CN113464120B - Tool face state prediction method and system, and sliding directional drilling method and system - Google Patents
Tool face state prediction method and system, and sliding directional drilling method and system Download PDFInfo
- Publication number
- CN113464120B CN113464120B CN202111035640.2A CN202111035640A CN113464120B CN 113464120 B CN113464120 B CN 113464120B CN 202111035640 A CN202111035640 A CN 202111035640A CN 113464120 B CN113464120 B CN 113464120B
- Authority
- CN
- China
- Prior art keywords
- data
- time
- directional drilling
- tool face
- torsional pendulum
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH DRILLING; MINING
- E21B—EARTH DRILLING, e.g. DEEP DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B47/00—Survey of boreholes or wells
- E21B47/02—Determining slope or direction
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH DRILLING; MINING
- E21B—EARTH DRILLING, e.g. DEEP DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B7/00—Special methods or apparatus for drilling
- E21B7/04—Directional drilling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides a method and a system for predicting a tool face state and a method and a system for sliding directional drilling, wherein the method for predicting the tool face state comprises the following steps: acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well body data of sliding directional drilling operation; selecting at least one type of collected data as features, and fusing according to the time stamps to form a time sequence data set; training an LSTM neural network model one by using the data set to obtain an LSTM-based tool face state prediction model; in the sliding directional drilling operation process, the tool face states at the current time and the previous 4 times, the target tool face, the forward torque and the reverse torque form a data sequence as input, and the prediction results of the tool face states at the future 3 times are obtained. The invention can judge the future motion state of the tool surface and give corresponding control parameters as the reference of engineers.
Description
Technical Field
The invention relates to the technical field of oil and gas drilling engineering, in particular to a tool face state prediction method, a tool face state prediction system, a sliding directional drilling method and a sliding directional drilling system.
Background
The tool face angle control is the key of the sliding directional well track control, and in the manual sliding directional drilling process, an engineer empirically predicts the future tool face change trend according to historical sliding directional parameters and the tool face angle change condition so as to adopt a corresponding control strategy. However, in the case of automatic sliding directional drilling, due to the lack of a mechanism for effectively predicting a future toolface, the control strategy is lack of predictability, and the ideal effect of toolface control is difficult to achieve. Therefore, a set of sliding orientation tool surface prediction method is formed, and the method has important significance for improving the control efficiency of the sliding orientation track.
For example, patent documents, entitled machine learning-based sliding directional drilling tool face state identification method and publication number CN112360341A, published on 2/12/2021 describe a machine learning-based sliding directional drilling tool face state identification method, which provides a discrete state model of a sliding directional drilling tool face, on the basis of which Onehot coding is performed on various states and a BP neural network is constructed, and a tool face state identification network is obtained after training, so that states such as tool face error data, a stable state, a change direction, a change speed and the like can be accurately identified, real-time monitoring of the tool face state is realized, and the problems of high difficulty and low accuracy in manual identification of the tool face state are solved, thereby assisting a field engineer in performing sliding directional drilling operation. However, this patent application has the following drawbacks: the real-time monitoring and judgment of the tool surface state can be realized only, and the motion state of the tool surface cannot be predicted and judged in advance; the engineer is only assisted in obtaining the real-time state of the tool face and is not helped to further give control parameters for the future motion state.
Disclosure of Invention
The present invention aims to address at least one of the above-mentioned deficiencies of the prior art. For example, it is an object of the present invention to provide a tool face state prediction method and system that can effectively determine the future motion state of a tool face.
In order to achieve the above object, an aspect of the present invention provides a method for predicting a tool face state, the method comprising:
s1, acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well body data of the sliding directional drilling operation;
s2, selecting at least one of torsional pendulum system data, measurement while drilling system data, logging system data and well data as data characteristics, and fusing according to time stamps to form a time sequence data set;
step S3, training LSTM neural network models one by using a time series data set to obtain a tool surface state prediction model based on LSTM;
and step S4, in the sliding directional drilling operation process, the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times are used as input data sequences, and the prediction results of the tool surface states at the future 3 times are obtained.
In an exemplary embodiment of the inventive method of predicting a toolface state, the torsional pendulum system data may include a forward torque, a reverse torque, a forward rotational speed, a reverse rotational speed, a forward hold time, and a reverse hold time; the measurement-while-drilling system data may include a target toolface and an actual toolface; the logging system data includes weight on bit and displacement; the well bore data may include well depth, horizontal leg length, and well angle.
In an exemplary embodiment of the method for predicting the tool face state, the data acquisition frequency of the torsional pendulum system may be 30 s/time to 1200 s/time; the data acquisition frequency of the measurement while drilling system can be 20-180 s/time; the data acquisition frequency of the logging system can be 4 s/time to 5 s/time.
In an exemplary embodiment of the method for predicting a toolface state of the present invention, step S2 may include: and performing data cleaning on the time-series data set to remove invalid data and redundant data.
In an exemplary embodiment of the method for predicting a toolface state of the present invention, the invalid data and the redundant data may include: data collected when working condition changes occur in the torsional pendulum orientation process, data collected when operation is wrong in the torsional pendulum orientation process, and data collected when the situation is inconsistent with basic rules or expectations occurs in the torsional pendulum orientation process.
In an exemplary embodiment of the method for predicting the toolface state of the present invention, step S2 may further include a well-dividing and time-sharing process: and dividing the cleaned time series data set according to the time period of each directional drilling operation to form a plurality of subdata sets to be used as a training data basis of the LSTM neural network model.
In an exemplary embodiment of the method for predicting the toolface state of the present invention, step S2 may further include data preprocessing: the sub data sets are standardized to avoid interference between data features of different magnitudes.
In an exemplary embodiment of the tool-face state prediction method of the present invention, the specific method for training the LSTM neural network model one by one using the time series data set may be as follows: and building at least one layer of LSTM neural network model, splicing the fully-connected neural network at the rear end, training the LSTM neural network models one by using a data set, and simultaneously carrying out hyper-parameter adjustment, optimization and evaluation to obtain the tool face state prediction model based on the LSTM.
In an exemplary embodiment of the method for predicting a toolface state of the present invention, step S4 may further include: and judging the future motion state of the tool surface according to the prediction result of the tool surface state at the future 3 moments to obtain the control parameters of the torsional pendulum system.
The invention further provides a sliding directional drilling method, which adopts the prediction results of the tool surface states at the 3 future moments to predict the control parameters of the torsional pendulum system at the future moments and controls the torsional pendulum system according to the prediction results of the control parameters of the torsional pendulum system at the future moments, wherein the control parameters of the torsional pendulum system comprise forward torque and reverse torque, and the prediction results of the tool surface states at the 3 future moments are obtained by adopting the tool surface state prediction method.
In still another aspect, the present invention provides a tool surface motion state prediction system, which may include a data acquisition unit, a time-series data set output unit, a neural network training unit, and a tool surface state prediction unit, wherein,
the data acquisition unit is configured to be capable of acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well bore data of the sliding directional drilling operation;
the time sequence data set output unit is connected with the data acquisition unit and is configured to be capable of selecting at least one of torsional pendulum system data, measurement while drilling system data, logging system data and well data as data characteristics and outputting a time sequence data set after fusion;
the neural network training unit is connected with the time sequence data set output unit and is configured to be capable of training the LSTM neural network model one by using the time sequence data set to obtain an LSTM-based tool face state prediction model;
the tool surface state prediction unit is connected with the neural network training unit and is configured to input the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times as an input data sequence to the LSTM-based tool surface state prediction model and output the prediction result of the tool surface state at the future 3 times.
Yet another aspect of the present invention provides a sliding directional drilling system, which may include: a system for predicting tool face motion as described above; and
and the torsional pendulum system control unit is connected with the tool surface state prediction unit and is configured to be capable of adjusting the control parameters of the torsional pendulum system at the future time according to the prediction results of the tool surface states at the future 3 times.
Compared with the prior art, the beneficial effects of the invention can comprise at least one of the following:
(1) the future motion state of the tool surface can be judged, and corresponding control parameters are given as reference of an engineer;
(2) the control performance of the reinforcement learning algorithm is improved, the problem that the existing reinforcement learning algorithm is poor in cross-well adaptability is solved, and the cross-well adaptability is good;
(3) the directional drilling device can assist a field engineer to perform sliding directional drilling operation, and improves the operation efficiency.
Drawings
The above and other objects and/or features of the present invention will become more apparent from the following description taken in conjunction with the accompanying drawings, in which:
fig. 1 is a diagram illustrating a relationship between a current toolface and a target toolface according to an exemplary embodiment of the method for predicting a toolface state of the present invention.
FIG. 2 shows a schematic structural diagram of a 2-layer LSTM neural network model of an exemplary embodiment of the tool-face state prediction method of the present invention.
FIG. 3 shows a flow diagram of a layer 2 LSTM neural network model of an exemplary embodiment of the tool face state prediction method of the present invention.
Detailed Description
Hereinafter, the tool face state prediction method and system, and the sliding directional drilling method and system of the present invention will be described in detail with reference to exemplary embodiments.
It should be noted that, for those skilled in the art, the term "pressure" is used herein to correspond to pressure in part.
The invention provides a tool face state prediction method.
In an exemplary embodiment of the present invention, a method of predicting a toolface state may include:
and S1, acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well body data of the sliding directional drilling operation.
Further, in step S1, the torsional pendulum system data may include a forward torque, a reverse torque, a forward rotation speed, a reverse rotation speed, a forward holding time, and a reverse holding time; the measurement-while-drilling system data may include a target toolface and an actual toolface; the logging system data may include weight on bit and displacement; well bore data may include well depth, horizontal leg length, and well angle.
Further, the data of the torsion pendulum system are collected at a frequency of 30 s/time to 1200 s/time, for example, the torsion pendulum system collects real-time torque and revolution frequency of 0.5 s/time, and the frequency of collecting given torque, revolution and holding time can be 30 s/time to 1200 s/time. The data acquisition frequency of the measurement while drilling system can be 20 s/time to 180 s/time. The data acquisition frequency of the logging system can be 4 s/time to 5 s/time.
And S2, selecting at least one of torsional pendulum system data, Measurement While Drilling (MWD) data, logging system data and well bore data as data characteristics, and fusing according to the time stamps to form a time sequence data set.
Further, in step S2, the inventors found through research that: the target toolface, actual toolface, forward torque, and reverse torque may be selected as relevant data characteristics for the torsional pendulum system, measurement-while-drilling system, logging system, and the prediction of toolface conditions.
Further, step S2 may also include data cleansing: and performing data cleaning on the time-series data set to remove invalid data and redundant data.
It should be noted that the invalid data and the redundant data may include: data collected when working condition changes occur in the torsional pendulum orientation process, data collected when operation is wrong in the torsional pendulum orientation process, and data collected when the situation is inconsistent with basic rules or expectations occurs in the torsional pendulum orientation process. This is because the data frequency of the torsional pendulum system is different, the tool face acquired by the MWD system is affected by the transmission environment, and there is a certain error probability, and the data acquisition frequency of the logging system is high and there is a certain data fluctuation. The existence of these data adversely affects the later model training and therefore culls them.
Further, step S2 may further include a well-dividing and time-sharing process: in order to avoid interference of different geological environments and well section depths, the cleaned time sequence data set can be divided according to the time period of each directional drilling operation to form a plurality of subdata sets which are used as a training data basis of an LSTM (Long Short-Term Memory) neural network model.
Still further, step S2 may also include data preprocessing: the subdata sets may be separately standardized to avoid interference between data features of different magnitudes.
And step S3, training the LSTM neural network models one by using the time sequence data set to obtain an LSTM-based tool face state prediction model.
Further, the specific method for training the LSTM neural network model one by one using the time series data set in step S3 may be as follows: and building at least one layer of LSTM neural network model, splicing the fully-connected neural network at the rear end for reducing the dimension, training the LSTM neural network models one by using a data set, and simultaneously performing hyper-parameter adjustment, optimization and evaluation to obtain an LSTM-based tool face state prediction model.
And step S4, in the sliding directional drilling operation process, the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times are used as input data sequences, and the prediction results of the tool surface states at the future 3 times are obtained.
Further, in step S4, the future motion state of the tool surface may be determined according to the predicted tool surface state at the future 3 times, so as to obtain the control parameters of the torsional pendulum system.
In another aspect, the invention provides a method of sliding directional drilling.
In an exemplary embodiment of the invention, a sliding directional drilling method may predict a control parameter of a torsional pendulum system at a future time using a prediction of a toolface state at 3 future times and control the torsional pendulum system according to the prediction of the control parameter of the torsional pendulum system at the future time. That is, the predicted results of the toolface state at 3 future times may be used to adjust the control parameters of the torsional pendulum system at the future times. The control parameters of the torsional pendulum system comprise forward torque and reverse torque, and the prediction results of the tool face state at the future 3 moments are obtained by adopting the tool face state prediction method.
In yet another aspect, the present invention provides a system for predicting tool face motion.
In an exemplary embodiment of the present invention, a system for predicting a tool face motion state may include a data acquisition unit, a time series data set output unit, a neural network training unit, and a tool face state prediction unit.
Wherein the data acquisition unit is configured to be able to acquire torsional pendulum system data, measurement while drilling system data, logging system data, and well bore data of the sliding directional drilling operation.
The time sequence data set output unit is connected with the data acquisition unit and is configured to be capable of selecting at least one of torsional pendulum system data, measurement while drilling system data, logging system data and well bore data as data characteristics, and outputting the time sequence data set after fusion.
The neural network training unit is connected with the time sequence data set output unit and is configured to train LSTM neural network models one by using the time sequence data set to obtain LSTM-based tool face state prediction models.
The tool surface state prediction unit is connected with the neural network training unit and is configured to input the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times as an input data sequence to the LSTM-based tool surface state prediction model and output the prediction result of the tool surface state at the future 3 times.
In yet another aspect, the present invention provides a sliding directional drilling system.
In an exemplary embodiment of the invention, a sliding directional drilling system may include: a system for predicting tool face motion as described above; and the torsional pendulum system control unit is connected with the tool surface state prediction unit and is configured to be capable of adjusting the control parameters of the torsional pendulum system at the future time according to the prediction results of the tool surface states at the future 3 times.
Yet another aspect of the invention provides a computer program for a method of predicting based on a toolface state.
In an exemplary embodiment of the present invention, at least one of the method for predicting a toolface state and the sliding directional drilling method of the present invention can be compiled into a corresponding extent code or instruction and programmed as a computer program. When the program code or instructions are executed by the processor, steps S2 to S4 of the method for predicting a toolface state can be implemented to obtain a prediction of a toolface state at 3 future times, or partial steps of the method for sliding directional drilling can be implemented to obtain a prediction of a control parameter of the torsional pendulum system at a future time.
Yet another aspect of the present invention provides a computer-readable storage medium storing a computer program.
In an exemplary embodiment of the invention, a computer program is stored in a computer readable storage medium, which when executed, may implement at least one of the inventive method of predicting a toolface state and sliding directional drilling method. The computer readable storage medium may be any data storage device that stores data that can be read by a computer system. For example, examples of computer-readable storage media may include: read-only memory, random access memory, read-only optical disks, magnetic tapes, floppy disks, optical data storage devices, and carrier waves (such as data transmission through the internet via wired or wireless transmission paths).
Yet another aspect of the present invention provides a computer apparatus for predicting a toolface state.
In one exemplary embodiment of the present invention, a computer apparatus for predicting a toolface state may include a processor and a memory. The memory is for storing a computer program for implementing at least one of the toolface state prediction method and the sliding directional drilling method of the present invention.
In order to better explain the above exemplary embodiments of the invention, a further description is given below in connection with the specific examples and the accompanying drawings.
Fig. 1 is a schematic diagram of a relationship between a current toolface and a target toolface, wherein a region pointed by a box a in fig. 1 represents the current toolface, and a region pointed by a box B represents the target toolface. In general, the target toolface is a toolface specified in the orientation, and the current toolface is a current toolface acquired by MWD, and reference numerals 1 to 5 in fig. 1 respectively indicate tool face angle values acquired from the MWD system at five time points from far to near in time. As can be seen from fig. 1, the toolface is continuously close to the target toolface area with time, and there is still a certain difference between the current toolface and the target toolface, and the engineer needs to adjust the toolface of the downhole tool to the area close to the target toolface through the action of the torsional pendulum system, for example, adjust the toolface to the range of ± 10 degrees of the target toolface.
In order to achieve the above object, the present example provides a tool face state prediction method capable of predicting a future change trend of a tool face by predicting an angular distance, approaching speed, or departing speed of a current tool face from a target tool face.
Specifically, a method for predicting a tool face state may include the steps of:
(1) data acquisition: collecting forward torque, reverse torque, forward rotating speed, reverse rotating speed, forward holding time and reverse holding time of a torsional pendulum system; acquiring a tool face of the MWD system; acquiring the bit pressure and the discharge capacity of a logging system; and collecting the well depth, the horizontal section length and the well inclination angle of the directional well.
It should be noted that the future trend of the tool face is closely related to some of the parameters mentioned above.
Firstly, parameters of a torsion pendulum system: if the tool face is in equilibrium (i.e., at a certain value, the tool face is just held still), the values of the forward torque, the forward rotational speed and the forward holding time increase, and the tool face tends to rotate clockwise to the target tool face. The greater the change in value, the greater the speed at which the tool face approaches or moves away clockwise from the target tool face. And on the contrary, the reverse torque, the rotating speed and the holding time have similar laws. That is, if the tool face is at equilibrium (i.e., at some value, the tool face is just held stationary), the values of the reverse torque, reverse rotational speed, and reverse hold time increase, and the tool face tends to rotate more counterclockwise toward the target tool face. The greater the change in value, the greater the speed at which the tool face approaches or moves away counterclockwise from the target tool face.
Logging system parameters: parameters closely related to the toolface include weight-on-bit and displacement, wherein the greater the weight-on-bit, the more counter-clockwise the toolface tends; the larger the displacement, the more the tool face tends to change counterclockwise.
Well bore data: the larger the numerical values of the well depth, the horizontal section length, the well inclination angle and the like are, the slower the speed of the change of the tool surface caused by the change of the parameters of the torsional pendulum system is.
(2) Data multi-feature fusion: and fusing the multiple characteristics from the three systems of torsional pendulum, MWD and logging according to the time stamp to form a time sequence data set.
Further, one or more types of the collected data can be selected as features and fused according to the time stamps to form a time series data set.
The data to be fused as the features can be selected and fused into the data set according to the experience of the directional engineer, that is, the determined parameters include the above parameters as the key parameters for training according to the experience of the directional engineer and the action features of the torsion pendulum system.
(3) Data cleaning: invalid, wrong and redundant data in the time sequence data set are removed, and the data training efficiency is improved.
The data sources include 3 systems: the torsional pendulum system, the MWD system and the logging system have the problems of large data fluctuation range and partial errors due to different data frequencies. For example, the data frequency of the torsion pendulum system is different, wherein the frequency of the torque and the rotating speed collected in real time reaches a value of 0.5 second, the value fluctuation is large, and the frequency of the given torque, the given rotating speed and the given holding time is 30 seconds to 10 minutes. The MWD system collects tool face frequencies in the interval 20 seconds to 3 minutes, and there is a certain probability of error due to the influence of the transmission environment. The frequency of data of the logging system is high, approximately 4-5 s of data, and certain data fluctuation exists.
Therefore, data cleaning can be performed after the data set is formed, invalid, wrong and redundant data in the time sequence data set are removed, and the data training efficiency is improved.
Invalid and redundant data mainly occur, due to the complexity and uncertainty of the torsional pendulum orientation process, the conditions of working condition change, human intention operation change and inconsistency with basic rules or expectations during operation occur, and the existence of the data can cause adverse effects on later model training, so the data needs to be removed.
(4) Well-dividing, time-interval processing and data preprocessing: in order to avoid the interference of different geological environments and well section depths, the cleaned data set is divided according to the directional time period of each time to form a plurality of subdata sets which are used as the training data basis of the neural network model. Meanwhile, the subdata sets are standardized respectively, and mutual interference among data characteristics of different magnitudes is avoided.
Because the torsional pendulum is oriented to different wells, people may not find how much difference from the surface because of the difference of well hole environment, stratum characteristics, well hole tracks, drilling fluid performance, top drive performance and the like, but in the actual use process, when the same tool face adjustment is realized, the numerical values of torsional pendulum parameters and logging parameters and the change rate have larger difference, and the data difference amplitude even reaches 0.6-1 time, so that a great problem exists if the data of all wells are mixed together for training. In addition, when the same well is in different time periods and different well depths, corresponding environments sometimes have larger changes, corresponding torsional pendulum parameters and logging parameters also have larger differences, and tool face characteristics also have different characteristics, so that time-sharing processing is also necessary.
The well-dividing and time-sharing processing can avoid the situation, the data set obtained by orienting each time of one well is used as a relatively independent subdata set after the data is cleaned, and the parameters of one well can not be changed too much in the orienting process at a certain time, so the problem does not exist.
After processing, a plurality of sub-data sets are formed and then trained one by one, so that the LSTM neural network model extracts the experience of accumulating each sub-data set.
For example, the specific method for standardizing each sub data set is as follows: adopting minimum-maximum normalization to the tool surface parameters according to a formula (1), carrying out linear transformation on original data, and mapping numerical values between [0,1 ]; and (3) normalizing other data by adopting standard deviation according to a formula (2), wherein the mean value of the processed data is 0, and the variance is 1.
In the formula (I), the compound is shown in the specification,
as a feature mean, σ is a standard deviation, x is raw data, and x is normalized data.
(5) Building and training a neural network: and (3) building a 2-layer LSTM neural network model, splicing the fully-connected neural network at the rear end for dimensionality reduction, inputting the data characteristics selected in the step (1) as the neural network, carrying out network training on each sub data set formed in the step (4), and obtaining an LSTM-based tool face state prediction model after super-parameter adjustment, optimization and evaluation, wherein the model is output as tool face states at three moments in the future (as shown in fig. 2 and 3).
Wherein, the FC in fig. 2 is called Fully Connected Neural Network, which represents a Fully Connected Network; y is the network output, and y0, y1, y2 and y3 respectively represent the output of the network at different moments; t is a network input, and the numerals indicate network inputs at different times, for example, T1 includes a current time input parameter and a previous time network output y0, T2 includes a current time input parameter and a previous time network output y1, T3 includes a current time input parameter and a previous time network output y2, T4 includes a current time input parameter and a previous time network output y3, and T50 includes a current time input parameter and a previous time network output y 49. P in FIG. 3 is the model output, P1, P2, PL represent the model output at different times, respectively; c is the state parameter of the internal unit structure of the model, C1, C2 and CL-1 respectively represent the state parameter of the internal unit structure of the model at different moments in the prediction process; h is the parameter of the internal hidden layer of the model, and H1, H2 and HL-1 respectively represent the parameter change of the internal hidden layer of the model at different moments.
Fig. 2 shows the complete data flow process from input parameters to output parameters, and the hidden layer comprises two LSTM layers and two DropOut layers, the two LSTM layers are identical in structure and both consist of cell state parameters and a gate structure, and the gate structure is a full connection layer plus a Sigmoid activation function. Fig. 2 describes the data flow in the prediction process, the input data at each moment consists of the current input parameters and the output data at the last moment, and the complete data is continuously predicted in a circulating manner.
Fig. 3 shows a process of network training, in which sliding orientation data is first processed into a time series data set, and then divided, normalized, and the like, and then input into a network to obtain prediction data, and according to the calculation loss of the prediction data and real data, network parameters are improved through an optimization algorithm, and then a new round of training is performed. And continuously iterating through the steps of inputting, outputting, calculating loss and optimizing network parameters until the error between the predicted data and the real data reaches a specified index, and finishing training to obtain a trained model.
In the sliding orientation operation process, the tool surface states at the current moment and the previous 4 moments and a data sequence consisting of a target tool surface, a forward torque and a reverse torque are used as input, the tool surface states at the next three moments are obtained, the future motion state of the tool surface can be judged according to the result, and corresponding control parameters are given.
It should be noted that the LSTM-based sliding orientation toolface simulation environment is divided into three stages:
(i) data interface adaptation
The tool surface state prediction model based on the LSTM inputs the tool surface states at the current time and the previous 4 times, the target tool surface, the forward torque and the reverse torque to form a data sequence, and outputs the tool surface states at the next three times. The reinforcement learning algorithm inputs are the actual tool face, the target tool face, the positive and negative torques at the current moment.
The data flow direction is: and (3) giving an input data sequence at the current moment, giving forward and reverse torques of corresponding control parameters by the reinforcement learning algorithm, giving a possible tool face angle at the next moment by the tool face state prediction algorithm, and combining the tool face angle, the forward and reverse torques of the corresponding control parameters given by the reinforcement learning algorithm and the target tool face at the moment as the input of the reinforcement learning algorithm again to complete the closed-loop feedback of data.
(ii) Reinforcement learning algorithm simulation training
And (3) performing simulated directional online learning by the reinforcement learning algorithm according to the data flow direction in the step (i), improving the self control performance through multiple iterations, and learning the adjustment modes under the condition of various drilling tool surface changes, so that the self control effect is closer to the manual control experience.
(iii) Field test by using reinforcement learning algorithm obtained by training
In the process of the field sliding orientation operation, the reinforcement learning algorithm takes the tool surface state at the current moment and a data sequence formed by the target tool surface, the forward torque and the reverse torque as input, obtains the forward torque and the reverse torque of the torsional pendulum control parameter, and performs automatic control.
In summary, the beneficial effects of the invention can include:
(1) the future motion state of the tool surface can be judged, and corresponding control parameters are given as reference of an engineer;
(2) the control performance of the reinforcement learning algorithm is improved, the problem that the existing reinforcement learning algorithm is poor in cross-well adaptability is solved, and the cross-well adaptability is good;
(3) the directional drilling device can assist a field engineer to perform sliding directional drilling operation, and improves the operation efficiency.
Although the present invention has been described above in connection with the exemplary embodiments and the accompanying drawings, it will be apparent to those of ordinary skill in the art that various modifications may be made to the above-described embodiments without departing from the spirit and scope of the claims.
Claims (9)
1. A sliding directional drilling method is characterized in that the sliding directional drilling method adopts the prediction results of the tool surface state at the future 3 moments to predict the control parameters of a torsional pendulum system at the future moment and controls the torsional pendulum system according to the prediction results of the control parameters of the torsional pendulum system at the future moment, wherein the control parameters of the torsional pendulum system comprise forward torque and reverse torque,
the prediction result of the control parameter of the torsion pendulum system at the future moment is obtained by adopting the following mode:
combining an actual tool surface, a target tool surface, forward torque and reverse torque at the current moment, giving out corresponding control parameters of forward torque and reverse torque by using a reinforcement learning algorithm, simultaneously giving out a tool surface angle at the next moment by using a tool surface state prediction method, giving out corresponding control parameters of forward torque and reverse torque by using the tool surface angle at the next moment and the reinforcement learning algorithm, and combining the current target tool surface with the forward torque and reverse torque of the corresponding control parameters as input of the reinforcement learning algorithm again to finish closed-loop feedback of data;
the prediction results of the tool face states at the future 3 moments are obtained by adopting a tool face state prediction method,
the prediction method comprises the following steps:
s1, acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well body data of the sliding directional drilling operation;
s2, selecting at least one type of data in the torsional pendulum system data logging system data and well data and measurement while drilling system data as data characteristics, and fusing according to time stamps to form a time sequence data set;
step S3, training LSTM neural network models one by using a time series data set to obtain a tool surface state prediction model based on LSTM;
and step S4, in the sliding directional drilling operation process, the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times are used as input data sequences, and the prediction results of the tool surface states at the future 3 times are obtained.
2. The method of sliding directional drilling according to claim 1, wherein the torsional pendulum system data comprises forward torque, reverse torque, forward rotational speed, reverse rotational speed, forward hold time, and reverse hold time; the measurement while drilling system data comprises a target tool face and an actual tool face; the logging system data includes weight on bit and displacement; the well bore data includes well depth, horizontal segment length, and well angle.
3. The method of sliding directional drilling according to claim 2, wherein the data of the torsional pendulum system is acquired at a frequency of 30 s/time to 1200 s/time; the data acquisition frequency of the measurement while drilling system is 20-180 s/time; the data acquisition frequency of the logging system is 4 s/time to 5 s/time.
4. The sliding directional drilling method according to claim 1, wherein step S2 includes: and performing data cleaning on the time-series data set to remove invalid data and redundant data.
5. The method of sliding directional drilling according to claim 4, wherein the invalid and redundant data comprises: data collected when working condition changes occur in the torsional pendulum orientation process, data collected when operation is wrong in the torsional pendulum orientation process, and data collected when the situation is inconsistent with basic rules or expectations occurs in the torsional pendulum orientation process.
6. The sliding directional drilling method according to claim 4, wherein step S2 further comprises a split-well and split-period process: and dividing the cleaned time series data set according to the time period of each directional drilling operation to form a plurality of subdata sets to be used as a training data basis of the LSTM neural network model.
7. The method of sliding directional drilling according to claim 6, wherein step S2 further comprises data preprocessing: the sub data sets are standardized to avoid interference between data features of different magnitudes.
8. The sliding directional drilling method of claim 1, wherein the specific method of training the LSTM neural network model one by one using the time series dataset is as follows: and building at least one layer of LSTM neural network model, splicing the fully-connected neural network at the rear end, training the LSTM neural network models one by using a data set, and simultaneously carrying out hyper-parameter adjustment, optimization and evaluation to obtain the tool face state prediction model based on the LSTM.
9. A sliding directional drilling system is characterized by comprising a data acquisition unit, a time sequence data set output unit, a neural network training unit, a tool face state prediction unit and a torsional pendulum system control unit,
the data acquisition unit is configured to be capable of acquiring torsional pendulum system data, measurement while drilling system data, logging system data and well bore data of the sliding directional drilling operation;
the time sequence data set output unit is connected with the data acquisition unit and is configured to be capable of selecting at least one type of data in torsional pendulum system data, logging system data and well data and measurement while drilling system data as data characteristics, and outputting a time sequence data set after fusion;
the neural network training unit is connected with the time sequence data set output unit and is configured to be capable of training the LSTM neural network model one by using the time sequence data set to obtain an LSTM-based tool face state prediction model;
the tool surface state prediction unit is connected with the neural network training unit and is configured to input the actual tool surface, the target tool surface, the forward torque and the reverse torque at the current time and the previous 4 times as input data sequences to the LSTM-based tool surface state prediction model and output the prediction results of the tool surface state at the future 3 times;
the torsional pendulum system control unit is connected with the tool face state prediction unit and is configured to be able to adjust the control parameters of the torsional pendulum system at a future moment according to the prediction results of the tool face state at the future 3 moments obtained by the sliding directional drilling method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111035640.2A CN113464120B (en) | 2021-09-06 | 2021-09-06 | Tool face state prediction method and system, and sliding directional drilling method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111035640.2A CN113464120B (en) | 2021-09-06 | 2021-09-06 | Tool face state prediction method and system, and sliding directional drilling method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113464120A CN113464120A (en) | 2021-10-01 |
| CN113464120B true CN113464120B (en) | 2021-12-03 |
Family
ID=77867486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111035640.2A Active CN113464120B (en) | 2021-09-06 | 2021-09-06 | Tool face state prediction method and system, and sliding directional drilling method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113464120B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020028302A1 (en) * | 2018-07-31 | 2020-02-06 | Shell Oil Company | Process for real time geological localization with reinforcement learning |
| WO2020028298A1 (en) * | 2018-07-31 | 2020-02-06 | Shell Oil Company | Method for geological steering control through reinforcement learning |
| KR102211421B1 (en) * | 2020-06-17 | 2021-02-02 | 에스케이건설 주식회사 | Method and system for determining tbm control parameters based on prediction geological condition ahead of tunnel face |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB9010096D0 (en) * | 1990-05-04 | 1990-06-27 | Baroid Technology Inc | Improvements in or relating to guiding of a tool along a subterranean path |
| US9297205B2 (en) * | 2011-12-22 | 2016-03-29 | Hunt Advanced Drilling Technologies, LLC | System and method for controlling a drilling path based on drift estimates |
| WO2013192524A2 (en) * | 2012-06-22 | 2013-12-27 | Hunt Advanced Drilling Technologies, L.L.C. | System an method for determining incremental progression between survey points while drilling |
| EP3665355A4 (en) * | 2017-08-10 | 2021-05-19 | Motive Drilling Technologies, Inc. | Apparatus and methods for automated slide drilling |
| US11136882B2 (en) * | 2017-09-21 | 2021-10-05 | Nabors Drilling Technologies Usa, Inc. | Automated drilling instructions for steerable drilling systems |
| US10731453B2 (en) * | 2018-01-16 | 2020-08-04 | Nabors Drilling Technologies Usa, Inc. | System and method of automating a slide drilling operation |
| US11143011B2 (en) * | 2018-04-26 | 2021-10-12 | Nabors Drilling Technologies Usa, Inc. | Real-time modification of a slide drilling segment based on continuous downhole data |
| CN109098659B (en) * | 2018-07-06 | 2021-01-01 | 中国石油集团川庆钻探工程有限公司 | Sliding drilling tool face adjusting method |
| WO2020191360A1 (en) * | 2019-03-21 | 2020-09-24 | Schlumberger Technology Corporation | Drilling system |
| US20200308952A1 (en) * | 2019-03-27 | 2020-10-01 | Nvicta LLC. | Method And System For Active Learning And Optimization Of Drilling Performance Metrics |
| WO2020236232A1 (en) * | 2019-05-21 | 2020-11-26 | Schlumberger Technology Corporation | Drilling control |
| CN110500034B (en) * | 2019-08-30 | 2021-03-19 | 中国石油集团川庆钻探工程有限公司 | Method for establishing neural network model, determining torsional pendulum drill string parameters and directionally drilling |
| CN111946258B (en) * | 2020-08-11 | 2022-08-02 | 中国石油天然气集团有限公司 | GRU-based sliding orientation intelligent control method |
| CN112360341B (en) * | 2020-09-27 | 2021-12-24 | 电子科技大学 | Machine learning-based sliding directional drilling tool face state identification method |
-
2021
- 2021-09-06 CN CN202111035640.2A patent/CN113464120B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020028302A1 (en) * | 2018-07-31 | 2020-02-06 | Shell Oil Company | Process for real time geological localization with reinforcement learning |
| WO2020028298A1 (en) * | 2018-07-31 | 2020-02-06 | Shell Oil Company | Method for geological steering control through reinforcement learning |
| KR102211421B1 (en) * | 2020-06-17 | 2021-02-02 | 에스케이건설 주식회사 | Method and system for determining tbm control parameters based on prediction geological condition ahead of tunnel face |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113464120A (en) | 2021-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2852765C (en) | Method and system for predicting a drill string stuck pipe event | |
| EP1984860B1 (en) | Methods, systems, and computer-readable media for fast updating of oil and gas field production models with physical and proxy simulators | |
| CN112083498B (en) | Multi-wave earthquake oil and gas reservoir prediction method based on deep neural network | |
| US11268370B2 (en) | Model-based parameter estimation for directional drilling in wellbore operations | |
| CN110807557A (en) | Drilling rate prediction method based on BP neural network and drilling rate optimization method based on BP neural network and particle swarm optimization | |
| NO347038B1 (en) | Automatic geosteering and evolutionary algorithm for use with same | |
| NO345487B1 (en) | Enhancing oilfield operations with cognitive computing | |
| Fan et al. | Distributed monitoring with integrated probability PCA and mRMR for drilling processes | |
| CN113464120B (en) | Tool face state prediction method and system, and sliding directional drilling method and system | |
| CN112069646A (en) | Method for accurately predicting mechanical drilling speed | |
| US20230116731A1 (en) | Intelligent time-stepping for numerical simulations | |
| CN111751878A (en) | Method and device for predicting transverse wave velocity | |
| US11555397B2 (en) | Detecting wellpath tortuosity variability and controlling wellbore operations | |
| Liu et al. | Safe model-based control from signal temporal logic specifications using recurrent neural networks | |
| Keller et al. | Rate of penetration estimation downhole with machine learning for drilling position control | |
| US11940584B2 (en) | Multi-sensor data assimilation and predictive analytics for optimizing well operations | |
| US20230142230A1 (en) | Systems and methods for modeling of dynamic waterflood well properties | |
| US20240044238A1 (en) | Optimal bottom hole assembly configuration | |
| Gao et al. | L2-SSA-LSTM prediction model of steering drilling wellbore trajectory | |
| Carpenter | Machine-Learning Model Developed for Rate-of-Penetration Optimization | |
| WO2023076330A1 (en) | Drill string stick/slip prediction and mitigation | |
| Ji et al. | A decision-theoretic active loop closing approach to autonomous robot exploration and mapping | |
| CN114519291A (en) | Method for establishing working condition monitoring and control model and application method and device thereof | |
| Sun et al. | Application of artificial neural network to the research of formation damage | |
| Zelinsky | Jason Chen |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |