Intelligent pressure control system and method based on gas cut pressure pulse

Abstract

The invention relates to an intelligent pressure control system and method based on a gas cut pressure pulse, and belongs to the technical field of oil and gas drilling, the method comprises the following steps: firstly, acquiring drilling wellhead pressure pulse detection data, pressure control drilling data and a safety pressure additional value, constructing an input vector, and inputting the input vector into a formation pressure prediction model, the formation pressure prediction model is obtained by training a long short-term memory network, adjacent well historical pressure control drilling wellhead pressure pulse detection data and adjacent well historical pressure control data; the predicted formation pressure under the current pressure control condition is obtained, and whether gas cut occurs or not is judged according to the relation between the predicted formation pressure and the bottom hole pressure; finally, if gas cut occurs, predicting the gas cut position according to wellhead pressure pulse detection data, and adjusting a pressure control scheme; based on the method, the formation pressure can be predicted in real time by detecting the real-time wellhead pressure pulse, and the problem that traditional gas cut recognition and formation pressure prediction are poor in precision and real-time performance can be solved.

Description

Technical Field

[0001] The invention relates to an intelligent pressure control system and method based on gas invasion pressure pulse, belonging to the technical field of oil and gas drilling. Background Art

[0002] Deep and ultra-deep drilling faces complex situations such as high formation temperature and pressure, the existence of multiple pressure system formations, and acid gas reservoirs. If the wellbore pressure is not properly controlled, complex situations such as overflow and gas invasion are very likely to occur. Therefore, accurately identifying gas invasion and precisely controlling wellbore pressure are key technologies to achieve safe drilling in ultra-deep formations.

[0003] Traditional gas invasion identification and formation pressure prediction methods are based on ground parameter monitoring, with poor prediction accuracy and real-time performance. The wellbore pressure control based on the formation pressure predicted by this method has problems such as low control accuracy, poor stability, and low reliability. In addition, since its control parameters are fixed, the back pressure value cannot be dynamically adjusted according to the complex flow state of the wellbore. It has poor robustness and there is a risk of formation leakage due to excessive back pressure. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention proposes an intelligent pressure control system and method based on gas invasion pressure pulses, which can solve the problems of poor accuracy and real-time performance of traditional gas invasion identification and formation pressure prediction.

[0005] The present invention adopts the following technical solutions: On the one hand, the present invention provides an intelligent pressure control system based on gas invasion pressure pulse, including a drilling module, a data measurement module and a pressure control module; The drilling module includes a mud pool, a mud tank, a gas-liquid separator, a vibrating screen, a hose, a wellhead spool, a well wall, an annular space, a drill string, a formation, a drill bit and a mud pump; The data measurement module includes a gas-liquid separator exhaust pipe flowmeter, a gas-liquid separator discharge pipe flowmeter, a mud pool level meter, a drilling fluid outlet flowmeter, a casing pressure sensor, a first flowmeter, a second flowmeter, a first pressure sensor, a second pressure sensor, a third pressure sensor, a fourth pressure sensor, a logging while drilling tool, a drilling fluid inlet flowmeter, a vertical pressure sensor, a computer and a database; The pressure control module includes well control choke manifold, wellhead back pressure compensation manifold, rotary blowout preventer, annular blowout preventer and well killing manifold; The lower end of the drill string is connected to the drill bit, and the upper end of the drill string is connected to the rotary blowout preventer and the water hose. The drill string forms an annular space with the well wall and the formation; the water hose is connected to the mud pump, and a vertical pressure sensor and a drilling fluid inlet flow meter are installed on the pipeline connecting the water hose and the mud pump. The vertical pressure sensor is used to obtain the riser pressure in the drill string, and the drilling fluid inlet flow meter is used to obtain the flow rate of the drilling fluid injected into the wellbore through the drill string. The mud pump is connected to the mud pool through the mud tank to provide drilling fluid to the wellbore during the drilling process.

[0006] Preferably, the wellhead spool is installed at the top of the wellbore wall, so that the annular space is connected with the wellhead back pressure compensation manifold, the well control throttle manifold, and the well killing manifold, and an annular blowout preventer is installed at the upper end of the wellbore wall; the wellhead back pressure compensation manifold and the well control throttle manifold are connected with the gas-liquid separator and the vibrating screen; The wellhead back pressure compensation manifold is installed with a drilling fluid outlet flowmeter, a first flowmeter, a second flowmeter, a first pressure sensor, a second pressure sensor, a third pressure sensor and a fourth pressure sensor; the drilling fluid outlet flowmeter is used to obtain the drilling fluid flow returned through the annular space; the wellhead back pressure compensation manifold and the vibrating screen connecting pipeline are installed with a drilling fluid outlet flowmeter, which is used to measure the drilling fluid flow returned through the annular space; the first flowmeter is installed in the pipeline connecting the wellhead back pressure compensation manifold and the well control throttle manifold, which is used to measure the drilling fluid flow from the wellhead back pressure compensation manifold to the well control throttle manifold passage; the second flowmeter is installed in the wellhead back pressure compensation manifold, which is used to measure the drilling fluid flow through the passage under the current pressure control drilling conditions; the fourth pressure sensor, the third pressure sensor, the second pressure sensor and the first pressure sensor are installed at the front end of each control valve of the wellhead back pressure compensation manifold, which are used to measure the pressure along the wellhead back pressure compensation manifold and the pressure before the control valve; The well control throttle manifold is equipped with a casing pressure sensor, which is used to obtain the casing pressure in the annular space; The gas-liquid separator is connected to the atmosphere on one hand, and a gas-liquid separator exhaust pipe flowmeter is installed on the pipeline connected to the atmosphere. On the other hand, it is connected to the mud pool, and a gas-liquid separator discharge pipe flowmeter is installed on the pipeline connected to the mud pool. The gas-liquid separator discharge pipe flowmeter and the gas-liquid separator exhaust pipe flowmeter are respectively used to obtain the liquid and gas phase flows separated after passing through the gas-liquid separator; the vibrating screen is connected to the mud pool, and a mud pool level meter is installed in the mud pool to measure the change of the mud pool liquid level; the logging while drilling tool is used to obtain downhole fluid measurement data, including but not limited to formation lithology, reservoir fluid properties, well depth, vertical depth, well inclination angle, drill bit position, and bottom hole pressure data; The computer is respectively connected with the gas-liquid separator exhaust pipe flowmeter, the gas-liquid separator discharge pipe flowmeter, the mud pool level meter, the drilling fluid outlet flowmeter, the casing pressure sensor, the first flowmeter, the second flowmeter, the first pressure sensor, the second pressure sensor, the third pressure sensor, the fourth pressure sensor, the logging while drilling tool, the drilling fluid inlet flowmeter, and the vertical pressure sensor, and the generated data is transmitted to the computer; the computer is connected with the well control throttle manifold, the wellhead back pressure compensation manifold, the rotary blowout preventer, the annular blowout preventer, and the well killing manifold for transmitting the pressure control parameters; the computer is connected with the database, and the database is used to store the drilling wellhead pressure pulse detection data, the pressure control drilling parameters and other information, read the historical wellhead pressure pulse detection data and the historical pressure control drilling data of the adjacent wells, etc.; the data measurement module also includes ground instruments.

[0007] On the other hand, the present invention provides an intelligent pressure control method based on gas intrusion pressure pulses, which is implemented by an intelligent pressure control system based on gas intrusion pressure pulses, and includes the following steps: S1: Acquire wellhead pressure pulse detection data, managed pressure drilling data and pressure control targets; S2: constructing input vectors based on wellhead pressure pulse detection data and managed pressure drilling data; S3: inputting the input vector into the formation pressure prediction model to obtain the predicted formation pressure under the current pressure control condition; S4: judging whether gas invasion occurs by predicting the relationship between formation pressure and bottom hole pressure. The bottom hole pressure is the sum of the wellbore hydrostatic pressure and the circulating pressure loss. When the predicted formation pressure is greater than the bottom hole pressure, it is judged that gas invasion occurs and S5 is executed; if not, step S3 is executed; S5: predict the gas invasion location based on the wellhead pressure pulse detection data and adjust the pressure control plan; S6: Acquire the wellhead pressure pulse detection data when adjusting the pressure control scheme and the pressure control drilling data after adjustment; S7: judging whether the pressure control target is met after the pressure control scheme is implemented through the wellhead pressure pulse detection data and the pressure control drilling data obtained in step S6, if so, executing step S8; if not, executing step S9; S8: Perform managed pressure drilling according to the updated managed pressure plan; S9: Update the input vector, correct the formation pressure prediction result, and return to step S5.

[0008] Preferably, in step S1, the drilling wellhead pressure pulse detection data is obtained by the casing pressure sensor and the vertical pressure sensor during the drilling process, including the pressure fluctuations caused by gas invasion or wellhead valve adjustment; The pressure-controlled drilling data includes drilling parameters and pressure-controlled parameters. The drilling parameters include drilling time, well depth, vertical depth, well inclination, drill bit position, mud pool liquid level, drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, drilling fluid outlet temperature, downhole fixed-point pressure, circulating pressure consumption, drill bit pressure drop, wellbore hydrostatic pressure, fixed-point circulating equivalent density, drill bit circulating equivalent density, drilling fluid type, drilling fluid rheology, drilling fluid inlet flow rate and drilling fluid outlet flow rate under the current drilling conditions; the pressure-controlled parameters include pump pressure, first flowmeter, second flowmeter, first pressure sensor, second pressure sensor, third pressure sensor and fourth pressure sensor; among them, drilling time, drilling fluid The type and rheological properties of drilling fluid are obtained from the drilling log; the well depth, vertical depth, well inclination angle, drill bit position, downhole fixed-point pressure, fixed-point circulating equivalent density, and drill bit circulating equivalent density are obtained from the logging while drilling tool; the drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, and drilling fluid outlet temperature are obtained from the surface instrument; the circulating pressure loss, drill bit pressure drop, and wellbore hydrostatic pressure are calculated from the wellbore multiphase flow calculation equation; the mud pool liquid level is obtained from the mud pool level gauge, the drilling fluid inlet flow rate is obtained from the drilling fluid inlet flow meter, and the drilling fluid outlet flow rate is obtained from the drilling fluid outlet flow meter; the pressure control parameter neutral pressure is obtained from the vertical pressure sensor, the casing pressure is obtained from the casing pressure sensor, and the pump pressure is obtained from the mud pump; The pressure control target is that the wellbore pressure of the managed pressure drilling is greater than or equal to the sum of the formation pressure and the safety additional pressure. The wellbore pressure of the managed pressure drilling can be calculated by the wellbore multiphase flow calculation equation, and the safety additional pressure is set by the oilfield company, industry standards or drilling team; In step S2, the input vector is a vector including drilling time, wellhead pressure pulse detection data, and pressure-controlled drilling data. n dimensional vector, n is the number of data, such as: ; In the formula, is the input vector, is the drilling time, It is the drilling wellhead pressure pulse detection data. This is the managed pressure drilling data.

[0009] Preferably, the formation pressure prediction model in step S3 adopts a long short-term memory network, and uses historical pressure-controlled drilling data of adjacent wells, historical wellhead pressure pulse detection data of adjacent wells, and formation pressure data to train the final formation pressure prediction model; The process of establishing the formation pressure prediction model includes: S31: Acquire historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data, wherein the historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data are obtained from the adjacent well drilling log, well history, geological logging and other data or calculated from the drilling parameters during the adjacent well drilling process; S32: pre-process the data obtained in S31 to construct sample data for training the model; S33: Building long short-term memory networks; S34: Construct a model evaluation index based on the long short-term memory network of the sample data set, and solve the parameters in the long short-term memory network based on the evaluation index.

[0010] Preferably, the preprocessing process of step S32 is: S321: Analyze and select the variation rules between the wellhead pressure pulse detection data and the formation pressure data under different pressure-controlled drilling data in the historical drilling of the adjacent well; S322: Deleting abnormal values ​​of historical wellhead pressure pulse detection data of adjacent wells, historical pressure-controlled drilling data of adjacent wells, and formation pressure data through a preset abnormal data judgment method; S323: expanding the missing values ​​of the historical wellhead pressure pulse detection data of the adjacent wells, the historical pressure-controlled drilling data of the adjacent wells, and the formation pressure data by setting a denoising diffusion probability model; S324: Calculate the variation rules of various parameters in the historical controlled pressure drilling data of the adjacent wells and the formation pressure data, calculate the correlation, and select the parameters with strong correlation with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells to reduce redundant features and noise and improve the generalization ability of the model; S325: normalizing the historical wellhead pressure pulse detection data of the adjacent well, the historical pressure-controlled drilling data of the adjacent well, and the formation pressure data; S326: splicing the processed historical wellhead pressure pulse detection data of the adjacent well, the historical pressure-controlled drilling data of the adjacent well and the formation pressure data to construct model sample data; In step S323, the denoising diffusion probability model includes a forward diffusion process and a reverse generation process. The forward diffusion process simulates the "degradation" process of data by gradually adding noise to real data; the reverse generation process gradually restores the noise to data through a reverse step-by-step denoising process. In step S324, the correlation is calculated using the following formula: ; in, For variables and formation pressure The distance correlation coefficient between The parameters of the historical drilling wellhead pressure pulse detection data and the historical pressure-controlled drilling data of the adjacent well; , is the standard deviation of the parameter’s own distance; For variables and formation pressure The distance covariance between them is obtained; the parameters with a correlation coefficient greater than 0.5 with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells are selected as input parameters; In step S325, the normalization process uses max-min normalization to map each parameter feature variable to between (0, 1); In step S326, the constructed sample data are arranged by drilling time, including historical pressure-controlled drilling data of nearby wells, historical wellhead pressure pulse detection data of nearby wells, and vectors of formation pressure. Multiple groups of sample data are constructed based on the pressure-controlled drilling data, wellhead pressure pulse detection data, and formation pressure data of different drilling conditions, different wells, and different periods.

[0011] Preferably, in step S33, the long short-term memory network includes long short-term memory network neurons, each of which includes a cell state, a forget gate, an input gate, and an output gate, wherein the cell state is responsible for information transmission between multiple time steps of the grid, and its state formula is defined as: ; In the formula, is the current cell state; is the cell state at the previous time step; Output of the forget gate; is the input gate output; Candidate memory for the current time step; The long short-term memory network model solves the shortcomings of the standard recurrent neural network model in dealing with long-term problems by introducing a gating mechanism. The forget gate controls the amount of data forgotten by the model, the input gate controls the amount of information input to the current cell state, and the output gate controls the amount of data output to the next time step of the cell state. The control equation is: ; in, , , They are the weight matrices of the forget gate, cell state, and output gate respectively; is the activation function; , , , They are the bias items of forget gate, input gate, cell state and output gate respectively; is the output gate activation value; is the hidden state of the current time step; Input for the current time step; In step S34, the evaluation index is the average relative error MRE and mean square error RMSLE : ; ; in, is the number of samples; , are the formation pressure value solved by the model and the real formation pressure value respectively; The model output is calculated by forward propagation, including the input gate , Forget Gate , output gate and the candidate memory for the current time step Data, combined with the input gate and forget gate to update the current cell state , calculate the hidden state of the current time step based on the output gate and cell state ; Calculate the loss between the model output and the true value according to the evaluation index; Calculate the gradient and update the parameters through back propagation, including calculating the evaluation criteria , , , , , , The gradient of the parameters is back-propagated to each unit according to the gradient to update the input gate , Forget Gate , output gate and the candidate memory for the current time step Parameters; the long short-term memory grid model is continuously optimized through multiple forward propagation, calculation of evaluation indicators, and back propagation iterations.

[0012] Preferably, in step S5, if gas invasion does not occur, the wellbore contains a single-phase drilling fluid, and the pressure wave propagation speed is: ; in, is the pressure wave propagation speed, , are the elastic modulus of drilling fluid and pipe wall, respectively. is the drilling fluid density, is the inner diameter of the tube wall, is the tube wall thickness, then: ; in, is the corresponding time of pressure wave propagation in the wellbore without gas invasion, n is the number of grid discretes along the wellbore depth, For the The grid length of the grid, is the propagation speed of the grid pressure wave; If gas invasion occurs, the gas invasion position in the wellbore is a two-phase fluid of invading gas and drilling fluid, and the pressure wave propagation speed is: ; in, is the propagation velocity of the wellbore pressure wave under invasion state, is the elastic modulus of the intruding gas, is the volume fraction of intruding gas, is the density of the mixed fluid of invading gas and drilling fluid; ; in, The corresponding time of pressure wave propagation in the wellbore caused by gas invasion; according to the return time difference of the pressure wave , calculate its gas invasion position; According to the predicted formation pressure and gas invasion position, the wellhead back pressure value is adjusted by adjusting the valve opening in the well control throttling manifold and the wellhead back pressure compensation manifold, so that the pressure controlled drilling wellbore pressure is greater than or equal to the sum of the predicted formation pressure obtained by S3 and the safety additional value. Among them, the wellhead back pressure value can be read by the vertical pressure sensor.

[0013] Preferably, in step S7, the pressure wave return time difference is calculated through the wellhead pressure pulse detection data and the pressure controlled drilling data obtained in step S6. If the pressure wave return time difference is reduced, the readings of the drilling fluid outlet flowmeter and the drilling fluid inlet flowmeter are the same, and the pressure controlled drilling wellbore pressure is greater than or equal to the sum of the predicted formation pressure and the safety additional pressure, then the pressure control target is met.

[0014] Preferably, in step S9, the input vector is updated to correct the formation pressure prediction result, and the specific process is as follows: S91: Acquire the current real-time drilling dynamic data, which includes the current gas invasion pressure pulse detection data and the pressure-controlled drilling data; S92: constructing a formation pressure prediction model as a real-time drilling dynamic sub-model using current real-time drilling dynamic data; S93: Freeze some layers of the historical formation pressure prediction model, retain its weights, fine-tune it using current real-time drilling dynamic data, and update the remaining weights; S94: weighting and fusing the output results of the real-time drilling dynamic sub-model and the historical formation pressure prediction model to obtain an output result that fuses the historical formation pressure prediction model information and the real-time drilling dynamic sub-model information; S95: Perform error-weighted integration on the current real-time drilling dynamic sub-model and the historical formation pressure prediction model to predict the formation pressure and gas invasion pressure pulse change law at the next moment.

[0015] Further preferably, in step S93, the historical formation pressure prediction model is trained by drilling historical static data, and the historical static data includes drilling historical gas invasion pressure pulse detection data and historical pressure-controlled drilling data; the input parameters, input gates, and forget gates in the historical formation pressure prediction model are frozen, and their weights are retained. The historical formation pressure prediction model is unchanged, wherein the historical formation pressure prediction model is the formation pressure prediction model established and used before step S7; the current gas invasion pressure pulse detection data and the pressure-controlled drilling data are used to fine-tune it to adapt to the current data changes, and the fine-tuning objectives are: ; in, Output of the fine-tuned model; is the true value of formation pressure; The updated model weights are: ; in, The weights are updated by the gradient descent algorithm; In step S94, the weighted fusion method is: ; In the formula, , The historical formation pressure prediction model is determined by the model evaluation index. and real-time drilling dynamics sub-model Weight; , are the outputs of the historical formation pressure prediction model and sub-model, respectively; Further preferably, in step S95, if the errors of the historical formation pressure prediction model and the real-time drilling dynamic sub-model at the current moment are respectively , , then the error weighted integration method is: .

[0016] For any details not provided in the present invention, please refer to the prior art.

[0017] The beneficial effects of the present invention are: The intelligent pressure control system and method based on gas invasion pressure pulse of the present invention can solve the problems of poor accuracy and real-time performance of traditional gas invasion identification and formation pressure prediction. Specifically, firstly, the wellhead pressure pulse detection data, pressure control drilling data and safety pressure added value are obtained, an input vector is constructed, and input into the formation pressure prediction model, wherein the formation pressure prediction model is trained by using long short-term memory network, historical pressure control drilling wellhead pressure pulse detection data of adjacent wells and historical pressure control data of adjacent wells. The predicted formation pressure under the current pressure control condition is obtained, and whether gas invasion occurs is determined by predicting the relationship between formation pressure and bottom hole pressure. Finally, if gas invasion occurs, the gas invasion position is predicted according to the wellhead pressure pulse detection data and the pressure control scheme is adjusted. Based on the above method, the real-time prediction of formation pressure can be realized by detecting the real-time wellhead pressure pulse; the formation pressure prediction model greatly reduces the formation pressure prediction error by processing and learning the historical wellhead pressure pulse detection data of adjacent wells and the historical pressure control drilling data of adjacent wells, making it more accurate. The real-time wellbore pressure control method based on gas invasion pressure pulse can dynamically adjust the back pressure value according to the wellbore flow state, accurately match the formation pressure, and ensure drilling safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0019] Figure 1 It is a structural schematic diagram of an intelligent pressure control system based on gas intrusion pressure pulse in a certain embodiment of the present invention; Figure 2 This is a flow chart of an intelligent pressure control method based on gas intrusion pressure pulses in a certain embodiment of the present invention; Figure 3 It is a flow chart of the process of establishing a formation pressure prediction model in a certain embodiment of the present invention; Figure 4 A sample data flow chart for constructing a training model in step S32 in an embodiment of the present invention; Figure 5 This is a flow chart of correcting the formation pressure prediction result in a certain embodiment of the present invention; Figure 6 A schematic diagram of the structure of a denoising diffusion probability model in an embodiment of the present invention; Figure 7 A schematic diagram of the structure of a long short-term memory network in a certain embodiment of the present invention; In the figure, 1, mud pool; 2, mud tank; 3, gas-liquid separator; 4, gas-liquid separator exhaust pipe flowmeter; 5, gas-liquid separator discharge pipe flowmeter; 6, mud pool level gauge; 7, vibrating screen; 8, drilling fluid outlet flowmeter; 9, well control throttling manifold; 10, casing pressure sensor; 11, wellhead back pressure compensation manifold; 12, first flowmeter; 13, second flowmeter; 14, first pressure sensor; 15, second pressure sensor; 16. Third pressure sensor; 17. Fourth pressure sensor; 18. Water hose; 19. Rotary blowout preventer; 20. Annular blowout preventer; 21. Wellhead spool; 22. Wellbore wall; 23. Annular space; 24. Drill string; 25. Formation; 26. LWD tool; 27. Drill bit; 28. Well kill manifold; 29. ​​Mud pump; 30. Drilling fluid inlet flowmeter; 31. Vertical pressure sensor; 32. Computer; 33. Database. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the drawings in the implementation of this specification, but are not limited to this. Anything not fully described in the present invention shall be based on the conventional technology in the art.

[0021] Example 1 An intelligent pressure control system based on gas intrusion pressure pulses, such as Figure 1 As shown, it includes a drilling module, a data measurement module and a pressure control module; The drilling module includes a mud pool 1, a mud tank 2, a gas-liquid separator 3, a vibrating screen 7, a hose 18, a wellhead spool 21, a well wall 22, an annular space 23, a drill string 24, a formation 25, a drill bit 27 and a mud pump 29; The data measurement module includes a gas-liquid separator exhaust pipe flowmeter 4, a gas-liquid separator discharge pipe flowmeter 5, a mud pool level meter 6, a drilling fluid outlet flowmeter 8, a casing pressure sensor 10, a first flowmeter 12, a second flowmeter 13, a first pressure sensor 14, a second pressure sensor 15, a third pressure sensor 16, a fourth pressure sensor 17, a logging while drilling tool 26, a drilling fluid inlet flowmeter 30, a vertical pressure sensor 31, a computer 32 and a database 33; The pressure control module includes a well control throttle manifold 9, a wellhead back pressure compensation manifold 11, a rotary blowout preventer 19, an annular blowout preventer 20 and a well killing manifold 28; The lower end of the drill string 24 is connected to the drill bit 27, and the upper end of the drill string 24 is connected to the rotary blowout preventer 19 and the water hose 18. The drill string 24 forms an annular space 23 with the well wall 22 and the formation 25; the water hose 18 is connected to the mud pump 29, and a vertical pressure sensor 31 and a drilling fluid inlet flowmeter 30 are installed on the pipeline connecting the water hose 18 and the mud pump 29. The vertical pressure sensor 31 is used to obtain the vertical pipe pressure in the drill string 24, and the drilling fluid inlet flowmeter 30 is used to obtain the flow rate of the drilling fluid injected into the wellbore through the drill string 24. The mud pump 29 is connected to the mud pool 1 through the mud tank 2 to provide drilling fluid to the wellbore during the drilling process; The wellhead spool 21 is installed on the top of the well wall 22, so that the annular space 23 is connected with the wellhead back pressure compensation manifold 11, the well control throttle manifold 9, and the well killing manifold 28. An annular blowout preventer 20 is installed on the upper end of the well wall; the wellhead back pressure compensation manifold 11 and the well control throttle manifold 9 are connected with the gas-liquid separator 3 and the vibrating screen 7; The wellhead back pressure compensation manifold 11 is installed with a drilling fluid outlet flowmeter 8, a first flowmeter 12, a second flowmeter 13, a first pressure sensor 14, a second pressure sensor 15, a third pressure sensor 16 and a fourth pressure sensor 17; the drilling fluid outlet flowmeter 8 is used to obtain the drilling fluid flow returned through the annular space 23; the wellhead back pressure compensation manifold 11 is connected to the vibrating screen 7 by a pipeline to install a drilling fluid outlet flowmeter 8, which is used to measure the drilling fluid flow returned through the annular space 23; the first flowmeter 12 is installed on the wellhead back pressure compensation manifold 11 is a pipeline connected to the well control throttling manifold 9, which is used to measure the drilling fluid flow rate from the wellhead back pressure compensation manifold 11 to the well control throttling manifold 9; the second flow meter 13 is installed on the wellhead back pressure compensation manifold, which is used to measure the drilling fluid flow rate through the passage under the current pressure controlled drilling conditions; the fourth pressure sensor 17, the third pressure sensor 16, the second pressure sensor 15, and the first pressure sensor 14 are installed at the front end of each control valve of the wellhead back pressure compensation manifold 11, which are used to measure the pressure along the wellhead back pressure compensation manifold 11 and the pressure before the control valve.

[0022] The well control throttle manifold 9 is equipped with a casing pressure sensor 10, which is used to obtain the casing pressure in the annular space 23; The gas-liquid separator 3 is connected to the atmosphere on one hand, and a gas-liquid separator exhaust pipe flowmeter 4 is installed on the pipeline connected to the atmosphere. On the other hand, it is connected to the mud pool 1, and a gas-liquid separator discharge pipe flowmeter 5 is installed on the pipeline connected to the mud pool. The gas-liquid separator discharge pipe flowmeter 5 and the gas-liquid separator exhaust pipe flowmeter 4 are respectively used to obtain the liquid and gas phase flows separated after passing through the gas-liquid separator 3; the vibrating screen 7 is connected to the mud pool 1, and a mud pool level meter 6 is installed in the mud pool 1 to measure the change of the mud pool liquid level; the logging while drilling tool 26 is used to obtain downhole fluid measurement data, including but not limited to formation lithology, reservoir fluid properties, well depth, vertical depth, well inclination, drill bit position, bottom hole pressure and other data; The computer 32 is respectively connected with the gas-liquid separator exhaust pipe flowmeter 4, the gas-liquid separator discharge pipe flowmeter 5, the mud pool level meter 6, the drilling fluid outlet flowmeter 8, the casing pressure sensor 10, the first flowmeter 12, the second flowmeter 13, the first pressure sensor 14, the second pressure sensor 15, the third pressure sensor 16, the fourth pressure sensor 17, the logging while drilling tool 26, the drilling fluid inlet flowmeter 30, and the vertical pressure sensor 31, and the generated data is transmitted to the computer; the computer 32 is connected with the well control throttle manifold 9, the wellhead back pressure compensation manifold 11, the rotary blowout preventer 19, the annular blowout preventer 20, and the well killing manifold 28 for transmitting the pressure control parameters; the computer 32 is connected with the database 33, and the database 33 is used to store the drilling wellhead pressure pulse detection data, the pressure control drilling parameters and other information, read the historical wellhead pressure pulse detection data and the historical pressure control drilling data of the adjacent wells, etc.

[0023] The data measurement module also includes ground instruments.

[0024] Example 2 An intelligent pressure control method based on gas intrusion pressure pulse is implemented by the intelligent pressure control system based on gas intrusion pressure pulse of embodiment 1, such as Figure 2 As shown, the following steps are included: S1: Acquire wellhead pressure pulse detection data, managed pressure drilling data and pressure control targets; S2: constructing input vectors based on wellhead pressure pulse detection data and managed pressure drilling data; S3: inputting the input vector into the formation pressure prediction model to obtain the predicted formation pressure under the current pressure control condition; S4: judging whether gas invasion occurs by predicting the relationship between formation pressure and bottom hole pressure. The bottom hole pressure is the sum of the wellbore hydrostatic pressure and the circulating pressure loss. When the predicted formation pressure is greater than the bottom hole pressure, it is judged that gas invasion occurs and S5 is executed; if not, step S3 is executed; S5: predict the gas invasion location based on the wellhead pressure pulse detection data and adjust the pressure control plan; S6: Acquire the wellhead pressure pulse detection data when adjusting the pressure control scheme and the pressure control drilling data after adjustment; S7: judging whether the pressure control target is met after the pressure control scheme is implemented through the wellhead pressure pulse detection data and the pressure control drilling data obtained in step S6, if so, executing step S8; if not, executing step S9; S8: Perform managed pressure drilling according to the updated managed pressure plan; S9: Update the input vector, correct the formation pressure prediction result, and return to step S5.

[0025] Example 3 An intelligent pressure control method based on gas invasion pressure pulse, as described in Example 2, except that in step S1, the drilling wellhead pressure pulse detection data is obtained by the casing pressure sensor 10 and the vertical pressure sensor 31 during the drilling process, including the pressure fluctuations caused by gas invasion or wellhead valve adjustment; The pressure-controlled drilling data includes drilling parameters and pressure-controlled parameters. The drilling parameters include drilling time, well depth, vertical depth, well inclination, drill bit position, mud pool liquid level, drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, drilling fluid outlet temperature, downhole fixed-point pressure, circulating pressure consumption, drill bit pressure drop, wellbore hydrostatic pressure, fixed-point circulating equivalent density, drill bit circulating equivalent density, drilling fluid type, drilling fluid rheology, drilling fluid inlet flow rate and drilling fluid outlet flow rate under the current drilling conditions. The pressure-controlled parameters include pump pressure, a first flowmeter 12, a second flowmeter 13, a first pressure sensor 14, a second pressure sensor 15, a third pressure sensor 16 and a fourth pressure sensor 17. Among them, the drilling time, the drilling fluid type and the drilling fluid rheology are determined by the drill. The well log is used to obtain the well depth, vertical depth, well inclination angle, drill bit position, downhole fixed-point pressure, fixed-point circulating equivalent density, and drill bit circulating equivalent density. The drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, and drilling fluid outlet temperature are obtained from the surface instrument; the circulating pressure loss, drill bit pressure drop, and wellbore hydrostatic pressure are calculated by the wellbore multiphase flow calculation equation, which is the existing technology and can be referred to; the mud pool liquid level is obtained by the mud pool level meter 6, the drilling fluid inlet flow rate is obtained by the drilling fluid inlet flow meter 30, and the drilling fluid outlet flow rate is obtained by the drilling fluid outlet flow meter 8; the pressure control parameter neutral pressure is obtained by the vertical pressure sensor 31, the casing pressure is obtained by the casing pressure sensor 10, and the pump pressure is obtained by the mud pump 29; The pressure control target is that the wellbore pressure of pressure-controlled drilling is greater than or equal to the sum of the formation pressure (the formation pressure can be obtained by referring to the well logging data, exploration well data, and geological data) and the safety additional pressure, where the wellbore pressure of pressure-controlled drilling can be calculated by the wellbore multiphase flow calculation equation, and the safety additional pressure is set by the oilfield company, industry standards or drilling team.

[0026] In step S2, the input vector is a vector including drilling time, wellhead pressure pulse detection data, and pressure-controlled drilling data. n dimensional vector, n is the number of data, such as: ; In the formula, is the input vector, is the drilling time, which is included in the managed pressure drilling data. It is the drilling wellhead pressure pulse detection data. This is the managed pressure drilling data.

[0027] Example 4 An intelligent pressure control method based on gas invasion pressure pulses, as described in Example 3, except that the formation pressure prediction model in step S3 adopts a long short-term memory network, and uses the historical pressure control drilling data of adjacent wells, the historical wellhead pressure pulse detection data of adjacent wells, and the formation pressure data to train the final formation pressure prediction model.

[0028] The process of establishing the formation pressure prediction model is as follows: Figure 3 As shown, including: S31: Acquire historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data, wherein the historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data can be obtained from the adjacent well drilling log, well history, geological logging and other data or calculated from the drilling parameters during the adjacent well drilling process; S32: pre-process the data obtained in S31 to construct sample data for training the model; S33: Building long short-term memory networks; S34: Construct a model evaluation index based on the long short-term memory network of the sample data set, and solve the parameters in the long short-term memory network based on the evaluation index.

[0029] Example 5 An intelligent pressure control method based on gas intrusion pressure pulses is as described in Example 4, except that the preprocessing process of step S32 is: S321: Analyze and select the change rules between the wellhead pressure pulse detection data and the formation pressure data under different pressure-controlled drilling data in the historical drilling of the adjacent wells. Through the preliminary classification of the historical drilling data of the adjacent wells, the data with similar drilling conditions are classified and trained, which can significantly reduce the training difficulty and improve the calculation accuracy. S322: Processing abnormal values ​​of historical wellhead pressure pulse detection data of adjacent wells, historical pressure-controlled drilling data of adjacent wells, and formation pressure data through a preset abnormal data judgment method; By presetting the abnormal data judgment method, for example, when the historical pressure-controlled drilling data of the adjacent well, the historical wellhead pressure pulse detection data and the formation pressure data of the adjacent well are 0 data, NaN data, etc., such abnormal data are deleted.

[0030] S323: expanding the missing values ​​of the historical wellhead pressure pulse detection data of the adjacent wells, the historical pressure-controlled drilling data of the adjacent wells, and the formation pressure data by setting a denoising diffusion probability model; S324: Calculate the variation rules of various parameters in the historical controlled pressure drilling data of the adjacent wells and the formation pressure data, calculate the correlation, and select the parameters with strong correlation with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells to reduce redundant features and noise and improve the generalization ability of the model; S325: normalizing the historical wellhead pressure pulse detection data of the adjacent well, the historical pressure-controlled drilling data of the adjacent well, and the formation pressure data; S326: The processed historical wellhead pressure pulse detection data, historical pressure-controlled drilling data and formation pressure data of the adjacent wells are spliced ​​to construct model sample data, such as Figure 4 shown.

[0031] In step S323, if Figure 6 As shown in Figure 1, the denoising diffusion probability model includes a forward diffusion process and a reverse generation process. The forward diffusion process simulates the “degradation” process of data by gradually adding noise to the real data. The process is defined as: ; in, For the A sample of steps, For the Step sample; is the noise intensity at each time step; represents the Gaussian distribution of noise in the diffusion process; is the unit matrix, which indicates that the noise in each unit is independent; The reverse generation process gradually restores the noise to data through the reverse step-by-step denoising process. The process is defined as: ; in, Given the current noise sample In the case of Distribution of is the mean of the model output; is the standard deviation; The cross-attention mechanism in the reverse generation process of the denoising diffusion probability model can effectively introduce conditional information, such as pressure-controlled drilling parameters, into the reverse generation process, so that the generation of missing values ​​in the reverse generation process is guided by the conditional information. The cross-attention weight can be defined as: ; in, is the weighted output, indicating the weighted information of the target data; Q is the query value from the condition information; K is the key from the target data; is the value of the target data; is the scaling factor.

[0032] Preferably, in step S324, the correlation is calculated using the following formula: ; in, For variables and formation pressure The distance correlation coefficient between The parameters of the historical drilling wellhead pressure pulse detection data and the historical pressure-controlled drilling data of the adjacent well; , is the standard deviation of the parameter’s own distance; For variables and formation pressure The distance covariance between them is obtained; the parameters with a correlation coefficient greater than 0.5 with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells are selected as input parameters; In step S325, the normalization process may adopt max-min normalization to map each parameter feature variable to between (0, 1): ; in, is the normalized data; , is the minimum and maximum value of a feature.

[0033] In step S326, the constructed sample data is arranged by drilling time, including historical pressure-controlled drilling data of adjacent wells, historical wellhead pressure pulse detection data of adjacent wells, and vectors of formation pressure: ; In the formula, is the sample data, It is the formation pressure; multiple sets of sample data can be constructed according to the pressure-controlled drilling data, wellhead pressure pulse detection data and formation pressure data of different drilling conditions, different wells and different periods.

[0034] Example 6 An intelligent pressure control method based on gas intrusion pressure pulse, as described in Example 5, except that in step S33, Figure 7 As shown, the LSTM network includes LSTM neurons. Each LSTM neuron contains a cell state, a forget gate, an input gate, and an output gate. The cell state is responsible for information transmission between multiple time steps of the grid, and its state formula is defined as: ; In the formula, is the current cell state; is the cell state at the previous time step; Output of the forget gate; is the input gate output; Candidate memory for the current time step; The long short-term memory network model solves the shortcomings of the standard recurrent neural network model in dealing with long-term problems by introducing a gating mechanism. The forget gate controls the amount of data forgotten by the model, the input gate controls the amount of information input to the current cell state, and the output gate controls the amount of data output to the next time step of the cell state. The control equation is: ; in, , , They are the weight matrices of the forget gate, cell state, and output gate respectively; is the activation function; , , , They are the bias items of forget gate, input gate, cell state and output gate respectively; is the output gate activation value; is the hidden state of the current time step; Input for the current time step.

[0035] Example 7 An intelligent pressure control method based on gas intrusion pressure pulse, as described in Example 6, except that in step S34, the evaluation index adopts the average relative error MRE and mean square error RMSLE : ; ; in, is the number of samples; , are the formation pressure value solved by the model and the real formation pressure value respectively; The model output is calculated by forward propagation, including the input gate , Forget Gate , output gate and the candidate memory for the current time step Data, combined with the input gate and forget gate to update the current cell state , calculate the hidden state of the current time step based on the output gate and cell state ; Calculate the loss between the model output and the true value according to the evaluation index; Calculate the gradient and update the parameters through back propagation, including calculating the evaluation criteria , , , , , , The gradient of the same parameter is back-propagated to each unit according to the gradient, and the input gate is updated , Forget Gate , output gate and the candidate memory for the current time step Parameters; the long short-term memory grid model is continuously optimized through multiple forward propagation, calculation of evaluation indicators, and back propagation iterations.

[0036] Example 8 An intelligent pressure control method based on gas invasion pressure pulse, as described in Example 7, except that in step S5, if gas invasion does not occur, the wellbore is a single-phase drilling fluid, and the pressure wave propagation speed is: ; in, is the pressure wave propagation speed, , are the elastic modulus of drilling fluid and pipe wall, respectively. is the drilling fluid density, is the inner diameter of the tube wall, is the tube wall thickness, then: ; in, is the corresponding time of pressure wave propagation in the wellbore without gas invasion, n is the number of grid discretes along the wellbore depth, For the The grid length of the grid, For the Grid pressure wave propagation speed; If gas invasion occurs, the gas invasion position in the wellbore is a two-phase fluid of invading gas and drilling fluid, and the pressure wave propagation speed is: ; in, is the propagation velocity of the wellbore pressure wave under invasion state, is the elastic modulus of the intruding gas, is the volume fraction of intruding gas, is the density of the mixed fluid of invading gas and drilling fluid; ; in, The corresponding time of pressure wave propagation in the wellbore caused by gas invasion; according to the return time difference of the pressure wave , calculate its gas invasion position; According to the predicted formation pressure and gas invasion position, the wellhead back pressure value is adjusted by adjusting the valve opening in the well control throttling manifold 9 and the wellhead back pressure compensation manifold 11, so that the pressure controlled drilling wellbore pressure is greater than or equal to the sum of the predicted formation pressure obtained by S3 and the safety additional value, wherein the wellhead back pressure value can be read by the vertical pressure sensor.

[0037] Example 9 An intelligent pressure control method based on gas invasion pressure pulses, as described in Example 8, except that, in step S7, the pressure wave return time difference is calculated through the wellhead pressure pulse detection data and the pressure controlled drilling data obtained in step S6. If the pressure wave return time difference is reduced, the readings of the drilling fluid outlet flowmeter and the drilling fluid inlet flowmeter are the same, and the pressure of the pressure controlled drilling wellbore is greater than or equal to the sum of the predicted formation pressure and the safety additional pressure, the pressure control target is met.

[0038] Example 10 An intelligent pressure control method based on gas invasion pressure pulse, as described in Example 9, except that in step S9, the input vector is updated and the formation pressure prediction result is corrected, such as Figure 5 As shown, the specific process is: S91: Acquire the current real-time drilling dynamic data, which includes the current gas invasion pressure pulse detection data and the pressure-controlled drilling data; S92: constructing a formation pressure prediction model as a real-time drilling dynamic sub-model using current real-time drilling dynamic data; S93: Freeze some layers of the historical formation pressure prediction model, retain its weights, fine-tune it using current real-time drilling dynamic data, and update the remaining weights; S94: weighting and fusing the output results of the real-time drilling dynamic sub-model and the historical formation pressure prediction model to obtain an output result that fuses the historical formation pressure prediction model information and the real-time drilling dynamic sub-model information; S95: Perform error-weighted integration on the current real-time drilling dynamic sub-model and the historical formation pressure prediction model to predict the formation pressure and gas invasion pressure pulse change law at the next moment.

[0039] In step S92, the method of constructing a formation pressure prediction model using the gas invasion pressure pulse detection data and the pressure-controlled drilling data at the current moment is the same as the training process of the formation pressure prediction model (S31-S34), but the data used are the gas invasion pressure pulse detection data and the pressure-controlled drilling data at the current moment (the same as the above S31-S34 method, except that the input data is changed to the current moment); In step S93, the historical formation pressure prediction model is trained by drilling historical static data, and the historical static data includes drilling historical gas invasion pressure pulse detection data and historical pressure control drilling data. Freeze the input parameters, input gates, forget gates and other feature extraction layers in the historical formation pressure prediction model, and keep their weights unchanged. The historical formation pressure prediction model is the formation pressure prediction model established and used before step S7. Use the current gas invasion pressure pulse detection data and pressure control drilling data to fine-tune it to adapt to the current data changes. The fine-tuning target is: ; in, Output of the fine-tuned model; is the true value of formation pressure.

[0040] The updated model weights are: ; in, is the weight updated by the gradient descent algorithm.

[0041] Preferably, in step S94, the weighted fusion method is: ; In the formula, , The historical formation pressure prediction model is determined by the model evaluation index. and real-time drilling dynamics sub-model Weight; , The historical formation pressure prediction model is and real-time drilling dynamics sub-model Output.

[0042] In step S95, if the errors of the historical formation pressure prediction model and the real-time drilling dynamic sub-model at the current moment are , , then the error weighted integration method is: .

[0043] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An intelligent pressure control system based on gas intrusion pressure pulse, characterized in that: It includes drilling module, data measurement module and pressure control module; The drilling module includes a mud pool, a mud tank, a gas-liquid separator, a vibrating screen, a hose, a wellhead spool, a well wall, an annular space, a drill string, a formation, a drill bit and a mud pump; The data measurement module includes a gas-liquid separator exhaust pipe flowmeter, a gas-liquid separator discharge pipe flowmeter, a mud pool level meter, a drilling fluid outlet flowmeter, a casing pressure sensor, a first flowmeter, a second flowmeter, a first pressure sensor, a second pressure sensor, a third pressure sensor, a fourth pressure sensor, a logging while drilling tool, a drilling fluid inlet flowmeter, a vertical pressure sensor, a computer and a database; The pressure control module includes well control choke manifold, wellhead back pressure compensation manifold, rotary blowout preventer, annular blowout preventer and well killing manifold; The lower end of the drill string is connected to the drill bit, and the upper end of the drill string is connected to the rotary blowout preventer and the water hose. The drill string forms an annular space with the well wall and the formation; the water hose is connected to the mud pump, and a vertical pressure sensor and a drilling fluid inlet flow meter are installed on the pipeline connecting the water hose and the mud pump. The vertical pressure sensor is used to obtain the riser pressure in the drill string, and the drilling fluid inlet flow meter is used to obtain the flow rate of the drilling fluid injected into the wellbore through the drill string. The mud pump is connected to the mud pool through the mud tank to provide drilling fluid to the wellbore during the drilling process.

2. The intelligent pressure control system based on gas intrusion pressure pulse according to claim 1 is characterized in that: The wellhead spool is installed at the top of the wellbore wall, so that the annular space is connected with the wellhead back pressure compensation manifold, well control throttle manifold and well killing manifold. An annular blowout preventer is installed at the top of the wellbore wall; the wellhead back pressure compensation manifold and well control throttle manifold are connected with the gas-liquid separator and vibrating screen; The wellhead back pressure compensation manifold is installed with a drilling fluid outlet flowmeter, a first flowmeter, a second flowmeter, a first pressure sensor, a second pressure sensor, a third pressure sensor and a fourth pressure sensor; the drilling fluid outlet flowmeter is used to obtain the drilling fluid flow returned through the annular space; the wellhead back pressure compensation manifold and the vibrating screen connecting pipeline are installed with a drilling fluid outlet flowmeter, which is used to measure the drilling fluid flow returned through the annular space; the first flowmeter is installed in the pipeline connecting the wellhead back pressure compensation manifold and the well control throttle manifold, which is used to measure the drilling fluid flow from the wellhead back pressure compensation manifold to the well control throttle manifold passage; The second flow meter is installed at the wellhead back pressure compensation manifold to measure the drilling fluid flow through the passage under the current pressure-controlled drilling conditions; The fourth pressure sensor, the third pressure sensor, the second pressure sensor and the first pressure sensor are installed at the front end of each control valve of the wellhead back pressure compensation manifold, and are used to measure the pressure along the wellhead back pressure compensation manifold and the pressure before the control valve; The well control throttle manifold is equipped with a casing pressure sensor, which is used to obtain the casing pressure in the annular space; The gas-liquid separator is connected to the atmosphere on one hand, and a gas-liquid separator exhaust pipe flowmeter is installed on the pipeline connected to the atmosphere. On the other hand, it is connected to the mud pool, and a gas-liquid separator discharge pipe flowmeter is installed on the pipeline connected to the mud pool. The gas-liquid separator discharge pipe flowmeter and the gas-liquid separator exhaust pipe flowmeter are respectively used to obtain the liquid and gas phase flows separated after passing through the gas-liquid separator; the vibrating screen is connected to the mud pool, and a mud pool level meter is installed in the mud pool to measure the change of the mud pool liquid level; the logging while drilling tool is used to obtain downhole fluid measurement data, including but not limited to formation lithology, reservoir fluid properties, well depth, vertical depth, well inclination angle, drill bit position, and bottom hole pressure data; The computer is respectively connected with the gas-liquid separator exhaust pipe flowmeter, the gas-liquid separator discharge pipe flowmeter, the mud pool level meter, the drilling fluid outlet flowmeter, the casing pressure sensor, the first flowmeter, the second flowmeter, the first pressure sensor, the second pressure sensor, the third pressure sensor, the fourth pressure sensor, the logging while drilling tool, the drilling fluid inlet flowmeter, and the vertical pressure sensor, and the generated data is transmitted to the computer; the computer is connected with the well control throttle manifold, the wellhead back pressure compensation manifold, the rotary blowout preventer, the annular blowout preventer, and the well killing manifold for transmitting the pressure control parameters; the computer is connected with the database, and the database is used to store the drilling wellhead pressure pulse detection data, the pressure control drilling parameter information, and read the historical wellhead pressure pulse detection data and the historical pressure control drilling data of the adjacent wells; the data measurement module also includes ground instruments.

3. An intelligent pressure control method based on gas intrusion pressure pulse, characterized in that: The intelligent pressure control system based on gas intrusion pressure pulse according to claim 2 is implemented, comprising the following steps: S1: Acquire wellhead pressure pulse detection data, managed pressure drilling data and pressure control targets; S2: constructing input vectors based on wellhead pressure pulse detection data and managed pressure drilling data; S3: inputting the input vector into the formation pressure prediction model to obtain the predicted formation pressure under the current pressure control condition; S4: judging whether gas invasion occurs by predicting the relationship between formation pressure and bottom hole pressure. The bottom hole pressure is the sum of the wellbore hydrostatic pressure and the circulating pressure loss. When the predicted formation pressure is greater than the bottom hole pressure, it is judged that gas invasion occurs and S5 is executed; if not, step S3 is executed; S5: predict the gas invasion location based on the wellhead pressure pulse detection data and adjust the pressure control plan; S6: Acquire the wellhead pressure pulse detection data when adjusting the pressure control scheme and the pressure control drilling data after adjustment; S7: judging whether the pressure control target is met after the pressure control scheme is implemented through the wellhead pressure pulse detection data and the pressure control drilling data obtained in step S6, if so, executing step S8; if not, executing step S9; S8: Perform managed pressure drilling according to the updated managed pressure plan; S9: Update the input vector, correct the formation pressure prediction result, and return to step S5.

4. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 3 is characterized in that: In step S1, the drilling wellhead pressure pulse detection data is obtained by the casing pressure sensor and the vertical pressure sensor during the drilling process, including the pressure fluctuations caused by gas invasion or wellhead valve adjustment; The pressure-controlled drilling data includes drilling parameters and pressure-controlled parameters. The drilling parameters include drilling time, well depth, vertical depth, well inclination, drill bit position, mud pool liquid level, drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, drilling fluid outlet temperature, downhole fixed-point pressure, circulating pressure consumption, drill bit pressure drop, wellbore hydrostatic pressure, fixed-point circulating equivalent density, drill bit circulating equivalent density, drilling fluid type, drilling fluid rheology, drilling fluid inlet flow rate and drilling fluid outlet flow rate under the current drilling conditions; the pressure-controlled parameters include pump pressure, first flowmeter, second flowmeter, first pressure sensor, second pressure sensor, third pressure sensor and fourth pressure sensor; among them, drilling time, drilling fluid The type and rheological properties of drilling fluid are obtained from the drilling log; the well depth, vertical depth, well inclination angle, drill bit position, downhole fixed-point pressure, fixed-point circulating equivalent density, and drill bit circulating equivalent density are obtained from the logging while drilling tool; the drilling fluid inlet density, drilling fluid outlet density, drilling fluid inlet temperature, and drilling fluid outlet temperature are obtained from the surface instrument; the circulating pressure loss, drill bit pressure drop, and wellbore hydrostatic pressure are calculated from the wellbore multiphase flow calculation equation; the mud pool liquid level is obtained from the mud pool level gauge, the drilling fluid inlet flow rate is obtained from the drilling fluid inlet flow meter, and the drilling fluid outlet flow rate is obtained from the drilling fluid outlet flow meter; the pressure control parameter neutral pressure is obtained from the vertical pressure sensor, the casing pressure is obtained from the casing pressure sensor, and the pump pressure is obtained from the mud pump; The pressure control target is that the wellbore pressure of the managed pressure drilling is greater than or equal to the sum of the formation pressure and the safety additional pressure. The wellbore pressure of the managed pressure drilling can be calculated by the wellbore multiphase flow calculation equation, and the safety additional pressure is set by the oilfield company, industry standards or drilling team; In step S2, the input vector is a vector including drilling time, wellhead pressure pulse detection data, and pressure-controlled drilling data. n dimensional vector.

5. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 4 is characterized in that: The formation pressure prediction model in step S3 adopts a long short-term memory network, and uses the historical pressure-controlled drilling data of adjacent wells, the historical wellhead pressure pulse detection data of adjacent wells, and the formation pressure data to train and obtain the final formation pressure prediction model; The process of establishing the formation pressure prediction model includes: S31: Acquire historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data, wherein the historical controlled pressure drilling data of the adjacent well, historical wellhead pressure pulse detection data of the adjacent well, and formation pressure data are obtained from the adjacent well drilling log, well history, geological logging data, or calculated from drilling parameters during the adjacent well drilling process; S32: pre-process the data obtained in S31 to construct sample data for training the model; S33: Building long short-term memory networks; S34: Construct a model evaluation index based on the long short-term memory network of the sample data set, and solve the parameters in the long short-term memory network based on the evaluation index.

6. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 5 is characterized in that: The preprocessing process of step S32 is: S321: Analyze and select the variation rules between the wellhead pressure pulse detection data and the formation pressure data under different pressure-controlled drilling data in the historical drilling of the adjacent well; S322: Deleting abnormal values ​​of historical wellhead pressure pulse detection data of adjacent wells, historical pressure-controlled drilling data of adjacent wells, and formation pressure data through a preset abnormal data judgment method; S323: expanding the missing values ​​of the historical wellhead pressure pulse detection data of the adjacent wells, the historical pressure-controlled drilling data of the adjacent wells, and the formation pressure data by setting a denoising diffusion probability model; S324: Calculate the variation rules of various parameters in the historical controlled pressure drilling data of the adjacent wells and the formation pressure data, calculate the correlation, and select the parameters with strong correlation with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells to reduce redundant features and noise and improve the generalization ability of the model; S325: normalizing the historical wellhead pressure pulse detection data of the adjacent well, the historical pressure-controlled drilling data of the adjacent well, and the formation pressure data; S326: splicing the processed historical wellhead pressure pulse detection data of the adjacent well, the historical pressure-controlled drilling data of the adjacent well and the formation pressure data to construct model sample data; In step S323, the denoising diffusion probability model includes a forward diffusion process and a reverse generation process. The forward diffusion process simulates the "degradation" process of data by gradually adding noise to real data; the reverse generation process gradually restores the noise to data through a reverse step-by-step denoising process. In step S324, the correlation is calculated using the following formula: ； in, For variables and formation pressure The distance correlation coefficient between The parameters of the historical drilling wellhead pressure pulse detection data and the historical pressure-controlled drilling data of the adjacent well; , is the standard deviation of the parameter’s own distance; For variables and formation pressure The distance covariance between them is obtained; the parameters with a correlation coefficient greater than 0.5 with the formation pressure data in the historical controlled pressure drilling data of the adjacent wells are selected as input parameters; In step S325, the normalization process uses max-min normalization to map each parameter feature variable to between (0, 1); In step S326, the constructed sample data are arranged by drilling time, including historical pressure-controlled drilling data of nearby wells, historical wellhead pressure pulse detection data of nearby wells, and vectors of formation pressure. Multiple groups of sample data are constructed based on the pressure-controlled drilling data, wellhead pressure pulse detection data, and formation pressure data of different drilling conditions, different wells, and different periods.

7. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 6 is characterized in that: In step S33, the long short-term memory network includes long short-term memory network neurons, each of which includes a cell state, a forget gate, an input gate, and an output gate, wherein the cell state is responsible for information transmission between multiple time steps of the grid, and its state formula is defined as: ； In the formula, is the current cell state; is the cell state at the previous time step; Output of the forget gate; is the input gate output; Candidate memory for the current time step; The long short-term memory network model solves the shortcomings of the standard recurrent neural network model in dealing with long-term problems by introducing a gating mechanism. The forget gate controls the amount of data forgotten by the model, the input gate controls the amount of information input to the current cell state, and the output gate controls the amount of data output to the next time step of the cell state. The control equation is: ； in, , , They are the weight matrices of the forget gate, cell state, and output gate respectively; is the activation function; , , , They are the bias items of forget gate, input gate, cell state and output gate respectively; is the output gate activation value; is the hidden state of the current time step; Input for the current time step; In step S34, the evaluation index is the average relative error MRE and mean square error RMSLE : ； ； in, is the number of samples; , are the formation pressure value solved by the model and the real formation pressure value respectively; The model output is calculated by forward propagation, including the input gate , Forget Gate , output gate and the candidate memory for the current time step Data, combined with the input gate and forget gate to update the current cell state , calculate the hidden state of the current time step based on the output gate and cell state ; Calculate the loss between the model output and the true value according to the evaluation index; Calculate the gradient and update the parameters through back propagation, including calculating the evaluation criteria , , , , , , The gradient of the parameters is back-propagated to each unit according to the gradient to update the input gate , Forget Gate , output gate and the candidate memory for the current time step Parameters; the long short-term memory grid model is continuously optimized through multiple forward propagation, calculation of evaluation indicators, and back propagation iterations.

8. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 7 is characterized in that: In step S5, if gas invasion does not occur, the wellbore contains a single-phase drilling fluid, and the pressure wave propagation speed is: ； in, is the pressure wave propagation speed, , are the elastic modulus of drilling fluid and pipe wall, respectively. is the drilling fluid density, is the inner diameter of the tube wall, is the tube wall thickness, then: ； in, is the corresponding time of pressure wave propagation in the wellbore without gas invasion, n is the number of grid discretes along the wellbore depth, For the The grid length of the grid, is the propagation speed of the grid pressure wave; If gas invasion occurs, the gas invasion position in the wellbore is a two-phase fluid of invading gas and drilling fluid, and the pressure wave propagation speed is: ； in, is the propagation velocity of the wellbore pressure wave under invasion state, is the elastic modulus of the intruding gas, is the volume fraction of intruding gas, is the density of the mixed fluid of invading gas and drilling fluid; ； in, The corresponding time of pressure wave propagation in the wellbore caused by gas invasion; according to the return time difference of the pressure wave , calculate its gas invasion position; According to the predicted formation pressure and gas invasion position, the wellhead back pressure value is adjusted by adjusting the valve opening in the well control throttling manifold and the wellhead back pressure compensation manifold, so that the pressure controlled drilling wellbore pressure is greater than or equal to the sum of the predicted formation pressure obtained by S3 and the safety additional value. Among them, the wellhead back pressure value can be read by the vertical pressure sensor.

9. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 8 is characterized in that: In step S7, the pressure wave return time difference is calculated through the wellhead pressure pulse detection data and the pressure controlled drilling data obtained in step S6. If the pressure wave return time difference is reduced, the drilling fluid outlet flowmeter and the drilling fluid inlet flowmeter readings are the same, and the pressure controlled drilling wellbore pressure is greater than or equal to the sum of the predicted formation pressure and the safety additional pressure, the pressure control target is met.

10. The intelligent pressure control method based on gas intrusion pressure pulse according to claim 9 is characterized in that: In step S9, the input vector is updated to correct the formation pressure prediction result. The specific process is as follows: S91: Acquire the current real-time drilling dynamic data, which includes the current gas invasion pressure pulse detection data and the pressure-controlled drilling data; S92: constructing a formation pressure prediction model as a real-time drilling dynamic sub-model using current real-time drilling dynamic data; S93: Freeze some layers of the historical formation pressure prediction model, retain its weights, fine-tune it using current real-time drilling dynamic data, and update the remaining weights; S94: weighting and fusing the output results of the real-time drilling dynamic sub-model and the historical formation pressure prediction model to obtain an output result that fuses the historical formation pressure prediction model information and the real-time drilling dynamic sub-model information; S95: Perform error-weighted integration on the current real-time drilling dynamic sub-model and the historical formation pressure prediction model to predict the formation pressure and gas invasion pressure pulse change law at the next moment; In step S93, the historical formation pressure prediction model is trained by drilling historical static data, and the historical static data includes drilling historical gas invasion pressure pulse detection data and historical pressure-controlled drilling data; the input parameters, input gates, and forget gates in the historical formation pressure prediction model are frozen, and their weights are retained. The historical formation pressure prediction model is unchanged, wherein the historical formation pressure prediction model is the formation pressure prediction model established and used before step S7; the current gas invasion pressure pulse detection data and the pressure-controlled drilling data are used to fine-tune it to adapt to the current data changes, and the fine-tuning objectives are: ； in, Output of the fine-tuned model; is the true value of formation pressure; The updated model weights are: ； in, The weights are updated by the gradient descent algorithm; In step S94, the weighted fusion method is: ； In the formula, , The historical formation pressure prediction model is determined by the model evaluation index. and real-time drilling dynamics sub-model Weight; , The historical formation pressure prediction model is and real-time drilling dynamics sub-model Output: In step S95, if the errors of the historical formation pressure prediction model and the real-time drilling dynamic sub-model at the current moment are , , then the error weighted integration method is: 。

Images