometer to be equal to the ambient magnetic field prediction value obtained by inverse prediction, that is, complete the compensation calibration.

[0078] Here, in the detection speed rotation working stage in the previous drilling detection interval, the environmental magnetic field prediction value obtained by reverse prediction is used to calibrate the environmental magnetic field signal detected by the magnetometer, the purpose is to overcome the interference of the current environment on the magnetometer, reduce the error of the magnetometer caused by the magnetic change of the formation, the magnetism of the drilling tool and electromagnetic interference, improve the accuracy of the drill bit motion attitude measurement, and increase the robustness. After the magnetometer is corrected in this way, when entering the super detection speed rotation working stage after entering the current drilling detection interval, the environmental magnetic field signal detected by the magnetometer can be regarded as an accurate value after eliminating environmental interference.

[0079] When the drill bit is in the super-detection speed rotation working stage during the drilling detection interval, a compensated and calibrated magnetometer is used to measure the ambient magnetic field of the drill bit, and the ambient magnetic field signal of the magnetometer is collected as the input of the calibrated and trained graph neural network model to predict the output motion posture of the drill bit, which is recorded as the motion posture information for online while-drilling measurement of the drill bit during the super-detection speed rotation working stage; and, based on the motion posture of the drill bit predicted and output by the graph neural network model, the drift error of the inertial sensor is corrected to ensure the accuracy of the motion posture detection of the inertial sensor in the next drilling detection interval, such as Figure 5 shown.

[0080] In the specific implementation, when the drill bit is in the stage of rotating at a speed beyond the detection speed, the calibrated ambient magnetic field is used to predict the output motion posture of the drill bit through the graph neural network model to correct the drift error of the inertial sensor, ensuring the accuracy of measurement under high-speed rotation conditions, reducing the drift error accumulated due to the long-term operation of the inertial sensor, and improving the reliability and accuracy of the drill bit posture measurement.

[0081] In the specific implementation, during the online measurement stage of drilling work, when the drill bit is in the stage of rotating at an over-detection speed, the drift error of the inertial sensor is corrected based on the motion posture of the drill bit predicted and output by the graph neural network model. The specific process includes:

[0082] S201, acquiring the motion posture of the inertial sensor under static conditions according to the collected motion posture signal of the inertial sensor, and solving and obtaining the initial roll angle and initial pitch angle of the drill bit.

[0083] The solution formula for the initial roll angle and initial pitch angle of the drill bit is:

[0084]

[0085] In the navigation coordinate system, the accelerometer f n The values ​​are:

[0086] n =[0 0g]

[0087] In the carrier coordinate system, the accelerometer f b The values ​​are:

[0088]

[0089] Where β is the initial pitch angle, γ is the initial roll angle, and a x 、a y 、a z are the measured values ​​of the accelerometer along the wingspan direction of the aircraft, perpendicular to the ground and in the vertical direction, respectively, and g is the acceleration due to gravity;

[0090] S202, determining a horizontal mathematical plane according to an initial roll angle and an initial pitch angle of the drill bit, acquiring a motion attitude signal of an inertial sensor under dynamic conditions, and calculating the motion attitude of the drill bit in a carrier coordinate system.

[0091] In this step, the specific process of calculating the motion posture of the drill bit in the carrier coordinate system includes:

[0092] S2021. Control the drill motor to rotate to the 0° and 180° positions to stabilize, and collect angular velocity data output by the inertial sensor under static conditions.

[0093] S2022. Based on the determined horizontal mathematical plane, the collected output data of the inertial sensor is converted into horizontal plane output data reflecting the posture of the drill bit relative to the horizontal plane, eliminating the influence of gravity, so as to solve the movement posture of the drill bit in the carrier coordinate system.

[0094] The solution formula for the drill bit's motion posture in the carrier coordinate system is:

[0095]

[0096] Where α is the azimuth of the drill bit in the carrier coordinate system, ω1 and ω2 are the angular velocities of the static output of the inertial sensor at the two positions of 0° and 180° respectively, and ω ie is the angular velocity of the earth's rotation, L is the geographical latitude;

[0097] S203, obtaining an attitude conversion matrix according to the conversion relationship between the quaternion attitude differential equation and the motion attitude of the drill bit in the carrier coordinate system.

[0098] After the system is initially aligned, it enters the navigation phase. The three attitude angles of the MWD are usually solved using the quaternion attitude differential equation in the quaternion attitude update algorithm. The quaternion attitude differential equation in the quaternion attitude update algorithm is:

[0099]

[0100] In the formula, is the differential of the quaternion Q, where Q is a quaternion, is the angular velocity in the inertial sensor carrier coordinate system. By converting the quaternion attitude to the rotation matrix, the updated attitude transformation matrix can be obtained. Thus, the conversion and updating operation matrix of azimuth, well inclination and tool face angle can be obtained.

[0101] By decomposing the quaternion attitude differential equation and the conversion relationship between the motion attitude of the drill bit in the carrier coordinate system, the expression of the attitude conversion matrix is ​​obtained as follows:

[0102]

[0103] In the formula, is the attitude transformation matrix, They are the inclination angle, azimuth angle and tool face angle of the drill bit in the navigation coordinate system.

[0104] S204. Using the attitude conversion matrix, the motion attitude of the drill bit in the navigation coordinate system predicted and output by the graph neural network model is converted into the motion attitude of the drill bit in the carrier coordinate system, which is used as the error correction value of the motion attitude detection of the inertial sensor to complete the correction of the drift error of the inertial sensor.

[0105] Among them, the formula for converting the motion posture of the drill bit in the navigation coordinate system predicted by the graph neural network model into the motion posture of the drill bit in the carrier coordinate system is:

[0106]

[0107] In the formula, (x b ,y b , z b ) is the carrier coordinate system, (x n ,y n , z n ) is the motion posture of the drill bit in the navigation coordinate system predicted and output by the graph neural network model.

[0108] As an example, Figure 6 As shown in the figure, in order to achieve the purpose of online measurement during specific implementation, it is assumed that a drilling detection interval m is set. When the drill bit is in the detection speed rotation working stage, the three-axis magnetometer, gyroscope, and accelerometer are used to measure the drill bit at a low speed, and the neural network is used for learning. First, the angle measured by the combination of the low-speed rotating gyroscope and accelerometer is used to correct the magnetometer signal. Since the degree of change in the magnetic field intensity is relatively stable, when the drill bit rotates at a high speed, the input magnetometer data can obtain a set of attitude angles. When the next m-length interval arrives, low-speed rotation, north search and deep learning are performed again to ensure the slow change of the magnetic field environment. Then, the drill tool motion attitude is obtained through the magnetometer output during high-speed rotation. The above steps are repeated to achieve the purpose of online real-time measurement. In actual working conditions, due to the uncertainty of the downhole environment, the value of m can be determined according to the magnetic field environment conditions detected in advance.

[0109] In summary, the present invention divides the drilling detection interval into a detection speed rotation working stage and a super detection speed rotation working stage. When the drill bit is in the detection speed rotation working stage, the motion attitude signal of the inertial sensor collected is used to reversely predict the predicted environmental magnetic field predicted value through the graph neural network model to compensate and calibrate the environmental magnetic field signal detected by the magnetometer, thereby reducing the error of the magnetometer caused by the magnetic change of the formation, the magnetism of the drill tool and electromagnetic interference, improving the accuracy of the drill bit motion attitude measurement, and increasing the robustness; when the drill bit is in the super detection speed rotation working stage, the calibrated environmental magnetic field is used to predict the output of the motion attitude of the drill bit through the graph neural network model to correct the drift error of the inertial sensor, ensuring the accuracy of measurement under high-speed rotation conditions, reducing the drift error accumulated due to the long-term operation of the inertial sensor, and improving the reliability and accuracy of the drill bit attitude measurement. Based on the graph neural network, the present invention effectively constructs the complex relationship between the magnetometer and the inertial sensor data, performs feature learning on the sensor data through deep learning, and uses the learned features and relationships to realize the real-time measurement of the drill tool motion attitude, and improve the reliability of the drill bit attitude angle.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes may be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A virtual angle sensor online measurement method based on graph neural network, characterized in that: It includes the model training phase and the online measurement phase of drilling work; In the model training stage, magnetometers and inertial sensors are used to measure the ambient magnetic field and motion posture of the drill bit under different magnetic field strength environments, and the ambient magnetic field signals of the magnetometer and the motion posture signals of the inertial sensor are collected. The ambient magnetic field signals and motion posture signals collected under different magnetic field strength environments are used as the input of the graph neural network, and a graph neural network model that can predict the correlation between the ambient magnetic field signals and the motion posture signals is trained. During the online measurement phase of drilling work, a drilling detection interval is preset, and in each drilling detection interval, the drill bit is controlled to first enter the detection speed rotation working phase, and then enter the super detection speed rotation working phase; During the drilling detection interval, when the drill bit is in the detection speed rotation working stage, the magnetometer and inertial sensor are used to measure the environmental magnetic field of the drill bit and the motion posture of the drill bit respectively for while-drilling measurement, and the motion posture signal of the collected inertial sensor is input into the graph neural network model for reverse prediction, and the environmental magnetic field prediction value obtained by reverse prediction is used to compensate and calibrate the environmental magnetic field signal detected by the magnetometer; When the drill bit is in the super-detection speed rotation working stage in the drilling detection interval, a compensated and calibrated magnetometer is used to measure the environmental magnetic field of the drill bit, and the environmental magnetic field signal of the magnetometer is collected as the input of the calibrated and trained graph neural network model to predict the output motion posture of the drill bit, which is recorded as the motion posture information for online while-drilling measurement of the drill bit in the super-detection speed rotation working stage; and, based on the motion posture of the drill bit predicted and output by the graph neural network model, the drift error of the inertial sensor is corrected to ensure the accuracy of the motion posture detection of the inertial sensor in the next drilling detection interval; Thus, the online while-drilling measurement of the motion posture of the drill bit in different drilling detection intervals during the online measurement process is realized.

2. The online measurement method of virtual angle sensor based on graph neural network according to claim 1 is characterized in that: During the model training phase, the collected environmental magnetic field signals and motion posture signals are also preprocessed to remove invalid, erroneous and repeated signals, and then used as input for training of the graph neural network.

3. The online measurement method of virtual angle sensor based on graph neural network according to claim 1 is characterized in that: During the online measurement phase of drilling work, when the drill bit is in the detection speed rotation working phase, the working speed of the drill bit is within the detection speed requirement range of the inertial sensor; When the drill bit is in the stage of rotating at a speed exceeding the detection speed, the operating speed of the drill bit is higher than the upper limit of the detection speed requirement range of the inertial sensor.

4. The online measurement method of virtual angle sensor based on graph neural network according to claim 1 is characterized in that: During the online measurement phase of drilling work, the specific method of using the predicted value of the ambient magnetic field obtained by reverse prediction to compensate and calibrate the ambient magnetic field signal detected by the magnetometer is as follows: The compensation calibration is completed by adjusting the value of the ambient magnetic field signal detected by the magnetometer to be equal to the ambient magnetic field prediction value obtained by reverse prediction.

5. The online measurement method of virtual angle sensor based on graph neural network according to claim 1 is characterized in that: During the online measurement phase of drilling work, the specific process of correcting the drift error of the inertial sensor based on the motion posture of the drill bit predicted and output by the graph neural network model includes: S201, acquiring the motion posture of the inertial sensor under static conditions according to the motion posture signal of the inertial sensor collected, and solving to obtain the initial roll angle and initial pitch angle of the drill bit; S202, determining a horizontal mathematical plane according to an initial roll angle and an initial pitch angle of the drill bit, acquiring a motion attitude signal of an inertial sensor under dynamic conditions, and calculating a motion attitude of the drill bit in a carrier coordinate system; S203, obtaining an attitude conversion matrix according to a conversion relationship between a quaternion attitude differential equation and the motion attitude of the drill bit in a carrier coordinate system; S204. Using the attitude conversion matrix, the motion attitude of the drill bit in the navigation coordinate system predicted and output by the graph neural network model is converted into the motion attitude of the drill bit in the carrier coordinate system, which is used as the error correction value of the motion attitude detection of the inertial sensor to complete the correction of the drift error of the inertial sensor.

6. The online measurement method of virtual angle sensor based on graph neural network according to claim 5 is characterized in that: In step S201, the calculation formula for obtaining the initial roll angle and initial pitch angle of the drill bit is: Where β is the initial pitch angle, γ is the initial roll angle, and a x 、a y 、a z They are the measured values ​​of the accelerometer along the wingspan direction of the aircraft, perpendicular to the ground and in the vertical direction respectively, and g is the acceleration due to gravity.

7. The online measurement method of virtual angle sensor based on graph neural network according to claim 5 is characterized in that: In step S202, the specific process of calculating the motion posture of the drill bit in the carrier coordinate system includes: S2021, controlling the drill motor to rotate to the 0° and 180° positions to stabilize, and collecting angular velocity data output by the inertial sensor under static conditions; S2022. Based on the determined horizontal mathematical plane, the collected output data of the inertial sensor is converted into horizontal plane output data reflecting the posture of the drill bit relative to the horizontal plane, so as to calculate the motion posture of the drill bit in the carrier coordinate system.

8. The online measurement method of virtual angle sensor based on graph neural network according to claim 7 is characterized in that: The solution formula for the motion posture of the drill bit in the carrier coordinate system is: Where α is the azimuth of the drill bit in the carrier coordinate system, ω1 and ω2 are the angular velocities of the static output of the inertial sensor at the two positions of 0° and 180° respectively, and ω ie is the angular velocity of the earth's rotation, and L is the geographical latitude.

9. The online measurement method of virtual angle sensor based on graph neural network according to claim 4 is characterized in that: In step S203, the expression of the posture conversion matrix is: In the formula, is the attitude transformation matrix, θ, They are the inclination angle, azimuth angle and tool face angle of the drill bit in the navigation coordinate system.

10. The online measurement method of virtual angle sensor based on graph neural network according to claim 4 is characterized in that: In step S204, the formula for converting the motion posture of the drill bit in the navigation coordinate system predicted and output by the graph neural network model into the motion posture of the drill bit in the carrier coordinate system is: In the formula, (x b ,y b , z b ) is the carrier coordinate system, (x n ,y n , z n ) is the motion posture of the drill bit in the navigation coordinate system predicted and output by the graph neural network model.

Images