Well track real-time intelligent prediction method

Abstract

The invention provides a real-time intelligent prediction method for a well track, and belongs to the technical field of oil and gas drilling, and the method comprises the steps: carrying out the data preprocessing based on the while-drilling inclinometry data and drilling recording data of a historical well and a drilled section, and constructing a data set; a GRU-BP parallel neural network is constructed, the GRU neural network and the BP neural network are searched and optimized, optimal parameters are compared and optimized, a well track intelligent prediction model based on the GRU-BP parallel neural network is established, and the prediction distance is optimized; and training a basic model based on the data set and the intelligent well track prediction model, and establishing a dynamic updating mechanism of the intelligent well track prediction model based on actually measured data of a drilling well section. According to the method, real-time data can be utilized to adapt to the change of an underground complex environment, and the model precision and generalization performance are effectively improved. And a technical support is provided for intelligent control of a well track in the future.

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas drilling, and in particular relates to a real-time intelligent prediction method for wellbore trajectory. Background Art

[0002] Oil and gas exploration and development are developing towards deep sea, deep layers and unconventional areas. Conventional vertical wells are difficult to achieve profitable development, and directional wells such as extended reach wells and horizontal wells have become the main development well types. Accurate prediction of wellbore trajectory is crucial for accurate targeting and safe drilling of directional wells, and is one of the important indicators of drilling quality. Improving the accuracy and timeliness of wellbore trajectory prediction will help increase the reservoir drilling rate and ensure drilling safety. However, there are many factors affecting wellbore trajectory and they are highly nonlinear. A single model is difficult to accurately predict, and the downhole sensor of the measurement while drilling is still some distance away from the drill bit, which makes it difficult for the trajectory parameters of the measured point to reflect the trajectory status at the drill bit in real time, increasing the difficulty of predicting wellbore trajectory.

[0003] Current research on wellbore trajectory prediction mainly focuses on geometric relationship extrapolation and mechanical analysis of bottom drilling tool assembly. Although the research on wellbore trajectory prediction has been relatively in-depth, it is still difficult to achieve accurate prediction for complex formations due to the large number of factors affecting the wellbore trajectory and its high nonlinearity, as well as the extremely complex mechanical behavior of the wellbore. For example, the geometric extrapolation method does not take into account the influence of the formation and the deformation of the drilling tool assembly, and its accuracy is insufficient, so it is currently rarely used; the mechanical analysis method has a complex mechanism and many assumptions, which cannot be met in many environments. At the same time, the complex interaction mechanism between the drill bit and the formation has not yet been fully clarified, and some empirical coefficients still need to be determined based on a large amount of actual drilling data.

[0004] In recent years, artificial intelligence technology has made great progress. It has multi-source, multi-modal, and multi-scale modeling capabilities, and has shown significant superiority and great potential in complex data processing, nonlinear fitting, and has been gradually applied to the drilling field as a new research paradigm, including lithology identification, wellbore trajectory optimization design, mechanical drilling rate prediction, and reservoir physical property parameter inversion. At present, there are still few intelligent prediction methods for wellbore trajectories, and the intelligent algorithms used are relatively old. They are all offline models used after training. The selection of input features is mainly based on time series features such as well inclination, azimuth, and wellbore curvature. It is believed that the influence of drilling parameters and formation parameters can be reflected by the trajectory parameters of the measured points. Therefore, non-time series features such as formation properties and bottom drilling tool assembly structure are not specifically considered, and the prediction accuracy needs to be improved. At the same time, the drilling environment is complex and changeable, and the offline model cannot use real-time data. The model has lag and insufficient generalization ability. Therefore, a high-precision and high-efficiency wellbore trajectory prediction method is urgently needed. Summary of the invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides a real-time intelligent prediction method for wellbore trajectory. The technical problem to be solved by the present invention is to solve the problem that the current intelligent model cannot utilize real-time data and has insufficient generalization performance, improve the accuracy and timeliness of wellbore trajectory prediction, and provide technical support for directional drilling.

[0006] In order to solve the above technical problems, the present invention provides a real-time intelligent prediction method for wellbore trajectory, comprising the following steps:

[0007] Step S1: Based on the inclination measurement while drilling data and drilling and logging data of historical wells and the drilled sections of the well being drilled, data preprocessing is performed to construct a data set for model training;

[0008] Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance;

[0009] Step S3: Use the data set formed in step S1 to train the wellbore trajectory intelligent prediction model established in step S2 to obtain a basic model, and then incrementally train the basic model based on the measured data of the drilling section to establish a dynamic update mechanism for the wellbore trajectory intelligent prediction model and update the model weights.

[0010] Furthermore, in step S1, the inclination measurement while drilling data is interpolated and the depth interval is unified with the drilling and logging data.

[0011] Furthermore, the LWD data are interpolated using cubic spline, cylindrical spiral method or natural curve method.

[0012] Furthermore, in step S1, abnormal data is detected and removed, missing values ​​are filled, and noise data is smoothed.

[0013] Furthermore, in step S1, abnormal data is detected and removed using 3σ, DBSCAN or a box plot algorithm combined with a sliding window, missing values ​​are filled using a linear filling or average filling method, and noise data is smoothed using an SG filter or a sliding average filter denoising algorithm.

[0014] Furthermore, in step S1, the non-time series features are digitized using one-hot encoding or sequential encoding.

[0015] Furthermore, in step S1, the Spearman correlation coefficient and the distance correlation coefficient are used to evaluate the importance of the time series features, and the drilling parameters and the logging while drilling parameters are optimized.

[0016] Furthermore, in step S1, the drilling parameters include drilling pressure, torque, drilling speed, rotary table speed and drilling fluid density.

[0017] Furthermore, in step S1, the logging while drilling parameters include well depth, vertical depth, well inclination, azimuth, wellbore curvature and natural gamma.

[0018] Furthermore, the calculation formula of the wellbore curvature is as follows:

[0019]

[0020]

[0021]

[0022] Where, K is the wellbore curvature, ° / 30m; K a is the well inclination change rate, ° / 30m; is the azimuth change rate, ° / 30m; Δa is the well inclination increment, °; is the azimuth increment, °; ΔH is the well depth increment, m; a c is the average well inclination angle of the measuring section, °.

[0023] Furthermore, in the step S1, the non-time-sequential features are optimized to select the bottom hole assembly structure, formation layering, drilling method and drill bit type.

[0024] Furthermore, in step S2, based on the GRU neural network and the BP neural network, a dual input structure is designed as the input of the GRU neural network and the BP neural network respectively, the time series features are input into the GRU neural network, and the non-time series features are input into the BP neural network; the last hidden layer of the GRU neural network and the last hidden layer of the BP neural network are concatenated for fusion training, followed by several hidden layers or an output layer.

[0025] Furthermore, in step S2, the parameter data output by the output layer is selected according to the result of the search and optimization.

[0026] Furthermore, in step S2, the output layer outputs the well inclination angle and azimuth angle, and then calculates the trajectory parameters according to the well depth, well inclination angle and azimuth angle.

[0027] Furthermore, in step S2, the trajectory parameters include vertical depth increment, horizontal projection increment, north-south coordinate increment, east-west coordinate increment, well inclination increment and azimuth increment.

[0028] Furthermore, in step S2, the trajectory parameter calculation formula based on the corrected average angle method is as follows:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] Where, ΔD is the vertical depth increment of the measuring section, m; ΔS is the horizontal projection increment of the measuring section, m; ΔN is the north-south coordinate increment of the measuring section, m; ΔE is the east-west coordinate increment of the measuring section, m; a c is the average azimuth of the measuring section, °; is the average azimuth of the measuring section, °; ​​a1 and a2 are the well inclination angles of the upper and lower measuring points, °; and is the azimuth of the upper and lower measuring points, °.

[0036] Furthermore, in step S2, orthogonal experiments and genetic algorithms are used to search and optimize the number of hidden layers, the number of neural units, the activation function, the optimizer and the training batches of the GRU neural network and the BP neural network, and the optimal parameters are compared and selected to establish an intelligent prediction model of wellbore trajectory based on the GRU-BP parallel network.

[0037] Furthermore, in step S2, the predicted distance is 10-30 meters.

[0038] Furthermore, in step S3, the inclination measurement while drilling data and drilling logging data of historical wells and the drilled sections of the drilling well are used as data sets to train a basic model, obtain the basic model neural network weights, and perform the following update process on the basic model: when the measured data of the drilling section is updated, the inclination measurement while drilling data between the measured point and the previous measured point is interpolated to make it have the same depth interval as the drilling logging data to form an updated data set, and the basic model neural network weights are used as initial weights, and the updated data set is input to perform incremental training. After the training is completed, the new basic model neural network weights are updated and stored, and predictions are performed using the new basic model; at the same time, wait for the next data update, and repeat the update process until the well section is drilled.

[0039] Furthermore, in step S3, the inclination measurement data between the actual measurement point and the previous measurement point is interpolated using a cubic spline interpolation function.

[0040] Furthermore, in step S3, the inclination measurement while drilling data includes the well inclination angle and azimuth angle.

[0041] The present invention provides a real-time intelligent prediction method for wellbore trajectory, which has the following beneficial effects compared with the prior art:

[0042] (1) It mainly uses historical data and current well surface drilling parameters and measurement while drilling data, which are easy to obtain and easy to promote and apply on site.

[0043] (2) Compared with the bottom drilling tool assembly mechanical analysis method and the finite element simulation method, the prediction based on the neural network does not need to consider the complex downhole mechanical action mechanism, and directly establishes an end-to-end mapping relationship between the influencing variables and the trajectory parameters, with higher accuracy and computational efficiency.

[0044] (3) A GRU-BP parallel network suitable for wellbore trajectory prediction was constructed. Compared with general neural networks, it can give full play to the respective advantages of GRU network and BP network and effectively extract the influence of temporal and non-temporal features on trajectory parameters.

[0045] (4) A dynamic model update mechanism was established. Compared with general intelligent algorithms, it can utilize real-time data, adapt to changes in the complex underground environment, and effectively improve model accuracy and generalization performance.

[0046] (5) Provide technical support for the future development of intelligent control of wellbore trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of an embodiment of a real-time intelligent prediction method for wellbore trajectory of the present invention.

[0048] Figure 2 It is a GRU-BP parallel neural network structure diagram of a real-time intelligent prediction method for wellbore trajectory of the present invention.

[0049] Figure 3 It is a schematic diagram of real-time intelligent prediction of wellbore trajectory of a real-time intelligent prediction method of wellbore trajectory of the present invention.

[0050] Figure 4 y is a schematic diagram of a model dynamic update based on incremental training of a real-time intelligent prediction method for wellbore trajectory of the present invention; t-k is the predicted value of the prediction result at time tk, y t-1 is the predicted value of the prediction result at time t-1, y t is the predicted value of the prediction result at time t, k is an integer less than t and greater than 0, and t is a time constant.

[0051] Figure 5 It is a prediction result diagram of well inclination and azimuth of a real-time intelligent prediction method for well trajectory of the present invention.

[0052] Figure 6It is a comparison diagram between the predicted values ​​and the measured values ​​of the well inclination angle and azimuth angle of a real-time intelligent prediction method for wellbore trajectory of the present invention.

[0053] Figure 7 The present invention is a flowchart of a real-time intelligent prediction method for wellbore trajectory. DETAILED DESCRIPTION

[0054] 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0057] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0058] In order to better understand the purpose, structure and function of the present invention, a real-time intelligent prediction method for wellbore trajectory of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0059] Embodiment 1:

[0060] like Figure 7 As shown, the present invention provides a real-time intelligent prediction method for wellbore trajectory, comprising the following steps:

[0061] Step S1: Based on the inclination measurement while drilling data and drilling and logging data of historical wells and the drilled sections of the well being drilled, data preprocessing is performed to construct a data set for model training;

[0062] Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance;

[0063] Step S3: Use the data set formed in step S1 to train the wellbore trajectory intelligent prediction model established in step S2 to obtain a basic model, and then incrementally train the basic model based on the measured data of the drilling section to establish a dynamic update mechanism for the wellbore trajectory intelligent prediction model and update the model weights.

[0064] Embodiment 2:

[0065] like Figure 7 As shown, the present invention provides a real-time intelligent prediction method for wellbore trajectory, comprising the following steps:

[0066] Step S1: Based on the inclination measurement while drilling data and drilling and logging data of historical wells and the drilled sections of the well being drilled, data preprocessing is performed to construct a data set for model training;

[0067] Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance;

[0068] Step S3: Use the data set formed in step S1 to train the wellbore trajectory intelligent prediction model established in step S2 to obtain a basic model, and then incrementally train the basic model based on the measured data of the drilling section to establish a dynamic update mechanism for the wellbore trajectory intelligent prediction model and update the model weights.

[0069] The difference between this embodiment and the first embodiment is that:

[0070] In the step S1, the measurement while drilling data is interpolated by using cubic spline, cylindrical spiral method or natural curve method, and the depth interval is unified with the drilling data.

[0071] In step S1, abnormal data are detected and removed by using algorithms such as 3σ, DBSCAN and box plot combined with sliding windows, missing values ​​are filled by linear filling and average filling methods, and high noise data are smoothed by using noise reduction algorithms such as SG filter and sliding average filter.

[0072] In the step S1, the non-time series features are digitized using one-hot encoding or sequential encoding.

[0073] In step S1, the Spearman correlation coefficient and distance correlation coefficient are used to evaluate the importance of the time series features, and the drilling parameters are optimized: drilling pressure, torque, drilling speed, rotary table speed and drilling fluid density; while drilling logging parameters: well depth, vertical depth, well inclination, azimuth, wellbore curvature and natural gamma.

[0074] The calculation formula of the wellbore curvature is as follows:

[0075]

[0076]

[0077]

[0078] Where, K is the wellbore curvature, ° / 30m; K a is the well inclination change rate, ° / 30m; is the azimuth change rate, ° / 30m; Δa is the well inclination increment, °; is the azimuth increment, °; ΔH is the well depth increment, m; a c is the average well inclination angle of the measuring section, °.

[0079] In the step S1, the non-time-sequential features are optimized according to drilling experience, and the bottom hole assembly structure, formation layering, drilling method and drill bit type are optimized.

[0080] Embodiment 3:

[0081] like Figure 7 As shown, the present invention provides a real-time intelligent prediction method for wellbore trajectory, comprising the following steps:

[0082] Step S1: Based on the inclination measurement while drilling data and drilling and logging data of historical wells and the drilled sections of the well being drilled, data preprocessing is performed to construct a data set for model training;

[0083] Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance;

[0084] Step S3: Use the data set formed in step S1 to train the wellbore trajectory intelligent prediction model established in step S2 to obtain a basic model, and then incrementally train the basic model based on the measured data of the drilling section to establish a dynamic update mechanism for the wellbore trajectory intelligent prediction model and update the model weights.

[0085] The difference between this embodiment and the above embodiments is that:

[0086] like Figure 1As shown, in step S2, according to the existing GRU neural network and BP neural network, a dual input structure is designed as the input of the GRU neural network and the BP neural network respectively, the time series features are input into the GRU neural network, and the non-time series features are input into the BP neural network; the last hidden layer of the GRU neural network and the last hidden layer of the BP neural network are spliced ​​together for fusion training, and then several hidden layers can be connected or directly connected to the output layer, such as Figure 2 As shown, the selection is made specifically according to the hyperparameter optimization results, the output layer neurons are set to 2, the well inclination angle and azimuth angle are output respectively, and then the trajectory parameters are calculated according to the well depth, well inclination angle and azimuth angle;

[0087] The trajectory parameters include vertical depth increment, horizontal projection increment, north-south coordinate increment, east-west coordinate increment, well inclination increment and azimuth increment.

[0088] The trajectory parameter calculation formula based on the corrected average angle method is as follows:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095] Where, ΔD is the vertical depth increment of the measuring section, m; ΔS is the horizontal projection increment of the measuring section, m; ΔN is the north-south coordinate increment of the measuring section, m; ΔE is the east-west coordinate increment of the measuring section, m; a c is the average azimuth of the measuring section, °; is the average azimuth of the measuring section, °; ​​a1 and a2 are the well inclination angles of the upper and lower measuring points, °; and is the azimuth of the upper and lower measuring points, °.

[0096] In step S2, orthogonal experiments, genetic algorithms, etc. are used to search and optimize the number of hidden layers, number of neural units, activation functions, optimizers, and training batches of the GRU neural network and the BP neural network, and the best parameters are compared and selected to establish a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network.

[0097] In step S2, the predicted distance is 10-30 meters.

[0098] Embodiment 4:

[0099] like Figure 7 As shown, the present invention provides a real-time intelligent prediction method for wellbore trajectory, comprising the following steps:

[0100] Step S1: Based on the inclination measurement while drilling data and drilling data of historical wells and drilled sections, data preprocessing is performed to construct a data set;

[0101] Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance;

[0102] Step S3: Based on the data set of step S1 and the wellbore trajectory intelligent prediction model, a basic model is trained, and based on the measured data of the drilling section, a dynamic update mechanism of the wellbore trajectory intelligent prediction model is established.

[0103] The difference between this embodiment and the above embodiments is that:

[0104] like Figure 3 and Figure 4 As shown, in step S3, the data of the historical wells and the drilled sections of the drilling well are used as data sets to train a basic model and obtain the basic model neural network weights. When the measured data of the drilling section is updated, the inclination data (well inclination and azimuth) between the measured point and the previous measured point is interpolated using the cubic spline interpolation function, and the depth interval is unified with the drilling data to form an updated data set. The basic model neural network weights are used as the initial weights, and the updated data set is input using the sliding window method to perform an incremental training. After the training is completed, the new basic model neural network weights are updated and stored, and the well inclination and azimuth are predicted with this new basic model, while waiting for the next data update, and the above process is repeated until the entire well section is drilled. The model update frequency can be set according to the measurement frequency of the on-site logging while drilling data.

[0105] Embodiment 5:

[0106] like Figure 7 As shown, a real-time intelligent prediction method for wellbore trajectory provided by an embodiment of the present invention comprises the following steps:

[0107] S101: Build a dataset.

[0108] Twelve wells in a well area are used as data sets, of which 11 wells are used for training and 1 well is used for testing. The cubic spline interpolation method is used to interpolate the inclination data, and the depth interval is unified with the drilling data. The time series features and non-time series features are optimized by the Spearman correlation coefficient method and drilling history experience. The 3σ method combined with the sliding window is used to detect outliers for time series features, the linear filling method is used to fill missing values, and the SG filter is used for smooth noise reduction. The non-time series features are digitized using one-hot encoding to construct a data set.

[0109] S102: Construct a GRU-BP parallel neural network.

[0110] according to Figure 2 The model was constructed as shown in the figure, and the network hyperparameters were optimized by orthogonal experimental method, including the number of hidden layers, number of neural units, activation function, Dropout, optimizer, etc. of GRU network and BP network. The final optimization results are shown in Table 1. At the same time, the prediction distance was optimized. According to the situation of this data set, the distance was selected as 15m in front of the measuring point.

[0111] Table 1. Hyperparameter optimization results of GRU-BP parallel network

[0112]

[0113] S103: Construct a dynamic model update mechanism to predict wellbore trajectories.

[0114] according to Figure 4 The dynamic update mechanism of the S102 model is constructed in this process. First, the basic model is trained using the data of 11 historical wells to obtain the basic model weights. The predicted well data is segmented using a sliding window to simulate the downhole measurement while drilling data update process. Each time a set of data is updated, the model is incrementally trained once, the model weights are updated, and the well inclination and azimuth at 15m below the current measurement point are predicted. The prediction results are shown in Table 2. Figure 5 and Figure 6 As shown, compared with the measured data, the average absolute error of azimuth is 0.1°, the maximum error is 0.85°, and the root mean square error is 0.02°; the average absolute error of azimuth is 0.18°, the maximum error is 1.18°, and the root mean square error is 0.08°.

[0115] Table 2 Wellbore trajectory prediction error

[0116]

[0117] 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 are included in the protection scope of the present invention.

Claims

1. A real-time intelligent prediction method for wellbore trajectory, characterized in that: The following steps are involved: Step S1: Based on the inclination measurement while drilling data and drilling and logging data of historical wells and the drilled sections of the well being drilled, data preprocessing is performed to construct a data set for model training; Step S2: constructing a GRU-BP parallel neural network, searching and optimizing the GRU neural network and the BP neural network, comparing and optimizing the optimal parameters, establishing a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network, and optimizing the prediction distance; Step S3: Use the data set formed in step S1 to train the wellbore trajectory intelligent prediction model established in step S2 to obtain a basic model, and then incrementally train the basic model based on the measured data of the drilling section to establish a dynamic update mechanism for the wellbore trajectory intelligent prediction model and update the model weights.

2. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In step S1, the inclination measurement while drilling data is interpolated and the depth interval is unified with the drilling and logging data.

3. The real-time intelligent prediction method for wellbore trajectory according to claim 2, characterized in that: The MWD data are interpolated using cubic spline, cylindrical spiral or natural curve methods.

4. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S1, abnormal data is detected and removed, missing values ​​are filled, and noise data is smoothed.

5. The real-time intelligent prediction method for wellbore trajectory according to claim 4, characterized in that: In step S1, abnormal data are detected and removed by using 3σ, DBSCAN or box plot algorithm combined with sliding window, missing values ​​are filled by linear filling or average filling method, and noise data are smoothed by using SG filter or sliding average filter noise reduction algorithm.

6. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S1, the non-time series features are digitized using one-hot encoding or sequential encoding.

7. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S1, the Spearman correlation coefficient and the distance correlation coefficient are used to evaluate the importance of the time series features, and the drilling parameters and the logging while drilling parameters are optimized.

8. The real-time intelligent prediction method for wellbore trajectory according to claim 7, characterized in that: In the step S1, the drilling parameters include drilling pressure, torque, drilling speed, rotary table speed and drilling fluid density.

9. The real-time intelligent prediction method for wellbore trajectory according to claim 7, characterized in that: In step S1, the logging while drilling parameters include well depth, vertical depth, well inclination, azimuth, wellbore curvature and natural gamma.

10. The real-time intelligent prediction method for wellbore trajectory according to claim 9, characterized in that: The calculation formula of the wellbore curvature is as follows: Where, K is the wellbore curvature, ° / 30m; K a is the well inclination change rate, ° / 30m; is the azimuth change rate, ° / 30m; Δa is the well inclination increment, °; is the azimuth increment, °; ΔH is the well depth increment, m; a c is the average well inclination angle of the measuring section, °.

11. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S1, the non-time-sequential features are optimized to select the bottom hole assembly structure, formation layering, drilling method and drill bit type.

12. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S2, based on the GRU neural network and the BP neural network, a dual input structure is designed as the input of the GRU neural network and the BP neural network respectively, the time series features are input into the GRU neural network, and the non-time series features are input into the BP neural network; the last hidden layer of the GRU neural network and the last hidden layer of the BP neural network are spliced ​​together for fusion training, followed by several hidden layers or an output layer.

13. The real-time intelligent prediction method for wellbore trajectory according to claim 12, characterized in that: In step S2, the parameter data output by the output layer is selected according to the result of the search and optimization.

14. The real-time intelligent prediction method for wellbore trajectory according to claim 13, characterized in that: In step S2, the output layer outputs the well inclination angle and azimuth angle, and then calculates the trajectory parameters according to the well depth, well inclination angle and azimuth angle.

15. The real-time intelligent prediction method for wellbore trajectory according to claim 14, characterized in that: In step S2, the trajectory parameters include vertical depth increment, horizontal projection increment, north-south coordinate increment, east-west coordinate increment, well inclination increment and azimuth increment.

16. The real-time intelligent prediction method for wellbore trajectory according to claim 15, characterized in that: In step S2, the trajectory parameter calculation formula based on the corrected average angle method is as follows: Where, ΔD is the vertical depth increment of the measuring section, m; ΔS is the horizontal projection increment of the measuring section, m; ΔN is the north-south coordinate increment of the measuring section, m; ΔE is the east-west coordinate increment of the measuring section, m; a c is the average well inclination angle of the measuring section, °; is the average azimuth of the measuring section, °; ​​a1 and a2 are the well inclination angles of the upper and lower measuring points, °; and is the azimuth of the upper and lower measuring points, °.

17. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In step S2, orthogonal experiments and genetic algorithms are used to search and optimize the number of hidden layers, the number of neural units, the activation function, the optimizer and the training batch of the GRU neural network and the BP neural network, and the best parameters are compared and selected to establish a wellbore trajectory intelligent prediction model based on the GRU-BP parallel network.

18. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In step S2, the predicted distance is 10-30 meters.

19. The real-time intelligent prediction method for wellbore trajectory according to claim 1, characterized in that: In the step S3, the inclination measurement while drilling data and drilling and logging data of the drilled sections of the historical wells and the drilling wells are used as data sets to train a basic model, obtain the basic model neural network weights, and perform the following update process on the basic model: when the measured data of the drilling section is updated, the inclination measurement while drilling data between the measured point and the previous measured point is interpolated to make it have the same depth interval as the drilling and logging data to form an updated data set, and the basic model neural network weights are used as initial weights, and the updated data set is input to perform an incremental training. After the training is completed, the new basic model neural network weights are updated and stored, and prediction is performed with the new basic model; Meanwhile, the system waits for the next data update and repeats the update process until the well section is drilled.

20. The real-time intelligent prediction method for wellbore trajectory according to claim 19, characterized in that: In step S3, the inclination data between the actual measured point and the previous measured point are interpolated using a cubic spline interpolation function.

21. The real-time intelligent prediction method for wellbore trajectory according to claim 20, characterized in that: In step S3, the inclination measurement while drilling data includes the well inclination angle and azimuth angle.

Images