Method and device for evaluating stick-slip vibration of bottom drilling tool assembly and drill bit

Abstract

The invention discloses a method and a device for evaluating stick-slip vibration strength of a bottom drilling tool assembly and a drill bit, and belongs to the technical field of stick-slip vibration strength evaluation. The method comprises the following steps: acquiring radial or tangential acceleration data during stick-slip vibration; performing data processing on the obtained radial or tangential acceleration data of the stick-slip vibration during the stick-slip vibration to obtain a characteristic value of the radial or tangential acceleration of the stick-slip vibration; predicting a stick-slip index by taking the obtained radial or tangential acceleration characteristic value of the stick-slip vibration as an input value of a pre-trained stick-slip index prediction neural network model; the invention further provides an evaluation device based on the evaluation method, rotation speed data does not need to be detected, the radial or tangential acceleration data during stick-slip vibration can be actually measured only through a single acceleration sensor, and the single acceleration sensor is easy and convenient to install and arrange and high in reliability. The problem that it is difficult to install a plurality of sensors on a shaft bottom drilling tool can be avoided, and the evaluation result is accurate.

Description

Technical Field

[0001] The invention relates to the technical field of stick-slip vibration intensity assessment, and in particular to a method and a device for assessing the stick-slip vibration intensity of a bottom drill assembly and a drill bit. Background Art

[0002] When drilling horizontal wells and deep wells, due to the long length of the downhole drill bit, the low torsional stiffness of the drill bit, and the inconsistency between the rotation speed of the bottom hole drill bit and the surface rotation speed, the torque of the drill bit may be difficult to overcome the friction between the well wall, and stick-slip vibration is likely to occur. In addition, if the drill bit penetrates too deep into the rock during drilling, the torque at the drill bit cannot meet the rock breaking requirements, causing the drill bit to get stuck, which is also likely to induce stick-slip vibration.

[0003] The stick-slip vibration of the drill bit includes two stages: sticking and slipping. In the sticking stage, the torque of the drill bit is not enough to overcome the friction between the drill bit and the formation, and the bottom hole drill bit speed drops to zero, but the upper drill bit continues to rotate under the drive of the ground, causing the drill bit to twist and accumulate energy. In the slipping stage, the friction between the drill bit and the formation is suddenly released, and the bottom hole drill bit speed can reach 2 to 15 times the ground speed.

[0004] When the drill bit experiences stick-slip vibration, the high-speed rotating drill bit will strongly collide with the bottom rock or well wall, which will accelerate the wear of the drill bit. At the same time, the severe periodic alternating stress will also cause premature fatigue failure of the downhole drill bit. In addition, the torque of the downhole drill bit usually fluctuates greatly during the stick-slip vibration process, which will not only greatly reduce the drilling efficiency, but also increase the instability and uncontrollability of the drilling system, increase the drilling risk, and accurately evaluate the stick-slip vibration intensity of the drill bit is the first step to alleviate and eliminate stick-slip vibration. The strength of stick-slip vibration is generally evaluated by the stick-slip index, and its value is equal to (maximum speed - minimum speed) / average speed. The larger the calculated stick-slip index, the stronger the stick-slip vibration, and the greater the damage to the drill bit.

[0005] At present, there are two main ways to monitor stick-slip vibration. One is to identify it by monitoring the rotation speed, torque and other parameters at the wellhead. However, this method has a certain lag, and due to the friction between the drill bit and the wellbore, the stick-slip vibration at the bottom of the well will be distorted after being transmitted to the ground, and the accuracy of monitoring deep formations is low. Another method is to set a speed sensor on the bottom of the wellbore to identify the stick-slip vibration through the rotation speed data of the drill bit and the bottom of the wellbore. At present, there are three main methods for measuring the rotation speed of downhole drill tools. One is to use two acceleration sensors for measurement. This method cannot measure the rotation speed close to zero; the second is to use two flux gates for measurement. This method cannot be used in the casing or when it is close to the casing of the adjacent well and is easily interfered; the third method is to install a gyroscope on the drill tool to directly measure the rotation speed. This method cannot measure higher rotation speeds, and the frequency of the data obtained is not high. All three methods have their own defects. To accurately measure the rotation speed, multiple methods need to be used in combination, which requires the installation and use of multiple measurement sensors at the same time, which is costly. In addition, the space available for installing sensors in the bottom of the wellbore is limited, which makes it difficult to install multiple measurement sensors on the bottom of the wellbore. Summary of the invention

[0006] The purpose of the present invention is to overcome the problems in the prior art and provide a method and device for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit of the present invention only requires acceleration measurement data at the bottom of the well to accurately predict the stick-slip index, and evaluates the stick-slip vibration strength of the bottom hole assembly and the drill bit through the stick-slip index.

[0007] The present invention provides a method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit, comprising the following steps:

[0008] Obtain radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit when stick-slip vibration occurs;

[0009] Processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or the drill bit when the stick-slip vibration occurs to obtain characteristic values ​​of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when the stick-slip vibration occurs;

[0010] The obtained characteristic values ​​of radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when stick-slip vibration occurs are used as input values ​​of a pre-trained stick-slip index prediction neural network model to predict the stick-slip index, and the stick-slip vibration intensity of the bottom hole assembly or the drill bit is evaluated based on the predicted stick-slip index.

[0011] As a preferred embodiment, the training method of the stick-slip index prediction neural network model includes:

[0012] Use the rotor dynamics model to simulate the radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit when stick-slip vibration occurs, and provide raw data for the neural network model;

[0013] Processing the original data to obtain the characteristic value of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit during the stick-slip vibration; using the rotational speed data during the simulation to calculate the stick-slip index of the bottom hole assembly or the drill bit during the stick-slip vibration; using the stick-slip index calculated by the rotational speed data as the output value of the neural network model, and using the characteristic value of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit during the stick-slip vibration as the input value of the neural network model;

[0014] A genetic algorithm is used to optimize the neural network model to obtain a stick-slip index prediction neural network model, and the stick-slip index prediction neural network model is trained to obtain a trained stick-slip index prediction neural network model.

[0015] As a preferred method, the method for processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit is as follows: firstly, removing obvious outliers in the radial acceleration or tangential acceleration data, and then using the moving average method to smooth the radial acceleration or tangential acceleration data; finally, normalizing the radial acceleration or tangential acceleration to limit the radial acceleration or tangential acceleration data to the interval of [0,1].

[0016] As a preferred embodiment, the characteristic value of the radial acceleration or tangential acceleration is 8 characteristic values, and the 8 characteristic values ​​are respectively kurtosis, shape factor, skewness, standard deviation, quartile, variance and two custom features; wherein the first custom feature is defined as the proportion of sample points falling outside the range of 1-n to 1+n times the average value relative to the total number of sample points, n<0.5; the second custom feature is defined as the proportion of sampling points whose difference between two consecutive accelerations exceeds m, m<0.1.

[0017] As a preferred method, before using a genetic algorithm to optimize the stick-slip index prediction neural network model, a neural network structure is designed, wherein the neural network structure includes 8 input nodes, 1 hidden layer, the number of nodes is X, and 1 output neuron is set, that is, the network structure is 8-X-1; the 8 input nodes are 8 eigenvalues ​​of radial acceleration or tangential acceleration, the 1 output neuron is the predicted stick-slip index, and the number of hidden layer nodes X is 10-20;

[0018] The number of hidden layer nodes is changed, and the genetic algorithm optimization is performed every time the number of hidden layer nodes is changed.

[0019] As a preferred method, the stick-slip index prediction neural network model is optimized by genetic algorithm, including:

[0020] Initialize the parameters to obtain the initial population;

[0021] For each individual, a neural network is trained according to its weight and bias, and the characteristic values ​​of radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when the stick-slip vibration occurs are input to obtain a set of predicted stick-slip indices; the root mean square error between the predicted stick-slip indices and the actually calculated stick-slip indices is used as the cost function of the genetic algorithm;

[0022] Sort the cost functions in ascending order, select the P*CR group with the smallest cost function as the parent generation, and randomly cross-combine to form a new P group of individuals;

[0023] The P group of individuals generated by the parent crossover are allowed to mutate with a mutation rate of MR. The mutated P group of individuals serve as the new generation population and enter the next round of optimization;

[0024] After each round of calculation of this function, the minimum cost function value obtained this time is compared with the minimum value obtained in the previous round; if the difference between the two minimum values ​​is less than 0.5%, the weight and deviation corresponding to the individual with the smallest cost function value at this time are the optimization results; conversely, if the difference between the two minimum values ​​is greater than 0.5%, it is necessary to enter the next round to continue optimization; set the maximum number of iterations, and when the number of population iterations reaches the maximum number of iterations, the weight and deviation combination corresponding to the individual with the lowest cost function value at this time is used as the final optimization result.

[0025] As a preferred method, the training of the stick-slip index prediction neural network model uses the mean square error (MSE) of the model as the standard for judging whether to stop the model training; a threshold value Emin is set, and when the difference between the MSE of this training and the MSE of the previous training is greater than Emin, the model training continues; when the difference between the MSE of this training and the MSE of the previous training is less than Emin, the training is stopped, and a maximum number of training times is set at the same time. When the number of training times reaches the maximum value, the model training is stopped, and finally the accuracy of the model is tested.

[0026] The present invention also provides a device based on the above-mentioned bottom hole assembly and drill bit stick-slip vibration intensity evaluation method, comprising:

[0027] An identification module for identifying whether a bottom hole assembly or a drill bit has stick-slip vibration;

[0028] An acquisition module, used for acquiring radial acceleration or tangential acceleration data of stick-slip vibration when stick-slip vibration occurs in the bottom hole assembly or the drill bit;

[0029] A data processing module is used to process the radial acceleration or tangential acceleration data of the stick-slip vibration to obtain the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration;

[0030] The prediction module is used to use the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration as the input value of the pre-trained stick-slip index prediction neural network model to predict the stick-slip index.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention obtains radial acceleration or tangential acceleration data of a bottom hole assembly or a drill bit during stick-slip vibration, processes and obtains characteristic values ​​of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit during stick-slip vibration, and inputs the characteristic values ​​into a pre-trained stick-slip index prediction neural network model to obtain a predicted stick-slip index. The present invention does not need to detect rotational speed data, and can measure the radial acceleration or tangential acceleration data of the bottom hole assembly or the drill bit during stick-slip vibration only through a single acceleration sensor. The installation and arrangement of a single acceleration sensor is simple, and the problem of difficulty in installing multiple sensors on a bottom hole drill tool due to limited space available for installing sensors in the bottom hole drill tool can be avoided.

[0033] The present invention uses a genetic algorithm to optimize the neural network model, thereby ensuring the accuracy of the prediction results of the stick-slip index prediction neural network model and rapid response. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is an overall implementation flow chart of an embodiment of the present invention.

[0035] Figure 2 : is the rotor dynamics model of an embodiment of the present invention.

[0036] Figure 3 This is the placement position of the acceleration sensor according to the embodiment of the present invention.

[0037] Figure 4 It is the sinusoidally changing rotation speed of the embodiment of the present invention.

[0038] Figure 5 This is a flowchart of neural network model optimization according to an embodiment of the present invention.

[0039] Figure 6 1 is the stick-slip vibration acceleration and the rotation speed of the embodiment of the present invention.

[0040] Figure 7 The stick-slip vibration radial acceleration and rotation speed of the embodiment of the present invention.

[0041] Figure 8 This is a prediction diagram of the neural network model for predicting the stick-slip index according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0043] It should be noted that the terms first, second, etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms include and have and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] The prior art generally uses the stick-slip index as an indicator to evaluate the severity of stick-slip vibration of the bottom drill assembly and drill bit in the well. The stick-slip index needs to be calculated using the rotation speed of the bottom drill assembly or the top of the drill bit. The stick-slip index is defined as (maximum rotation speed of the drill bit - minimum rotation speed of the drill bit) / average rotation speed of the drill bit. In many cases, there is no rotation speed data of the downhole drill bit or bottom drill. The problem actually solved by the present invention is how to evaluate the intensity of stick-slip vibration if there is only acceleration measurement data of the bottom drill assembly or drill bit, but no rotation speed measurement of the bottom drill assembly or drill bit. The method for evaluating the stick-slip vibration intensity of the bottom drill assembly and drill bit provided by the present invention can replace the rotation speed measurement of the bottom drill assembly or drill bit, thereby reducing the number of rotation speed sensors, thereby reducing the cost of downhole measurement and improving the accuracy of the evaluation results.

[0045] This invention is based on the scientific research and innovation results of the National Natural Science Foundation Youth Fund "Research on Intelligent Identification Methods of BHA Vibration Modes and Excitation Sources under Irregular Well Walls", Grant No.: 52304002.

[0046] The present invention provides a method for evaluating the stick-slip vibration intensity of a bottom drill assembly and a drill bit, comprising the following steps: obtaining radial acceleration or tangential acceleration data of the bottom drill assembly or the drill bit during stick-slip vibration; as an example, when it is necessary to evaluate the stick-slip vibration intensity of the bottom drill assembly, obtaining the radial acceleration or tangential acceleration data of the bottom drill assembly; when it is necessary to evaluate the stick-slip vibration intensity of the drill bit, obtaining the radial acceleration or tangential acceleration data of the drill bit.

[0047] The radial acceleration or tangential acceleration data of the present invention is obtained by measuring with an acceleration sensor. Specifically, when measuring the radial acceleration or tangential acceleration of the bottom drill assembly, the acceleration sensor is installed on the bottom drill assembly; when measuring the radial acceleration or tangential acceleration of the drill bit, the acceleration sensor is installed on the drill bit.

[0048] Processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or the drill bit when the stick-slip vibration occurs to obtain characteristic values ​​of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when the stick-slip vibration occurs;

[0049] The obtained characteristic values ​​of radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when stick-slip vibration occurs are used as input values ​​of a pre-trained stick-slip index prediction neural network model to predict the stick-slip index, and the stick-slip vibration intensity of the bottom hole assembly or the drill bit is evaluated based on the predicted stick-slip index.

[0050] The present invention is based on the radial or tangential high-frequency acceleration data measured at the bottom of the well, and uses a neural network to predict the stick-slip index. The method involves the acquisition of training data for the neural network prediction model, the design and training of the neural network structure, and the application of the trained model. The content of the present invention involves four modules. The first module is to simulate and generate stick-slip vibration acceleration data to provide raw data for training the neural network model; the second module is to process the acceleration data and calculate the stick-slip index and eigenvalue. The stick-slip index is the output result of the neural network model, and the eigenvalue is the input value of the neural network model; the third module is to optimize and train the neural network model for stick-slip index prediction; the fourth module is to identify, extract and process the actual stick-slip vibration data, and input the model to predict the stick-slip index.

[0051] Exemplarily, the training method of the stick-slip index prediction neural network model includes: using a rotor dynamics model to simulate the radial acceleration or tangential acceleration data of the bottom drill assembly or drill bit when stick-slip vibration is generated, and providing raw data for the neural network model; processing the raw data to obtain the characteristic value of the radial acceleration or tangential acceleration of the bottom drill assembly or drill bit during stick-slip vibration; using the rotational speed data during simulation to calculate the stick-slip index of the bottom drill assembly or drill bit during stick-slip vibration; using the stick-slip index calculated by the rotational speed data as the output value of the neural network model, and using the characteristic value of the radial acceleration or tangential acceleration of the bottom drill assembly or drill bit during stick-slip vibration as the input value of the neural network model; using a genetic algorithm to optimize the neural network model to obtain a stick-slip index prediction neural network model, and training the stick-slip index prediction neural network model to obtain a trained stick-slip index prediction neural network model. In this embodiment, the stick-slip vibration acceleration data is simulated and generated by the rotor dynamics model. The stick-slip vibration acceleration data is generated by the rotor dynamics model, which is not limited by the site conditions and generates sufficient acceleration data. It should be understood that in addition to the rotor dynamics model, other models can also be used to obtain, such as a finite element model. This embodiment takes the stick-slip vibration acceleration data generated by the rotor dynamics model simulation as an example for explanation:

[0052] like Figure 2 As shown in the figure, the rotor dynamics model is used to simulate the vibration of BHA (bottom hole assembly) in the wellbore. In the figure, φ is the wellbore inclination, O is the geometric center of the wellbore, S is the center of the disk, G is the center of mass of the disk, e is the distance between the center of mass of the disk and the center of the disk, k1 is the bending stiffness of the bottom hole assembly, k2 is the contact stiffness between the drill bit and the wellbore wall, FN and FT are the normal contact force and tangential friction force from the wellbore wall, Ω is the revolution speed of BHA rotating around the center line of the wellbore, and ω is the rotation speed of BHA rotating around its own center.

[0053] The equation of motion for the BHA center point S is:

[0054]

[0055] Where: m is the mass concentrated on the disk; r is the radial displacement; are the first and second derivatives of radial displacement with respect to time, respectively; g is the acceleration due to gravity; θ is the angle around the axis of the wellbore; are the first and second derivatives of the rotation angle θ with respect to time; C is the damping coefficient; t represents time.

[0056] The outer circle in the figure is the wellbore, and the inner circle is the BHA. The BHA rotates from a to a' under the combined action of revolution and rotation. The acceleration sensor is located at the eccentric point P of the BHA, and can measure its acceleration at e x 、e yThe acceleration in two directions is radial acceleration and tangential acceleration. The radial acceleration and tangential acceleration are calculated by the following formula:

[0057]

[0058] a rad =a Pi cosβ+a Pj sinβ;

[0059] a tan =-a Pi sinβ+a Pj cosβ.

[0060] Where: a Pi is the acceleration of point P in the i direction; a Pj is the acceleration of point P in the j direction; a rad is the radial acceleration of point P; a tan is the tangential acceleration of point P; D is the distance from the acceleration sensor to the center of the disk.

[0061] By changing the BHA specifications, rotation speed and other parameters, you can simulate different vibrations of the drill tool in the wellbore. Use the following formula to set the rotation speed to simulate the actual stick-slip vibration speed, such as Figure 4 As shown, the acceleration during stick-slip vibration can be simulated.

[0062]

[0063] Solving the equation of motion yields the radial displacement r and rotation angle θ of the BHA center point, as well as their first-order With the second-order derivative The radial acceleration and tangential radial acceleration of the BHA can be calculated.

[0064] The acceleration sensor can be arranged at an eccentric point P of the BHA, such as Figure 3 shown.

[0065] Where: s is the ground speed, which is a constant; T1 is the slip time, during which the BHA rotation speed changes sinusoidally; T2 is the sticky time, during which the BHA rotation speed is 0; T=T1+T2 is a whole stick-slip cycle.

[0066] It should be noted that the radial acceleration and tangential acceleration in this embodiment are two different single-point uniaxial accelerations, and the specific acceleration training model used should be consistent with the actually measured single-point uniaxial acceleration.

[0067] The method for processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit is as follows: first remove the obvious outliers in the radial acceleration or tangential acceleration data, then use the moving average method to smooth the radial acceleration or tangential acceleration data; finally, normalize the radial acceleration or tangential acceleration to limit the radial acceleration or tangential acceleration data to the interval of [0,1]. Specifically, the stick-slip index of the stick-slip vibration is calculated using the rotation speed data during simulation:

[0068] Stick-slip index = (maximum speed - minimum speed) / average speed

[0069] The stick-slip index can be used as an indicator to evaluate the strength of stick-slip vibration. The larger the calculated stick-slip index is, the more severe the stick-slip vibration is, and vice versa.

[0070] Since there are a lot of noise points in the radial acceleration or tangential acceleration data generated by the simulation, it will interfere with the extraction of radial acceleration or tangential acceleration features. Before extracting the radial acceleration or tangential acceleration features, the radial acceleration or tangential acceleration data needs to be properly processed to reduce the influence of these noise points. The specific processing method is to use the box plot method to remove obvious outliers in the radial acceleration or tangential acceleration data, and then use the moving average method to smooth the radial acceleration or tangential acceleration data. In order to eliminate the influence of the eigenvalue dimension and the differences caused by different BHA specifications and drilling parameters, the maximum and minimum value method is used to normalize the radial acceleration or tangential acceleration data, and the radial acceleration or tangential acceleration data is limited to the interval of [0,1]. The eight eigenvalues ​​of the radial acceleration or tangential acceleration of stick-slip vibration are extracted: kurtosis, shape factor, skewness, standard deviation, quartile, variance and two custom features. The first custom feature is defined as the proportion of sample points that fall outside the range of 1-n to 1+n times the average value relative to the total number of sample points, n < 0.5. The second custom feature is defined as the proportion of sampling points where the difference between two consecutive accelerations exceeds m, m < 0.1.

[0071] The specific method of optimizing and training the stick-slip index prediction neural network model in this embodiment is as follows:

[0072] Exemplarily, a neural network structure is designed, which includes 8 input nodes, 1 hidden layer, the number of nodes is X, and 1 output neuron is set, that is, the network structure is 8-X-1. The 8 input nodes are 8 eigenvalues ​​of single-point uniaxial acceleration (radial or tangential), and 1 output neuron is the predicted stick-slip index. The number of hidden layer nodes X can be 10-20.

[0073] As an example, the data of the sample set is divided into 3 parts, 70% of the data samples are used as training sets, 15% of the samples are used as validation sets, and the last 15% of the samples are used as test sets. The training set is used to train the classification model, the validation set is used to test the accuracy of the model after each training and help optimize the model, and the test set is used to test the accuracy of the model after training.

[0074] A genetic algorithm (GA algorithm) is used to optimize the weights and biases of the neural network. For a neural network with a structure of 8-X-1, (8*X+X) weights and (X+1) biases need to be optimized, that is, each individual in the GA algorithm contains (10X+1) parameters.

[0075] Initialize the parameters before using the genetic algorithm for optimization. Set the population size P of the GA algorithm, which can be 50-300, the maximum number of iterations NG, which can be 100-300, the crossover percentage CR, which can be 0.1-0.8, and the mutation percentage MR, which can be 0.01-0.1. Provide an initial combination of weights and biases, and add random values ​​on this basis to obtain P groups of random weight and bias combinations as the initial population.

[0076] For each individual, the neural network is trained according to its weight and bias, and the eight eigenvalues ​​of single-point uniaxial acceleration, i.e. radial acceleration or tangential acceleration, are input to obtain a set of predicted stick-slip indices. The root mean square error RMSE between the predicted stick-slip index and the actually calculated stick-slip index is calculated as the cost function of the GA algorithm. The lower its value, the more accurate the prediction result.

[0077] Sort the cost functions in ascending order, select the P*CR group individuals with the smallest cost function as the parent generation, and randomly cross-combine them to form a new P group of individuals.

[0078] The P group of individuals generated by the parent crossover are allowed to mutate, with a mutation rate of MR, that is, the ratio of the parameters after mutation to the parameters before mutation is between 1-MR and 1+MR. The P group of individuals after mutation is called the new generation of population and enters the next round of optimization.

[0079] After each round of calculation of this function, the minimum cost function value obtained this time is compared with the minimum value obtained in the previous round. If the difference between the two minimum values ​​is less than 0.5%, it means that the GA algorithm has converged, and the weight and deviation corresponding to the individual with the smallest cost function value are the optimization result. On the contrary, if the difference between the two minimum values ​​is greater than 0.5%, it means that the algorithm has not converged yet, and the optimal weight and deviation combination has not been found, and it is necessary to enter the next round to continue optimization. Set the maximum number of iterations NG, that is, when the number of population iterations reaches the maximum number of iterations NG, regardless of whether the algorithm has converged at this time, the iteration of the GA algorithm will end, and the weight and deviation combination corresponding to the individual with the lowest cost function value at this time will be used as the final optimization result.

[0080] The number of nodes X in the hidden layer is optimized. The number of nodes X takes values ​​in sequence within the value range, and the GA algorithm is optimized once for each value, and different optimization results are obtained. The number of nodes corresponding to the result with the lowest value of the cost function is found as the optimization result.

[0081] The neural network model for predicting the stick-slip index constructed by training is used. The mean square error (MSE) of the model is used as the criterion for judging whether the model training should be stopped. A threshold value Emin is set. When the difference between the MSE of this training and the MSE of the last training is greater than Emin, it means that the model has not converged and needs to continue to be trained. When the difference between the MSE of this training and the MSE of the last training is less than Emin, it means that the model has reached the convergence condition. At this time, the training is stopped and the final accuracy test of the model is performed using the test set. At the same time, a maximum number of training times is set. If the model still cannot meet the convergence condition when the number of training times reaches the maximum value, the training of the model is stopped and the accuracy of the model is tested using the test set. After training, a neural network model for predicting the stick-slip index is obtained.

[0082] The method for predicting the downhole stick-slip index using the trained stick-slip index prediction neural network model is as follows:

[0083] Before using the trained stick-slip index prediction neural network model to predict the stick-slip index, it is first necessary to identify whether stick-slip vibration occurs in the well. For specific identification of whether stick-slip vibration occurs in the well, please refer to the method mentioned in the paper "Wang X, Wang X, Lu J, et al. Automated classification of driver string vibrations using machine learning algorithms [J]. Geoenergy Science and Engineering, 239 (2024), 212995." This method relies on the measured single-point single-axis (radial or tangential) acceleration data to identify whether stick-slip vibration occurs in the well, and the identification algorithm used is a machine learning method.

[0084] After the occurrence of stick-slip vibration is identified, the radial acceleration or tangential acceleration and speed during the specific stick-slip vibration are selected as the data to be analyzed, and a complete set of stick-slip processes is regarded as a stick-slip vibration, such as Figure 6 shown.

[0085] Figure 6 The middle dotted box is a selected stick-slip vibration, including two stages: sticking and slipping. From the speed approaching zero, the acceleration fluctuation decreases abnormally, and the fluctuation amplitude approaches zero, to the speed and acceleration fluctuation suddenly increasing, this part is the sticking stage; from the end of the sticking stage to the speed and acceleration fluctuation amplitude approaching zero again, this part is the slipping stage.

[0086] After obtaining the radial acceleration or tangential acceleration and speed data of stick-slip vibration, the data is processed. The speed data can be used to directly obtain the stick-slip index; the radial acceleration or tangential acceleration data needs to be processed by removing outliers, smoothing and normalizing the radial acceleration or tangential acceleration data using the above method, and then the processed radial acceleration or tangential acceleration data is used to obtain 8 eigenvalues.

[0087] A set of stick-slip vibration data can extract a set of 8 eigenvalues ​​and 1 stick-slip index. Taking the 8 eigenvalues ​​of radial acceleration or tangential acceleration as input, the trained stick-slip index prediction neural network model is used to predict the stick-slip index. Each set of stick-slip vibration eigenvalues ​​can get a predicted stick-slip index.

[0088] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation plans.

[0089] Identify and collect stick-slip vibration data in a certain area, such as Figure 7A complete set of stick-slip processes is regarded as one stick-slip vibration, and a total of 200 sets of stick-slip vibration data are collected, including rotation speed data and radial acceleration data, and the stick-slip index prediction neural network model trained in this embodiment is used to predict the stick-slip index.

[0090] The actual stick-slip index of 200 sets of stick-slip vibrations is calculated based on the measured speed data.

[0091] The radial acceleration data is processed using the data processing method in this embodiment. The box plot method is used to remove obvious outliers in the radial acceleration data, the moving average method is used to smooth the radial acceleration data, and finally the maximum and minimum method is used to normalize the radial acceleration data to limit the radial acceleration data to the interval [0,1].

[0092] Eight eigenvalues ​​of radial acceleration data during stick-slip vibration are calculated: kurtosis, shape factor, skewness, standard deviation, quartiles, variance, the proportion of sample points falling outside the range of 0.9 to 1.1 times the mean, and the proportion of sample points where the difference between two consecutive accelerations exceeds 0.1.

[0093] The eight eigenvalues ​​of radial acceleration are used as input, and the stick-slip index prediction neural network model trained by the method of this embodiment is used to predict the stick-slip index. The comparison between the predicted result and the actual stick-slip index is as follows: Figure 8 shown.

[0094] The dotted line in the figure is the stick-slip index calculated using the speed data, and the solid line is the stick-slip index predicted by the stick-slip index prediction neural network model trained using the method of this embodiment. It can be seen that the error between the predicted value and the actual value is small, and the prediction is relatively accurate. The correlation coefficient between the predicted result and the actual result is calculated to be 0.89059, indicating that the correlation between the predicted result and the actual result is strong, and the method of this embodiment predicts accurately.

[0095] This embodiment also provides a bottom hole assembly and a drill bit stick-slip vibration strength evaluation device, comprising:

[0096] An identification module for identifying whether a bottom hole assembly or a drill bit has stick-slip vibration;

[0097] An acquisition module is used to obtain radial acceleration or tangential acceleration data of stick-slip vibration when the bottom drill assembly or the drill bit undergoes stick-slip vibration; as an example, the acquisition module may be an acceleration sensor, and the acceleration sensor is installed on the bottom drill assembly or the drill bit to obtain radial acceleration or tangential acceleration data of the bottom drill assembly or the drill bit when the stick-slip vibration occurs.

[0098] A data processing module is used to process the radial acceleration or tangential acceleration data of the stick-slip vibration to obtain the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration;

[0099] The prediction module is used to use the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration as the input value of the pre-trained stick-slip index prediction neural network model to predict the stick-slip index.

[0100] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit, characterized in that: The following steps are involved: Obtain radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit during stick-slip vibration; Processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or the drill bit when the stick-slip vibration occurs to obtain characteristic values ​​of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when the stick-slip vibration occurs; The obtained characteristic values ​​of radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when stick-slip vibration occurs are used as input values ​​of a pre-trained stick-slip index prediction neural network model to predict the stick-slip index, and the stick-slip vibration intensity of the bottom hole assembly or the drill bit is evaluated based on the predicted stick-slip index.

2. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 1, characterized in that: The training method of the stick-slip index prediction neural network model includes: Use the rotor dynamics model to simulate the radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit when stick-slip vibration occurs, and provide raw data for the neural network model; Processing the original data to obtain the characteristic value of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit during the stick-slip vibration; using the rotational speed data during the simulation to calculate the stick-slip index of the bottom hole assembly or the drill bit during the stick-slip vibration; using the stick-slip index calculated by the rotational speed data as the output value of the neural network model, and using the characteristic value of the radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit during the stick-slip vibration as the input value of the neural network model; A genetic algorithm is used to optimize the neural network model to obtain a stick-slip index prediction neural network model, and the stick-slip index prediction neural network model is trained to obtain a trained stick-slip index prediction neural network model.

3. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 1 or 2, characterized in that: The method for processing the acquired radial acceleration or tangential acceleration data of the bottom hole assembly or drill bit is as follows: firstly, removing obvious outliers in the radial acceleration or tangential acceleration data, and then using the moving average method to smooth the radial acceleration or tangential acceleration data; finally, normalizing the radial acceleration or tangential acceleration, and limiting the radial acceleration or tangential acceleration data to the interval of [0,1].

4. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 3, characterized in that: The eigenvalues ​​of the radial acceleration or tangential acceleration are 8 eigenvalues, and the 8 eigenvalues ​​are kurtosis, shape factor, skewness, standard deviation, quartile, variance and two custom features; The first custom feature is defined as the proportion of sample points that fall outside the range of 1-n to 1+n times the average value relative to the total number of sample points, n<0.5; the second custom feature is defined as the proportion of sampling points where the difference between two consecutive accelerations exceeds m, m<0.

1.

5. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 3, characterized in that: Before using the genetic algorithm to optimize the stick-slip index prediction neural network model, a neural network structure is designed, wherein the neural network structure includes 8 input nodes, 1 hidden layer, the number of nodes is X, and 1 output neuron is set, that is, the network structure is 8-X-1; the 8 input nodes are 8 eigenvalues ​​of radial acceleration or tangential acceleration, the 1 output neuron is the predicted stick-slip index, and the number of hidden layer nodes X is 10-20; The number of hidden layer nodes is changed, and the genetic algorithm optimization is performed every time the number of hidden layer nodes is changed.

6. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 2, wherein: The neural network model for stick-slip index prediction optimized by genetic algorithm includes: Initialize the parameters to obtain the initial population; For each individual, a neural network is trained according to its weight and bias, and the characteristic values ​​of radial acceleration or tangential acceleration of the bottom hole assembly or the drill bit when the stick-slip vibration occurs are input to obtain a set of predicted stick-slip indices; the root mean square error between the predicted stick-slip indices and the actually calculated stick-slip indices is used as the cost function of the genetic algorithm; Sort the cost functions in ascending order, select the P*CR group with the smallest cost function as the parent generation, and randomly cross-combine to form a new P group of individuals; The P group of individuals generated by the parent crossover are allowed to mutate with a mutation rate of MR. The mutated P group of individuals serve as the new generation population and enter the next round of optimization; After each round of calculation of this function, the minimum cost function value obtained this time is compared with the minimum value obtained in the previous round; if the difference between the two minimum values ​​is less than 0.5%, the weight and deviation corresponding to the individual with the smallest cost function value at this time are the optimization results; conversely, if the difference between the two minimum values ​​is greater than 0.5%, it is necessary to enter the next round to continue optimization; set the maximum number of iterations, and when the number of population iterations reaches the maximum number of iterations, the weight and deviation combination corresponding to the individual with the lowest cost function value at this time is used as the final optimization result.

7. The method for evaluating the stick-slip vibration strength of a bottom hole assembly and a drill bit according to claim 2, wherein: The training of the stick-slip index prediction neural network model uses the mean square error (MSE) of the model as the standard for judging whether to stop model training; a threshold value Emin is set. When the difference between the MSE of this training and the MSE of the previous training is greater than Emin, the model continues to be trained; when the difference between the MSE of this training and the MSE of the previous training is less than Emin, the training is stopped. At the same time, a maximum number of training times is set. When the number of training times reaches the maximum value, the model training is stopped, and finally the accuracy of the model is tested.

8. A bottom hole assembly and drill bit stick-slip vibration strength evaluation device, characterized in that: include: An identification module for identifying whether a bottom hole assembly or a drill bit has stick-slip vibration; An acquisition module, used for acquiring radial acceleration or tangential acceleration data of stick-slip vibration when stick-slip vibration occurs in the bottom hole assembly or the drill bit; A data processing module is used to process the radial acceleration or tangential acceleration data of the stick-slip vibration to obtain the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration; The prediction module is used to use the characteristic value of the radial acceleration or tangential acceleration of the stick-slip vibration as the input value of the pre-trained stick-slip index prediction neural network model to predict the stick-slip index.

Images