Drilling tool jamming early warning method and system based on multistage model

Abstract

The invention relates to a drill jamming early warning method and system based on a multistage model. Working state parameters in the drilling period are obtained and preprocessed; the torque, the rotating speed, the hook load and the riser pressure in the preprocessed working state parameters serve as input samples, reconstructed samples are obtained through the trained Bi-LSTM auto-encoder model, and if the error between the reconstructed samples and the input samples is larger than a set threshold value, drilling is abnormal; under the condition that well drilling is abnormal, the preprocessed working state parameters, bit pressure, pipe column information, well track and well body structure information serve as input of a friction torque-hydraulic model, predicted values of torque, hook load and riser pressure are obtained, the deviation between the predicted values and corresponding actual values is calculated, and the deviation of the predicted values is calculated; and according to the ratio of the deviation value to a set threshold value, determining a corresponding drill jamming early warning grade. A Bi-LSTM auto-encoder and a friction torque-hydraulic model are utilized to form a two-stage drilling tool jamming early warning method, data driving and mechanism modeling are combined, and the early warning accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas well drilling well control safety, and in particular to a pipe-stuck early warning method and system based on a multi-level model. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Drilling engineering is the main engineering activity for obtaining underground resources and conducting geological exploration. With the increase of well depth, risks such as drill bit sticking are more likely to occur during the drilling process. The occurrence of drill bit sticking accidents will not only prolong the drilling cycle and reduce development efficiency, but may also cause damage to drill tools and more serious drilling accidents such as blowouts and well collapse if handled improperly, and even cause casualties.

[0004] The existing technology uses two types of methods to warn of the risk of possible stuck drill. One type is a stuck drill warning method that models the mechanism of stuck drill. This method can achieve stuck drill warning without a sample training model, but it relies on monitoring the value of a single current time point or a certain indicator, ignoring the use of time series information of monitoring parameters, affecting the accuracy and field applicability of stuck drill warning; the other type is a data-driven stuck drill warning method that can use field measured data to learn the information of the drilling sequence and establish the correlation between various parameters. However, due to the small number of stuck drill samples and the large number of normal samples, the number of the two types of samples is unbalanced, resulting in insufficient model training and low warning accuracy. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a drill stuck warning method and system based on a multi-level model, and utilizes a bidirectional long short-term memory network (Bi-LSTM) autoencoder and a friction torque-hydraulic model to form a two-level drill stuck warning method. The method combines data-driven and mechanism modeling, and adopts a normal sample training method to fully train the Bi-LSTM autoencoder to identify drilling anomalies. After the drilling anomaly is identified, the friction torque-hydraulic model is used to calculate the drill stuck warning index to determine whether the anomaly is a drill stuck risk.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a method for early warning of a stuck pipe based on a multi-level model, comprising the following steps:

[0008] Obtain and pre-process working status parameters during drilling;

[0009] The torque, speed, hook load and riser pressure in the preprocessed working state parameters are used as input samples, and the reconstructed samples are obtained through the trained Bi-LSTM autoencoder model. If the error between the reconstructed sample and the input sample is greater than the set threshold, it is a drilling anomaly.

[0010] In the case of abnormal drilling, the preprocessed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as the input of the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value. The corresponding pipe sticking warning level is determined according to the ratio between the deviation value and the set threshold.

[0011] Furthermore, the Bi-LSTM autoencoder model uses the torque, rotation speed, hook load, riser pressure, drilling pressure, tubing information, wellbore trajectory and wellbore structure information in historical normal well condition samples as training input, and the model output is the reconstructed sample of the training sample. The reconstruction error sample set is obtained by calculating the mean absolute error between the reconstructed sample and the input sample. The reconstruction error threshold is set by taking the percentile of the reconstruction error set, and the threshold is used as the dividing standard between normal samples and abnormal samples.

[0012] Furthermore, during the operation of the friction torque-hydraulic model, the pipe string is divided into multiple calculation units, the midpoint of the calculation unit is taken as the origin, and the main normal direction, the secondary normal direction and the tangent direction are used as the coordinate axis directions to establish a coordinate system.

[0013] Furthermore, the components of gravity on the tangent, principal normal and binormal, and the force equations on the three coordinate axes are determined in the coordinate system.

[0014] Furthermore, according to the force equations on the three coordinate axes, based on the components of the wellbore positive pressure in the main normal and sub-normal directions and the axial forces at the upper and lower ends of the calculation unit, the pressure of the calculation unit on the wellbore is obtained.

[0015] Furthermore, the friction and torque of the calculation unit are obtained by using the pressure of the calculation unit on the well wall, the axial and tangential friction coefficient components, the outer diameter of the pipe string, the micro-element rotation speed and the unit axial speed.

[0016] Furthermore, the riser pressure is obtained according to the full well pressure loss, as shown in the following formula:

[0017] P=P d +P g +P c +P p +P s ;

[0018]

[0019] Where, P is the riser pressure, P d is the drill bit pressure drop, P g is the surface manifold pressure loss, P c is the pressure loss inside and outside the drill collar, P p is the pressure loss inside and outside the drill pipe, P s is the voltage loss of special components, ρ d is the drilling fluid density, Q is the drilling pump displacement, C is the nozzle flow coefficient, d e is the nozzle equivalent diameter, B is the calculation coefficient of the corresponding type of drill pipe, η is the plastic viscosity of the drilling fluid, D p , D c is the outer diameter of the drill pipe and drill collar, D h is the diameter of the corresponding well section, L p , L c is the length of the drill pipe and drill collar, L1, L2, L3, L4 are different well sections in the drilling process, and d1, d2, d3, d4 are the inner diameters of the corresponding well sections.

[0020] A second aspect of the present invention provides a stuck pipe warning system based on a multi-level model, comprising:

[0021] The data acquisition and preprocessing module is configured to: obtain and preprocess working state parameters during drilling;

[0022] The abnormality identification module is configured as follows: taking the pre-processed working state parameters of torque, speed, hook load and riser pressure as input samples, and obtaining reconstructed samples through the trained Bi-LSTM autoencoder model. If the error between the reconstructed sample and the input sample is greater than a set threshold, it is a drilling abnormality;

[0023] The risk warning module is configured as follows: in the case of abnormal drilling, the pre-processed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as inputs to the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value, and determine the corresponding stuck drill warning level according to the ratio between the deviation value and the set threshold.

[0024] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the multi-level model-based drill stuck warning method.

[0025] A fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the multi-level model-based drill stuck warning method when executing the program.

[0026] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:

[0027] A two-level stuck drill warning is formed based on the Bi-LSTM autoencoder model and the friction torque-hydraulic model. The Bi-LSTM autoencoder model, which is more likely to obtain a large number of normal samples, is used to identify drilling anomalies after training, thereby reducing the false alarm rate of stuck drill monitoring. On the basis of identifying the drilling anomaly, the friction torque-hydraulic model is used to further identify whether the drilling anomaly is a stuck drill risk, thereby improving the accuracy of the stuck drill warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 It is a schematic diagram of a two-stage drill stuck warning process based on a Bi-LSTM autoencoder and a friction torque-hydraulic model provided by one or more embodiments of the present invention;

[0030] Figure 2 is a schematic diagram of drill string force analysis provided by one or more embodiments of the present invention;

[0031] Figure 3 is a schematic diagram of a curve of a change in a load on a large hook when a drill is stuck, provided by one or more embodiments of the present invention;

[0032] Figure 4 is a schematic diagram of a torque variation curve when a drill is stuck provided by one or more embodiments of the present invention;

[0033] Figure 5 is a schematic diagram of a rotation speed variation curve when a drill is stuck provided by one or more embodiments of the present invention;

[0034] Figure 6 is a schematic diagram of a riser pressure change curve when a stuck drill occurs provided by one or more embodiments of the present invention;

[0035] Figure 7 is a schematic diagram of an output curve of a Bi-LSTM autoencoder model provided by one or more embodiments of the present invention;

[0036] Figure 8 is a schematic diagram of a friction torque-hydraulic model output curve provided by one or more embodiments of the present invention;

[0037] Fig. 9 Schematic diagram of the BiLSTM autoencoder network architecture provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] Bi-LSTM, Bi-directional Long Short-Term Memory, is a neural network that combines the characteristics of Long Short-Term Memory (LSTM) with bidirectional data processing capabilities. LSTM effectively remembers long-term dependencies by introducing gating mechanisms (input gate, forget gate, and output gate) to control the flow of information. Bi-LSTM further enhances the model's capabilities on this basis. It consists of two LSTM layers: one is a forward LSTM that processes data step by step from the beginning to the end of the sequence; the other is a reverse LSTM that processes from the end to the beginning of the sequence. Through this bidirectional processing, Bi-LSTM is able to simultaneously obtain the contextual information of each element in the sequence.

[0041] The following embodiment provides a drill stuck warning method and system based on a multi-level model, using a bidirectional long short-term memory network autoencoder and a friction torque-hydraulic model to form a two-level drill stuck warning method. This method combines data-driven and mechanism modeling, and uses a normal sample training method to fully train the Bi-LSTM autoencoder to identify drilling anomalies. After the drilling anomaly is identified, the friction torque-hydraulic model is used to calculate the drill stuck warning index to determine whether the anomaly is a drill stuck risk.

[0042] Embodiment 1:

[0043] The method for early warning of a stuck pipe based on a multi-level model includes the following steps:

[0044] Obtain and pre-process working status parameters during drilling;

[0045] The torque, speed, hook load and riser pressure in the preprocessed working state parameters are used as input samples, and the reconstructed samples are obtained through the trained Bi-LSTM autoencoder model. If the error between the reconstructed sample and the input sample is greater than the set threshold, it is a drilling anomaly.

[0046] In the case of abnormal drilling, the preprocessed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as the input of the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value. The corresponding pipe sticking warning level is determined according to the ratio between the deviation value and the set threshold.

[0047] The drill stuck warning method of this embodiment forms a two-level warning based on the Bi-LSTM autoencoder and the friction torque-hydraulic model. The implementation process of the warning method after training is as follows: Figure 1 As shown, the specific implementation process includes the following steps:

[0048] (1) Data acquisition and preprocessing

[0049] The parameters of the drilling equipment during operation are obtained and preprocessed. Data preprocessing includes three parts: outlier point identification and elimination, missing data filling, and dimension difference elimination. Outlier points refer to points where the change gradient of the sampling value cannot be reached within a sampling cycle in the actual system.

[0050] There are points in the monitoring parameter data whose values ​​deviate from the normal trend, so the outlier points are first eliminated using the box plot method. For a drilling sequence X with a length of n, the identification formula for outlier points is as follows:

[0051] IQR=Q3-Q1 (1)

[0052] L min =Q1-1.5*IQR (2)

[0053] L max =Q3+1.5*IQR (3)

[0054] After the values ​​of drilling sequence X are sorted in ascending order, the first quartile is the value at 25% of the sorted drilling sequence, the second quartile is the value at 50% of the sorted drilling sequence, the third quartile is the value at 75% of the sorted drilling sequence, and the fourth quartile is the value at 100% of the sorted drilling sequence. Q1 and Q3 are the first and third quartiles, respectively, IQR is the interquartile range, L min and L max They are the lower data limit and the upper data limit respectively.

[0055] If x i >L max or x i <L min , then the i-th sampling point is considered to be an outlier point, where x i ∈X.

[0056] During the drilling data transmission process, due to equipment problems or electromagnetic interference and other problems, data missing problems may easily occur in the monitoring parameters. The present invention uses the effective value collected at the previous moment to replace the missing value at the current moment.

[0057] In order to eliminate the dimensional differences between the monitoring parameters, the maximum-minimum normalization method is used to normalize the data. The calculation formula is:

[0058]

[0059] Where x is the current value, x′ is the normalized value, min(x) and max(x) represent the minimum and maximum values ​​of the drilling data sequence, respectively.

[0060] (2) Dataset Construction

[0061] The four monitoring parameters of torque, speed, hook load, and riser pressure are used as the input of the Bi-LSTM autoencoder. After completing the data normalization, a sliding window with a length of 30 and a step size of 1 is used to intercept samples from the original data to form a normal training sample set for the Bi-LSTM autoencoder model. At the same time, static information such as drilling pressure, string information, wellbore trajectory, and wellbore structure are collected as the input of the friction torque-hydraulic model.

[0062] (3) Construction of drilling anomaly recognition model based on Bi-LSTM autoencoder

[0063] A drilling anomaly intelligent recognition model is established based on the Bi-LSTM autoencoder. The model consists of three parts: input layer, hidden layer and output layer. The network model structure is shown in Table 1.

[0064] Table 1 Model structure table

[0065]

[0066] (4) Identification of stuck drill risk based on friction torque-hydraulic model

[0067] The friction torque-hydraulic model is used to calculate the predicted values ​​of torque, hook load and riser pressure, and the deviation between the predicted values ​​and the corresponding actual values ​​is calculated. The warning levels of the two are obtained according to the ratio of the deviation value to the threshold. The final stuck drill warning index is obtained by combining the warning levels of torque, hook load and riser pressure.

[0068] In order to predict the load and torque of the hook, the pipe string must first be divided into multiple calculation units. If the axial force at the lower end of the calculation unit is known, the pressure of the calculation unit can be calculated.

[0069] When the drilling tool is moving, the lower axial force of the bottom calculation unit is zero, and the upper axial force of the lower calculation unit is equal to the lower axial force of the adjacent upper calculation unit. Therefore, the axial tension and torque of the wellhead can be calculated starting from the bottom.

[0070] Perform force analysis on the calculation unit, such as Figure 2 As shown, the midpoint of the calculation unit is taken as the origin, and a coordinate system is established with the main normal direction, the binormal direction and the tangent direction as the coordinate axis directions.

[0071]

[0072] Where W is the gravity of the unit, W t , W n , W b are the components of gravity on the tangent, principal normal, and secondary normal, respectively; α2 and α1 are the well inclination angles at the lower and upper ends of the calculation unit, respectively. The change in the inclination azimuth of the upper and lower calculation units.

[0073] Determine the force equations on the three coordinate axes based on the forces on the three coordinate axes:

[0074]

[0075] Among them, N b 、N n are the components of the positive pressure of the wellbore in the main normal and secondary normal directions, T1 and T2 are the axial forces at the upper and lower ends, respectively. After deformation, the pressure of the calculation unit on the wellbore is obtained:

[0076]

[0077] After the normal pressure of the calculation unit on the well wall is obtained by formula (6), the friction and torque of the calculation unit can be solved as shown in the following formula:

[0078]

[0079] Among them, F is the friction resistance of the unit (i.e. the load of the large hook), M is the calculation unit torque, μ a , μ c are the axial and tangential friction coefficient components, D o is the outer diameter of the pipe, n is the micro-speed, V a is the unit axial velocity.

[0080] The riser pressure is calculated based on the full well pressure loss as follows:

[0081] P=P d +P g +P c +P p +P s (10)

[0082] Where, P is the riser pressure, P d is the drill bit pressure drop, P g is the surface manifold pressure loss, P c is the pressure loss inside and outside the drill collar, P p is the pressure loss inside and outside the drill pipe, P s Voltage loss for special components.

[0083]

[0084] Among them, ρ d is the drilling fluid density, Q is the drilling pump displacement, C is the nozzle flow coefficient, d e is the nozzle equivalent diameter, B is the calculation coefficient of the corresponding type of drill pipe, η is the plastic viscosity of the drilling fluid, D p , D c is the outer diameter of the drill pipe and drill collar, D h is the diameter of the corresponding well section, L p , L c is the length of the drill pipe and drill collar, L1, L2, L3, L4 are different well sections in the drilling process, and d1, d2, d3, d4 are the inner diameters of the corresponding well sections.

[0085] (5) Establishment of early warning indicators for stuck drill

[0086] The predicted values ​​of torque and hook load are calculated based on the friction torque-hydraulic model, and the deviation between the actual value and the predicted value is calculated:

[0087]

[0088] Among them, x actual is the actual value of the parameter, x predicted is the predicted value of the parameter.

[0089] The threshold settings in Table 2 are obtained under different working conditions. Taking the hook load as an example, when the hook load deviation is less than the threshold, it means that the hook load parameter is less likely to change abnormally due to the risk of drill sticking; when the hook load deviation is much greater than the threshold, it means that the hook load is more likely to change abnormally due to the risk of drill sticking. The warning level of the hook load is obtained according to the ratio of the hook load deviation to the threshold. The corresponding relationship between the ratio of the parameter deviation to the threshold and the parameter warning level is shown in Table 3.

[0090] Table 2 Threshold setting table

[0091] Working conditions Torque deviation threshold Hook load deviation threshold Riser pressure deviation threshold Rotary drilling 10％ -4％ 12％ Sliding drilling - -5％ 14％ Eye 6％ -20％ 6％ Reverse eye 4％ 10％ 20％ Drill Down - -16％ - Drilling - 12％ -

[0092] Table 3 Hook load and torque warning level setting table

[0093] Deviation / Threshold <1 1-3 >3 <![CDATA[Level hl ]]> 0 Actual value of hook load deviation / threshold 3 <![CDATA[Level tq ]]> 0 Actual value of torque deviation / threshold 3 <![CDATA[Level spp ]]> 0 Actual value of the riser pressure deviation / threshold 3

[0094] In this embodiment, the warning level is divided into three levels. When the ratio between the parameter deviation and the threshold is less than 1, it is the first level; when the ratio between the parameter deviation and the threshold is within 1-3, it is the second level; when the ratio between the parameter deviation and the threshold exceeds 3, it is the third level.

[0095] Finally, according to the big hook load warning level Level hl, Torque warning level Level tq , Standpipe pressure warning level Level spp and the maximum warning level Max , get the drill stuck warning indicator Level, the calculation formula is:

[0096]

[0097] (6) Training a drilling anomaly recognition model based on Bi-LSTM autoencoder

[0098] The BiLSTM autoencoder learns the characteristics and distribution patterns of the input samples by encoding the input samples into low-dimensional representations and then decoding the low-dimensional representations into reconstructed samples of the input samples.

[0099] like Fig. 9 As shown in the figure, the output layer of the autoencoder has the same structure as the input layer. The input samples are compressed into low-dimensional representations during the encoding process, and the low-dimensional representations are reconstructed into original samples during the decoding process. In this way, the input samples are reconstructed and trained to learn the data distribution law of the input samples and realize feature extraction of the input samples. The weights and biases of the BiLSTM autoencoder are continuously updated during the training process so that the reconstructed samples finally output by the model are the same as the input samples. Therefore, the historical normal well condition samples are input as training samples into the Bi-LSTM autoencoder model, and the model outputs the reconstructed samples of the normal well condition samples. The reconstructed error sample set is obtained by calculating the mean absolute error between the reconstructed samples and the input samples. Then, the percentile of the reconstructed error set is taken to set the reconstruction error threshold, and the threshold is used as the dividing standard between normal samples and abnormal samples. The calculation formula of the mean absolute error is as follows:

[0100]

[0101] Among them, W R and W are the reconstructed sample and input sample respectively, w i R and w i are the reconstructed parameters and original parameter values ​​at the i-th time point, k is the number of time points in the sequence, and m is the number of parameters in the sequence.

[0102] A two-stage stuck drill warning method based on Bi-LSTM autoencoder and friction torque-hydraulic model makes full use of a large number of easily available normal samples to train Bi-LSTM autoencoder to identify drilling anomalies and reduce the false alarm rate of stuck drill monitoring. On the basis of identifying drilling anomalies, the friction torque-hydraulic model is used to further identify whether the drilling anomaly is a stuck drill risk, thereby improving the accuracy of stuck drill warning.

[0103] After the model training is completed, a warning of the risk of stuck drill is executed, such as Figure 2As shown in the figure, the Bi-LSTM autoencoder model is used to monitor the data online. The trained Bi-LSTM autoencoder has learned the data distribution law of normal well condition samples. If the model input is a sample of normal well condition, the reconstruction error between the output reconstructed sample and the input sample is small. Therefore, if the reconstruction error output by the model is greater than the set threshold, it indicates that the input sample does not conform to the data distribution law of normal well condition samples, and it is judged as drilling abnormality. At this time, the friction torque-hydraulic model is loaded to output the drill bit early warning indicator; if the reconstruction error output by the model is less than the set threshold, it is judged as normal well condition.

[0104] In this embodiment, the network structure parameters of the Bi-LSTM autoencoder are shown in Table 1. Under this structure, the input parameters are torque, hook load, rotation speed, and standpipe pressure, and the time dimension is 30. Then the shape of the input data is (None, 30, 4). When the input data is input into the model, the output shape of the first Bi-LSTM layer is (None, 30, 32), and the output shape of the first Dropout layer is (None, 30, 32); the output shape after the second Bi-LSTM layer is (None, 16), and the output shape of the second Dropout layer is (None, 16); the output shape after the RepeatVector layer is (None, 30, 16), and then the output shape after the third Bi-LSTM layer and Dropout layer is (None, 30, 16), and the output shape after the fourth Bi-LSTM layer is (None, 30, 32). Finally, the output of the TimeDistributed layer is (None, 30, 4), which is the reconstructed sample.

[0105] Taking the stuck pipe risk of a drilling well as an example for analysis, the Bi-LSTM autoencoder model is used to identify drilling anomalies, and then the friction torque model is used to warn the stuck pipe risk under drilling anomalies. The parameter changes before and after the stuck pipe occur are shown as follows: Figure 3-Figure 6 As shown in the figure, the curve diagrams of the model output are as follows: Figure 7 and Figure 8 As shown, this case shows that the method of this embodiment can identify drilling anomalies in a timely manner and reduce the false alarm of the risk of stuck pipe.

[0106] Embodiment 2:

[0107] A multi-level model-based early warning system for stuck pipe, including:

[0108] The data acquisition and preprocessing module is configured to: obtain and preprocess working state parameters during drilling;

[0109] The abnormality identification module is configured as follows: taking the pre-processed working state parameters of torque, speed, hook load and riser pressure as input samples, and obtaining reconstructed samples through the trained Bi-LSTM autoencoder model. If the error between the reconstructed sample and the input sample is greater than a set threshold, it is a drilling abnormality;

[0110] The risk warning module is configured as follows: in the case of abnormal drilling, the pre-processed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as inputs to the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value, and determine the corresponding stuck drill warning level according to the ratio between the deviation value and the set threshold.

[0111] Embodiment three:

[0112] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the multi-level model-based drill stuck early warning method as described in the first embodiment are implemented.

[0113] Embodiment 4:

[0114] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the multi-level model-based drill stuck warning method described in the first embodiment are implemented.

[0115] The steps involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0116] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for early warning of a stuck drill based on a multi-level model, characterized in that: The following steps are involved: Obtain and pre-process working status parameters during drilling; The torque, speed, hook load and riser pressure in the preprocessed working state parameters are used as input samples, and the reconstructed samples are obtained through the trained Bi-LSTM autoencoder model. If the error between the reconstructed sample and the input sample is greater than the set threshold, it is a drilling anomaly. In the case of abnormal drilling, the preprocessed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as the input of the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value. The corresponding pipe sticking warning level is determined according to the ratio between the deviation value and the set threshold.

2. The method for early warning of a stuck pipe based on a multi-level model according to claim 1, characterized in that: The Bi-LSTM autoencoder model uses the torque, speed, hook load, riser pressure, drilling pressure, tubing information, wellbore trajectory and wellbore structure information in historical normal well condition samples as training input. The model output is the reconstructed sample of the training sample. The reconstruction error sample set is obtained by calculating the mean absolute error between the reconstructed sample and the input sample. The reconstruction error threshold is set by taking the percentile of the reconstruction error set, and the threshold is used as the criterion for dividing normal samples and abnormal samples.

3. The method for early warning of a stuck pipe based on a multi-level model according to claim 1, characterized in that: During the operation of the friction torque-hydraulic model, the pipe string is divided into multiple calculation units, the midpoint of the calculation unit is taken as the origin, and the main normal direction, the secondary normal direction and the tangent direction are used as the coordinate axis directions to establish a coordinate system.

4. The method for early warning of a stuck pipe based on a multi-level model as claimed in claim 3, characterized in that: In the moving coordinate system, determine the components of gravity on the tangent, principal normal and binormal, and the force equations on the three coordinate axes.

5. The method for early warning of a stuck pipe based on a multi-level model as claimed in claim 4, characterized in that: According to the force equations on the three coordinate axes, based on the components of the positive pressure of the wellbore in the main normal and sub-normal directions and the axial forces at the upper and lower ends of the calculation unit, the pressure of the calculation unit on the wellbore wall is obtained.

6. The method for early warning of a stuck pipe based on a multi-level model as claimed in claim 5, characterized in that: The friction and torque of the calculation unit are obtained by using the pressure of the calculation unit on the well wall, the axial and tangential friction coefficient components, the outer diameter of the pipe string, the micro-element rotation speed and the unit axial speed.

7. The method for early warning of a stuck pipe based on a multi-level model according to claim 1, characterized in that: The riser pressure is obtained according to the full well pressure loss, as shown in the following formula: P=P d +P g +P c +P p +P s ; Where, P is the riser pressure, P d is the drill bit pressure drop, P g is the surface manifold pressure loss, P c is the pressure loss inside and outside the drill collar, P p is the pressure loss inside and outside the drill pipe, P s is the voltage loss of special components, ρ d is the drilling fluid density, Q is the drilling pump displacement, C is the nozzle flow coefficient, d e is the nozzle equivalent diameter, B is the calculation coefficient of the corresponding type of drill pipe, η is the plastic viscosity of the drilling fluid, D p , D c is the outer diameter of the drill pipe and drill collar, D h is the diameter of the corresponding well section, L p , L c is the length of the drill pipe and drill collar, L1, L2, L3, L4 are different well sections in the drilling process, and d1, d2, d3, d4 are the inner diameters of the corresponding well sections.

8. A multi-level model-based early warning system for stuck pipe, characterized in that: include: The data acquisition and preprocessing module is configured to: obtain and preprocess working state parameters during drilling; The abnormality identification module is configured as follows: taking the pre-processed working state parameters of torque, rotation speed, hook load and standpipe pressure as input samples, obtaining reconstructed samples through the trained Bi-LST M autoencoder model, and if the error between the reconstructed sample and the input sample is greater than a set threshold, it is a drilling abnormality; The risk warning module is configured as follows: in the case of abnormal drilling, the pre-processed working state parameters, as well as drilling pressure, tubing information, wellbore trajectory and wellbore structure information are used as inputs to the friction torque-hydraulic model to obtain the predicted values ​​of torque, hook load and riser pressure, and calculate the deviation between the predicted value and the corresponding actual value, and determine the corresponding stuck drill warning level according to the ratio between the deviation value and the set threshold.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps in the multi-level model-based drill stuck early warning method as described in any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the multi-level model-based drill stuck early warning method as described in any one of claims 1 to 7 are implemented.

Images