A physical model and data double-driven annular multiphase flow well bottom pressure real-time evaluation method

Abstract

This invention discloses a real-time assessment method for bottom-hole pressure in annular multiphase flow driven by both physical models and data, comprising the following steps: S1: Establishing an annular multiphase flow mechanism model; S2: Initializing the annular multiphase flow mechanism model; S3: Calculating the prior estimate of the bottom-hole pressure at the current moment based on the physical parameters of the target well at the previous time step; S4: Calculating the theoretical riser pressure and obtaining the pressure residual; if the pressure residual is less than a threshold, the prior estimate is the final estimation result; otherwise, proceed to step S5; S5: Inputting the pressure residual and the operating characteristics of the target well into a neural network, mapping and outputting the physical parameter correction; S6: Correcting the physical parameters, performing a secondary solution, and obtaining the corrected value of the bottom-hole pressure at the current moment, which is the final estimation result; S7: Repeating steps S3-S6 to obtain the real-time assessment result of the bottom-hole pressure in annular multiphase flow. This invention can quickly and accurately assess the bottom-hole pressure of annular multiphase flow in real time.

Description

Technical Field

[0001] This invention relates to the field of oil and gas drilling engineering technology, and in particular to a real-time assessment method for bottom hole pressure in annular multiphase flow driven by both physical models and data. Background Technology

[0002] As global oil and gas exploration and development expands into deeper, ultra-deep, and deepwater formations, drilling operations face increasingly complex working conditions. In deep and ultra-deep well drilling, high formation temperatures and narrow pressure windows greatly increase the risk of gas intrusion and even blowouts. Controlled pressure drilling (MPD) technology, a key solution to the challenges of drilling in narrow pressure windows, relies on precise control of bottom hole pressure (BHP). However, achieving precise pressure control requires real-time and accurate knowledge of the multiphase flow state within the wellbore.

[0003] Currently, numerical modeling methods based on physical models are the mainstream approach in theoretical research and offline analysis. These methods calculate bottom hole pressure by solving equations for the conservation of mass, momentum, and energy. However, under the high temperature and high pressure conditions of deep wells, especially when using oil-based drilling fluids, this method has significant limitations: Existing models often employ simplified solubility calculations, making it difficult to accurately describe the complex phase equilibrium behavior of acidic gases (such as CO2 and H2S) or methane in oil-based drilling fluids. In deep formations, large amounts of gas dissolve in the oil phase without volume expansion, making early gas intrusion difficult to detect by surface sensors. However, when the fluid flows back to the precipitation point near the wellhead, the gas volume expands rapidly, leading to uncontrolled well control risks.

[0004] Furthermore, physical models are highly dependent on empirical parameters such as friction coefficient and drift velocity distribution coefficient. In long open-hole sections of deep wells, due to the nonlinear changes in wellbore roughness, cuttings transport, and fluid rheology with temperature and pressure, the preset empirical parameters often deviate significantly from actual operating conditions. This leads to the accumulation of pressure calculation errors with well depth, making it difficult to meet the pressure control accuracy requirements of MPD technology.

[0005] In recent years, with the development of artificial intelligence technology, intelligent monitoring methods based on machine learning (such as neural networks and random forests) have begun to emerge. These methods attempt to predict downhole conditions by mining the correlation features of real-time data such as surface riser pressure (SPP) and flow rate difference. Although data-driven methods are computationally fast, they have the following shortcomings: Pure data models are essentially "black box" fitting, heavily reliant on the completeness of training data. However, complex faults in actual drilling, such as overflows and lost circulation, are "small sample" events with sparse data. When encountering fault conditions the model has never seen before or new wellbore structures, the model's predictions are highly susceptible to distortion. Pure neural network models do not follow physical conservation laws (such as mass conservation) and may output predictions that violate common sense (e.g., predicting negative pressure values ​​or impossible flow velocities), leading field engineers to be hesitant to fully trust their decision-making recommendations.

[0006] In summary, a single physical model is insufficient to address the parameter uncertainties under complex phase transitions, while a single data-driven approach cannot guarantee physical reliability under extreme conditions. Therefore, there is an urgent need for an online estimation method that combines the interpretability of the physical model with the adaptive capabilities of data-driven approaches. Summary of the Invention

[0007] To address the aforementioned issues, this invention aims to provide a real-time assessment method for bottom hole pressure in annular multiphase flow driven by both physical models and data.

[0008] The technical solution of the present invention is as follows: A real-time assessment method for bottom-hole pressure in annular multiphase flow driven by both physical models and data includes the following steps: S1: Establish a multiphase flow mechanism model in the annulus; S2: Obtain the basic parameters of the target well, and complete the initialization settings of the annular multiphase flow mechanism model based on the basic parameters; S3: Based on the physical parameters of the target well at a given time step, and in conjunction with the annular multiphase flow mechanism model, calculate the prior estimate of the bottom hole pressure at the current moment; the physical parameters include the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient. S4: Calculate the theoretical riser pressure based on the prior estimate of the bottom hole pressure at the current moment, and compare the theoretical riser pressure with the measured riser pressure to obtain the pressure residual; If the pressure residual is less than the pressure residual threshold, then the prior estimate of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. If the pressure residual is greater than or equal to the pressure residual threshold, proceed to step S5; S5: Input the pressure residual and the working condition characteristics of the target well into the neural network, and map the output physical parameter correction amount; S6: Correct the physical parameters according to the physical parameter correction amount, and substitute the corrected physical parameters back into the annular multiphase flow mechanism model for a second solution to obtain the corrected value of the bottom hole pressure at the current moment. The corrected value of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. S7: Repeat steps S3-S6 to obtain real-time assessment results of the bottom hole pressure in the annular multiphase flow.

[0009] Preferably, in step S1, the annular multiphase flow mechanism model includes: (1) Mass conservation equation (1) (2) in: (3) (4) In the formula: ρ g E represents the gas phase density. g R is the gas content of the gas cross section; A is the annular cross-sectional area; R s E1 is the gas solubility; E1 is the liquid holdup; t is time; u g U1 is the gas phase velocity; u1 is the liquid phase velocity; z is the well depth; q g ρ is the gas production rate; ρ1 is the drilling fluid density; C o U is the gas-liquid two-phase flow distribution coefficient; m u is the mixing velocity assuming no relative motion between gas and liquid; gm U represents the gas phase drift velocity. sl For liquid phase reduced velocity; u sg For gas phase conversion velocity; (2) Equation of conservation of mixed momentum (5) in: (6) In the formula: g is the acceleration due to gravity; θ is the well inclination angle; P is the pressure of the gas-liquid two-phase mixture; f r ρ is the annular frictional pressure drop; f is the multiphase flow mixing friction coefficient; ρ m For the density of the mixed fluid; D h The annular hydraulic diameter; (3) Energy conservation equation (7) In the formula: Q m C1 is the volumetric flow rate of the drilling fluid in the annulus; C1 is the specific heat capacity of the drilling fluid in the annulus; T a Temperature in the annular space; d w U is the wellbore diameter; w To consider the combined heat transfer coefficient of the cement ring and the sleeve; T w Formation temperature; d po h is the inner diameter of the drill string. poT is the heat transfer coefficient between the annular fluid and the outer wall of the drill string; d This refers to the temperature of the drill string's outer wall. (4) Initial conditions (8) In the formula: P(h,0) is the initial pressure at a well depth of h; F r (h) represents the annular friction at a depth of h; P0 represents the initial pressure at the wellhead of the pressure control equipment; u g (h,0), u1(h,0), E g E1(h,0) and E1(h,0) are the initial gas phase velocity, initial liquid phase velocity, initial gas cross section gas content, and initial liquid holdup at a well depth of h, respectively. (5) Boundary conditions (9) In the formula: P(0,t) is the pressure at the wellhead position at time t; v WHP The back pressure is the pressure generated per unit time by the strategy; Q g (H,t), u sg (H,t) and u sl (H,t) represent the remaining free gas volume, gas-phase reduced velocity, and liquid-phase reduced velocity at the bottom of the well at time t, respectively; q sc B represents the gas production under standard conditions. g Q is the gas volume coefficient in the gas phase; g This refers to the amount of free gas remaining after the gas that has entered the wellbore dissolves per unit time.

[0010] Preferably, in step S2, the basic parameters include surface logging parameters, formation parameters, and wellbore structure parameters. The surface logging parameters include riser pressure, inlet / outlet flow rate, mechanical drilling rate, and drill string rotation speed.

[0011] As a preferred option, in the initial iteration, in step S3, the multiphase flow mixing friction coefficient of the target well at a time step is obtained by querying the Moody diagram using the Reynolds number, and the gas-liquid two-phase flow distribution coefficient of the target well at a time step is determined according to the flow regime. If the gas-liquid state is bubbly flow or slug flow, then the gas-liquid two-phase flow distribution coefficient is taken as 1.2; If the gas-liquid state is annular flow, then the gas-liquid two-phase flow distribution coefficient is taken as 1.05.

[0012] Preferably, the Reynolds number is calculated using the following formula: (10) In the formula: Re is the Reynolds number; The effective viscosity is calculated based on the power-law fluid or Bingham fluid rheological model.

[0013] Preferably, in step S4, the theoretical riser pressure is calculated using the following formula: (11) In the formula: The theoretical riser pressure; This is a priori estimate of the bottom hole pressure; This refers to the frictional pressure loss along the friction path; This refers to the localized pressure drop generated when drilling fluid flows at high speed through the drill bit nozzle; This is the hydrostatic pressure.

[0014] Preferably, in step S5, the neural network is any one of a long short-term memory network, a gated recurrent unit, a temporal convolutional network, or a physical information neural network.

[0015] Preferably, in step S5, a logic filter is provided after the output layer of the neural network, and the logic filter has the following constraints: Constraint 1: If the wellhead circulation inflow and outflow are balanced and the mud level in the mud pit remains stable, then the forced gas-liquid two-phase flow distribution coefficient correction is 0. Constraint 2: The gas-liquid two-phase flow distribution coefficient is between [0.8, 1.5].

[0016] Preferably, in step S6, the physical parameters are corrected using the following formula: (12) (13) In the formula: and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient at the current moment, respectively. and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient of the previous time step, respectively. and All are correction gain coefficients; and These are the correction values ​​for the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient.

[0017] Preferably, the correction gain coefficient is greater than 0 and less than 1.

[0018] The beneficial effects of this invention are: This invention integrates physical models and data-driven approaches. While preserving the interpretability of pure physical models, it uses data-driven methods to automatically compensate for model deviations under complex deep well conditions, thereby achieving adaptive and accurate tracking of annular flow states and precise calculation of bottom hole pressure during gas intrusion. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the physical model of the annular flow during deep well pressure control and throttling circulation in a specific embodiment. Figure 2 A schematic diagram of the physical model of the mass conservation control volume in a specific embodiment; Figure 3 This is a schematic diagram of the physical model of the momentum conservation control volume in a specific embodiment; Figure 4 This is a schematic diagram of the physical model of the energy conservation control volume in a specific embodiment; Figure 5 This is a schematic diagram of the spatial solution domain and the spatial-temporal discrete grid in a specific embodiment; Figure 6 This is a schematic diagram of a parameter error compensation neural network architecture for the "state-residual" mapping in a specific embodiment; Figure 7 This is a schematic diagram comparing the correction of gas-liquid two-phase flow distribution coefficients using different methods under complex conditions in a specific embodiment. Figure 8 This is a schematic diagram comparing the correction of multiphase flow mixing friction coefficient using different methods under complex working conditions in a specific embodiment. Figure 9 This is a schematic diagram comparing the wellhead pressure of the present invention with that of a traditional physical model in a specific embodiment; Figure 10 This is a schematic diagram comparing the bottom hole pressure of the present invention with that of a traditional physical model in a specific embodiment; Figure 11 This is a schematic diagram comparing the gas intrusion amount of the present invention with that of a traditional physical model in a specific embodiment. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and technical features described in this application can be combined with each other. It should also be pointed out that, unless otherwise indicated, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terms "comprising" or "including" and similar words used in this invention refer to elements or objects preceding the word that encompass the elements or objects listed following the word and their equivalents, without excluding other elements or objects.

[0022] like Figure 1 As shown, this invention provides a real-time assessment method for bottom hole pressure in annular multiphase flow driven by both physical models and data, comprising the following steps: S1: Establish an annular multiphase flow mechanism model, which includes: (1) Mass conservation equation (1) (2) in: (3) (4) In the formula: ρ g E represents the gas phase density. g R is the gas content of the gas cross section; A is the annular cross-sectional area; R s E1 is the gas solubility; E1 is the liquid holdup; t is time; u g U1 is the gas phase velocity; u1 is the liquid phase velocity; z is the well depth; q g ρ is the gas production rate; ρ1 is the drilling fluid density; C o U is the gas-liquid two-phase flow distribution coefficient; m u is the mixing velocity assuming no relative motion between gas and liquid; gm U represents the gas phase drift velocity. sl For liquid phase reduced velocity; u sg For gas phase conversion velocity; (2) Equation of conservation of mixed momentum (5) in: (6) In the formula: g is the acceleration due to gravity; θ is the well inclination angle; P is the pressure of the gas-liquid two-phase mixture; f r ρ is the annular frictional pressure drop; f is the multiphase flow mixing friction coefficient; ρ m For the density of the mixed fluid; D h The annular hydraulic diameter; (3) Energy conservation equation (7) In the formula: Q m C1 is the volumetric flow rate of the drilling fluid in the annulus; C1 is the specific heat capacity of the drilling fluid in the annulus; T a Temperature in the annular space; d w U is the wellbore diameter; w To consider the combined heat transfer coefficient of the cement ring and the sleeve; T w Formation temperature; d po h is the inner diameter of the drill string. po T is the heat transfer coefficient between the annular fluid and the outer wall of the drill string; d This refers to the temperature of the drill string's outer wall. (4) Initial conditions (8) In the formula: P(h,0) is the initial pressure at a well depth of h; F r (h) represents the annular friction at a depth of h; P0 represents the initial pressure at the wellhead of the pressure control equipment; u g (h,0), u1(h,0), E g E1(h,0) and E1(h,0) are the initial gas phase velocity, initial liquid phase velocity, initial gas cross section gas content, and initial liquid holdup at a well depth of h, respectively. (5) Boundary conditions (9) In the formula: P(0,t) is the pressure at the wellhead position at time t; v WHP The back pressure is the pressure generated per unit time by the strategy; Q g (H,t), u sg (H,t) and u sl (H,t) represent the remaining free gas volume, gas-phase reduced velocity, and liquid-phase reduced velocity at the bottom of the well at time t, respectively; q sc B represents the gas production under standard conditions. g Q is the gas volume coefficient in the gas phase; g This refers to the amount of free gas remaining after the gas that has entered the wellbore dissolves per unit time.

[0023] In the above embodiments, when establishing the annular multiphase flow mechanism model, the physical model of the annular multiphase flow is as follows: Figure 1 As shown. To highlight the characteristics of the gas-liquid two-phase flow in the annulus during the controlled pressure throttling circulation well control process in deep wells, the following basic assumptions are made for this model: (1) The gas and liquid phases are limited to exhibiting dynamic migration characteristics only along the well axis direction; (2) Each phase on the same cross section satisfies the transient temperature-pressure equilibrium condition; (3) The radial dimension phase parameters follow an isotropic distribution law; (4) The gas dissolution / precipitation process is treated with a quasi-steady-state approximation; (5) Ignore drill cuttings carried in the drilling fluid and treat the drilling fluid as a liquid phase; (6) The drill string is centered, and the annular eccentricity is ignored.

[0024] To address the physics of annular multiphase flow evolution, a mathematical model of gas-liquid annular multiphase flow needs to be constructed. When formation fluid intrusion occurs and free gas exists in the annulus, the annular flow state transitions from a steady-state single-phase flow to a gas-liquid mixed-phase flow state. Based on this, and combining the continuous medium assumption, and comprehensively considering interphase mass transfer, gas dissolution and precipitation, and bottom hole pressure-gas intrusion coupling, a set of multiphase flow governing equations can be derived.

[0025] For the mass conservation equation, as Figure 2 As shown, a micro-element segment is selected as the mass conservation control volume to analyze the mass flux and interphase mass transfer process of the gas phase (including dissolved gas and free gas) and the liquid phase. The obtained mass conservation equation for the mixed gas phase is shown in Equation (1), and the obtained mass conservation equation for the liquid phase is shown in Equation (2).

[0026] For the aforementioned equation of conservation of mixed momentum, as Figure 3 As shown, according to the law of conservation of momentum, the rate of change of momentum of the control body per unit time is the momentum flowing into and out of the control body and the external force on the control body. The resulting mixed momentum conservation equation is shown in equation (5).

[0027] For the energy conservation equation, as Figure 4 As shown, when establishing the transient heat transfer model of the drilling system annulus, the differential unit dz along the wellbore axis at well depth z at time t is taken as the control volume. The thermodynamic model is the temperature field model in the annulus, and the energy conservation equation shown in equation (7) is obtained based on the law of conservation of energy in the annulus.

[0028] In the early stage of overflow, the gas phase does not invade the annulus and the annulus remains in normal circulation state, so the initial conditions shown in equation (8) can be obtained; when gas invasion occurs, the free gas flow rate and the converted velocity of the gas phase and liquid phase at the bottom of the well are determined according to the amount of gas invasion and solubility at the bottom of the well, so the boundary conditions shown in equation (9) can be obtained.

[0029] S2: Obtain the basic parameters of the target well, and complete the initialization settings of the annular multiphase flow mechanism model based on the basic parameters.

[0030] In one specific embodiment, the basic parameters include surface logging parameters, formation parameters, and wellbore structure parameters. The surface logging parameters include riser pressure, inlet / outlet flow rate, mechanical drilling rate, and drill string rotation speed.

[0031] S3: Based on the physical parameters of the target well at a time step, and combined with the annular multiphase flow mechanism model, calculate the prior estimate of the bottom hole pressure at the current moment; the physical parameters include the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient.

[0032] In a specific embodiment, during the initial iteration, in this step, the multiphase flow mixing friction coefficient of the target well at a time step is obtained by querying the Moody diagram using the Reynolds number, and the gas-liquid two-phase flow distribution coefficient of the target well at a time step is determined according to the flow regime; if the gas-liquid state is bubbly flow or slug flow, the gas-liquid two-phase flow distribution coefficient is 1.2; if the gas-liquid state is annular flow, the gas-liquid two-phase flow distribution coefficient is 1.05.

[0033] It should be noted that the methods for determining the multiphase flow mixing friction coefficient and the gas-liquid two-phase flow distribution coefficient in the above embodiments are only a preferred method of the present invention, and other methods for obtaining these parameters in the prior art can also be applied to the present invention. Furthermore, the initial values ​​can also be set manually based on experience, but such a setting will increase the computational load.

[0034] In one specific embodiment, the Reynolds number is calculated using the following formula: (10) In the formula: Re is the Reynolds number; The effective viscosity is calculated based on the power-law fluid or Bingham fluid rheological model.

[0035] It should be noted that the method for obtaining the Reynolds number is existing technology, and the above embodiment is only a preferred method for obtaining this parameter. Other existing methods for obtaining this parameter can also be applied to this invention.

[0036] In a specific embodiment, when calculating the bottom hole pressure using the annular multiphase flow mechanism model, non-uniform step sizes are used for discretization in both the time and spatial domains. Specifically: A non-uniform spatial step size is used for spatial domain discretization during mesh generation. The annulus is divided into n+1 spatial nodes from the bottom to the top, resulting in n segments of mesh with one node at the bottom and n+1 nodes at the top, as shown below. Figure 5 The solution domain is shown on the left. Let the first segment be Z(1) and the last segment be Z(n). Figure 5As shown in the right-hand time domain, when dividing the time domain into grids, the time it takes for the drilling fluid to flow through the grid in the lower part where no gas appears is taken as the time step. Assuming that gas appears at position a, as the gas is released, the time it takes for the gas to move in the space segment is taken as the step. When the gas reaches the wellhead, the overall circulation of the wellbore tends to stabilize, so the time of the last segment reaching the wellhead is taken as the time step for the subsequent circulation to discharge the contaminated drilling fluid.

[0037] S4: Calculate the theoretical riser pressure based on the prior estimate of the bottom hole pressure at the current moment, and compare the theoretical riser pressure with the measured riser pressure to obtain the pressure residual; If the pressure residual is less than the pressure residual threshold, then the prior estimate of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. If the pressure residual is greater than or equal to the pressure residual threshold, proceed to step S5.

[0038] In this step, when the pressure residual is less than the pressure residual threshold, the current parameters match the actual flow pattern, so no correction is needed, and the prior estimate of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. However, when the pressure residual is greater than or equal to the pressure residual threshold, it indicates that the accuracy requirements are not met, and the downhole fluid properties or flow pattern have drifted, so the physical parameters need to be corrected.

[0039] In one specific embodiment, the theoretical riser pressure is calculated using the following formula: (11) In the formula: The theoretical riser pressure; This is a priori estimate of the bottom hole pressure; This refers to the frictional pressure loss along the friction path; This refers to the localized pressure drop generated when drilling fluid flows at high speed through the drill bit nozzle; This is the hydrostatic pressure.

[0040] S5: Input the pressure residual and the operating characteristics of the target well into the neural network, and map the output physical parameter correction amount.

[0041] In one specific embodiment, the neural network is any one of a Long Short-Term Memory (LSTM) network, a gated recurrent unit (GRU) network, a temporal convolutional network, and a physical information neural network. Optionally, establishing the LSM network includes the following steps: 1. Construct the input feature space Considering that changes in gas-liquid distribution within the wellbore during deep well gas invasion will simultaneously cause fluctuations in surface circulation parameters and changes in the downhole pressure field, the prediction bias of the physical model itself contains information about flow drift. Therefore, this embodiment selects key variables covering three dimensions—surface boundary conditions, downhole response state, and model feedback signal—to construct the feature input vector of the neural network.

[0042] (1) Definition of feature space Construct an input vector containing M=5 dimensional features. It covers three dimensions: surface, downhole, and model feedback. Riser pressure reflects the total energy level of the circulation system; The difference in flow rate between the inlet and outlet reflects the entry and exit of formation fluids; Mechanical drilling rate, as a substitute indicator for cuttings production, indirectly characterizes the effect of annular solids content on friction. Drill wheel rotation speed characterizes the shear rate of the fluid and affects the effective viscosity of non-Newtonian fluids; Physical model residual is defined as the difference between the measured riser pressure and the pressure calculated by the physical model at the previous moment, and is used as an error feedback signal.

[0043] (2) Time series data preprocessing Because wellbore flow exhibits significant hysteresis, data from a single moment cannot reflect the overall system picture. This embodiment employs a sliding window technique to construct a time-series tensor: setting the time window length to L (in a specific embodiment, L=50, corresponding to data from the past 25 seconds, with a sampling frequency of 2Hz) and constructing the input tensor. : (14) 2. Design the neural network architecture like Figure 6 As shown, the neural network architecture is an embedded neural network error compensation model, comprising: (1) Feature extraction layer A double-layer stacked LSTM unit is used to capture the long-term and short-term dependencies of time series. The LSTM unit controls the information flow through forget gates, input gates, and output gates. Its core operation formula is as follows: (15) In the formula: , and These are the forget gate, input gate, and output gate vectors at time t, respectively. The feature vector input at the current moment contains multi-source logging parameters and physical model residuals; and These are the hidden state vectors from the previous time step and the current time step, respectively; , and These represent the cell state at the previous time step, the cell state at the current time step, and the candidate memory state, respectively. , , , This is the weight matrix for each gated computation unit; , , , This is the corresponding bias vector; The activation function is a gating function; optionally, the activation function is a sigmoid activation function. It is a hyperbolic tangent nonlinear activation function.

[0044] (2) Mapping output layer LSTM layer output The parameters are mapped to correction coefficients for physical parameters through a fully connected layer. To prevent excessive parameter correction from leading to large errors in the calculated physical model data, a Tanh activation function is used in the output layer and multiplied by the maximum allowable correction magnitude. : (16) Where: output vector ,set up , and These are the weight matrix and bias vector of the fully connected mapping layer, respectively.

[0045] In one specific embodiment, the gas-liquid two-phase flow distribution coefficient is at most corrected. The maximum correction for the multiphase flow mixing friction coefficient In this embodiment, the above constraints can prevent the distortion of the bottom hole pressure calculated by the model due to excessive correction of the gas-liquid two-phase flow distribution coefficient (the gas-liquid two-phase flow distribution coefficient itself has a very small range of values); the friction coefficient of multiphase flow mixing has a large uncertainty, and the friction coefficient calculated at different locations may vary greatly, so a weak constraint is given.

[0046] To ensure that the output of the neural network conforms to physical principles, in a specific embodiment, a logic filter is provided after the output layer of the neural network, and the logic filter has the following constraints: Constraint 1: If the wellhead circulation inflow and outflow are balanced, the mud level in the mud pit remains stable (i.e., If the amount of mud entering and exiting is basically the same, indicating that it is under normal circulation conditions, then the forced gas-liquid two-phase flow distribution coefficient correction is 0. Constraint 2: The gas-liquid two-phase flow distribution coefficient is between [0.8, 1.5].

[0047] In the above embodiments, constraint one can prevent the network from adjusting the slippage parameter under normal operating conditions, and constraint two can prevent unreasonable parameters from occurring.

[0048] 3. Training and Adaptation of Neural Networks A supervised learning model is used to train the neural network offline, and an online sliding window technique is combined to achieve adaptive updating of the model, so as to ensure the generalization performance of the system under different geological conditions and drilling conditions.

[0049] A large amount of simulation data was generated using an annular multiphase flow mechanism model, covering various operating conditions such as gas intrusion, well leakage, and well washing, to obtain the correspondence between "prior residuals and physical parameter offsets" under different flow regimes. For each simulation time step, an optimized algorithm was used to back-calculate a set of ideal parameters that make the error of the annular multiphase flow model close to zero. As training labels.

[0050] Loss function: Mean squared error (MSE) is used. (17) Where: L is the mean squared error loss value during offline training; N is the total number of sample batches participating in network training; Correction vector for physical parameters output by the neural network; These are labels for ideal correction parameters calculated based on the physical model.

[0051] In actual drilling operations, measured riser pressure data is used to calculate the prediction error of the physical model. If the error exceeds a threshold (optionally, the threshold is 5%), then data from the most recent historical period is used to update the weights of the fully connected layers of the network using fast gradient descent, with the goal of minimizing the pressure residual. (18) In the formula: The loss function value during the online adaptive fine-tuning phase characterizes the model's fit to the current transient pressure. The measured riser pressure data at time t; The riser pressure is calculated from the bottom hole pressure prediction obtained by the annular multiphase flow mechanism model under the current parameter conditions, and is used to compare with the measured value to generate a residual signal.

[0052] In this way, the neural network can continuously adapt to the specific formation characteristics of the current well section, gradually improving the reliability and accuracy of the prediction.

[0053] S6: Correct the physical parameters according to the physical parameter correction amount, and resubmit the corrected physical parameters into the annular multiphase flow mechanism model for a second solution to obtain the corrected value of the bottom hole pressure at the current moment. The corrected value of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment.

[0054] In one specific embodiment, the physical parameters are corrected using the following formula: (12) (13) In the formula: and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient at the current moment, respectively. and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient of the previous time step, respectively. and All are correction gain coefficients; and These are the correction values ​​for the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient.

[0055] In one specific embodiment, the corrected gain coefficient is greater than 0 and less than 1.

[0056] S7: Repeat steps S3-S6 to obtain real-time assessment results of the bottom hole pressure in the annular multiphase flow.

[0057] In a specific embodiment, taking the ShuangXX well in the Shuangyushi structure of Sichuan as an example, the real-time bottom pressure assessment method of annular multiphase flow driven by the physical model and data described in this invention is used to assess the bottom pressure in real time.

[0058] In this embodiment, the test well section is located in the Sankai Ziliu Well Group (3280m~3670m). The average temperature of this block is 15.4℃, the geothermal gradient is 1.83℃ / 100m, the drilling fluid density is 1.41g / cm³, the drilling fluid discharge rate is 45L / s, the formation permeability is 20mD, and the reservoir thickness is 2m.

[0059] The simulation reproduced the high-pressure gas intrusion condition encountered by the well at a depth of 3638.32m. To address the calculation errors caused by the use of fixed parameters in traditional physical models (distribution coefficient fixed at 1.2 or 1.05, and friction coefficient changing monotonically only with Reynolds number), this invention utilizes neural network error compensation to perform online inversion and correction of key parameters based on real-time riser pressure residuals. The dynamic parameter adjustment process and the final control effect are as follows: Figures 7-11 As shown.

[0060] from Figures 7-8 As can be seen, in response to the problem of fixed parameters in traditional models, this invention can adaptively adjust the gas-liquid two-phase flow distribution coefficient within the range of 1.05 to 1.40 using a neural network. This breaks the limitation of traditional models that set the distribution coefficient to a fixed value (if the gas-liquid state is bubbly or slug flow, the gas-liquid two-phase flow distribution coefficient is 1.2; if the gas-liquid state is annular flow, the gas-liquid two-phase flow distribution coefficient is 1.05). At the same time, it compensates for the nonlinear deviation of the multiphase flow mixing friction coefficient in real time, effectively correcting the physical model distortion problem caused by flow drift.

[0061] from Figures 9-11 It can be seen that the maximum pressure required by the pure physical model of deep well pressure control and throttling circulation well control is 7.72 MPa, which is further reduced to 5.85 MPa after neural network optimization. This is in stark contrast to the maximum casing pressure of 11.1 MPa required by conventional well control methods; compared to the 4.3m pressure caused by conventional well control methods... 3 Overflow rate can be controlled within 1.5m using the throttling and circulation well control method. 3 Based on this, the present invention utilizes neural network error compensation to further reduce the theoretical overflow to 0.62m. 3 Within this range, the system performance is further improved by more than 30%, which fully demonstrates the significant effectiveness of this invention in gas invasion control and significantly improves the accuracy of well control monitoring and the efficiency of treatment under complex deep well conditions.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A real-time assessment method for bottom-hole pressure in annular multiphase flow driven by both physical models and data, characterized in that, Includes the following steps: S1: Establish a multiphase flow mechanism model in the annulus; S2: Obtain the basic parameters of the target well, and complete the initialization settings of the annular multiphase flow mechanism model based on the basic parameters; S3: Based on the physical parameters of the target well at a given time step, and in conjunction with the annular multiphase flow mechanism model, calculate the prior estimate of the bottom hole pressure at the current moment; the physical parameters include the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient. S4: Calculate the theoretical riser pressure based on the prior estimate of the bottom hole pressure at the current moment, and compare the theoretical riser pressure with the measured riser pressure to obtain the pressure residual; If the pressure residual is less than the pressure residual threshold, then the prior estimate of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. If the pressure residual is greater than or equal to the pressure residual threshold, proceed to step S5; S5: Input the pressure residual and the working condition characteristics of the target well into the neural network, and map the output physical parameter correction amount; S6: Correct the physical parameters according to the physical parameter correction amount, and substitute the corrected physical parameters back into the annular multiphase flow mechanism model for a second solution to obtain the corrected value of the bottom hole pressure at the current moment. The corrected value of the bottom hole pressure at the current moment is the final estimate of the bottom hole pressure at the current moment. S7: Repeat steps S3-S6 to obtain real-time assessment results of the bottom hole pressure in the annular multiphase flow.

2. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, In step S1, the annular multiphase flow mechanism model includes: (1) Mass conservation equation （1） （2） in: （3） （4） In the formula: ρ g E represents the gas phase density. g R is the gas content of the gas cross section; A is the annular cross-sectional area; R s E1 is the gas solubility; E1 is the liquid holdup; t is time; u g U1 is the gas phase velocity; u1 is the liquid phase velocity; z is the well depth; q g ρ is the gas production rate; ρ1 is the drilling fluid density; C o U is the gas-liquid two-phase flow distribution coefficient; m u is the mixing velocity assuming no relative motion between gas and liquid; gm U represents the gas phase drift velocity. sl For liquid phase reduced velocity; u sg For gas phase conversion velocity; (2) Equation of conservation of mixed momentum （5） in: （6） In the formula: g is the acceleration due to gravity; θ is the well inclination angle; P is the pressure of the gas-liquid two-phase mixture; f r ρ is the annular frictional pressure drop; f is the multiphase flow mixing friction coefficient; ρ m For the density of the mixed fluid; D h The annular hydraulic diameter; (3) Energy conservation equation （7） In the formula: Q m C1 is the volumetric flow rate of the drilling fluid in the annulus; C1 is the specific heat capacity of the drilling fluid in the annulus; T a Temperature in the annular space; d w U is the wellbore diameter; w To consider the combined heat transfer coefficient of the cement ring and the sleeve; T w Formation temperature; d po h is the inner diameter of the drill string. po T is the heat transfer coefficient between the annular fluid and the outer wall of the drill string; d This refers to the temperature of the drill string's outer wall. (4) Initial conditions （8） In the formula: P(h,0) is the initial pressure at a well depth of h; F r (h) represents the annular friction at a depth of h; P0 represents the initial pressure at the wellhead of the pressure control equipment; u g (h,0), u1(h,0), E g E1(h,0) and E1(h,0) are the initial gas phase velocity, initial liquid phase velocity, initial gas cross section gas content, and initial liquid holdup at a well depth of h, respectively. (5) Boundary conditions （9） In the formula: P(0,t) is the pressure at the wellhead position at time t; v WHP The back pressure is the pressure generated per unit time by the strategy; Q g (H,t), u sg (H,t) and u sl (H,t) represent the remaining free gas volume, gas-phase reduced velocity, and liquid-phase reduced velocity at the bottom of the well at time t, respectively; q sc B represents the gas production under standard conditions. g Q is the gas volume coefficient in the gas phase; g This represents the amount of free gas remaining after the gas that has entered the wellbore dissolves per unit time.

3. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, In step S2, the basic parameters include surface logging parameters, formation parameters, and wellbore structure parameters. The surface logging parameters include riser pressure, inlet / outlet flow rate, mechanical drilling speed, and drill string rotation speed.

4. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, When iterating for the first time, in step S3, the multiphase flow mixing friction coefficient of the target well at one time step is obtained by querying the Moody diagram using the Reynolds number, and the gas-liquid two-phase flow distribution coefficient of the target well at one time step is determined according to the flow regime. If the gas-liquid state is bubbly flow or slug flow, then the gas-liquid two-phase flow distribution coefficient is taken as 1.2; If the gas-liquid state is annular flow, then the gas-liquid two-phase flow distribution coefficient is taken as 1.

05.

5. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 4, characterized in that, The Reynolds number is calculated using the following formula: （10） In the formula: Re is the Reynolds number; The effective viscosity is calculated based on the power-law fluid or Bingham fluid rheological model.

6. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, In step S4, the theoretical riser pressure is calculated using the following formula: （11） In the formula: The theoretical riser pressure; This is a priori estimate of the bottom hole pressure; This refers to the frictional pressure loss along the friction path; This refers to the localized pressure drop generated when drilling fluid flows at high speed through the drill bit nozzle; This is the hydrostatic pressure.

7. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, In step S5, the neural network is any one of a long short-term memory network, a gated recurrent unit, a temporal convolutional network, or a physical information neural network.

8. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 1, characterized in that, In step S5, a logic filter is provided after the output layer of the neural network, and the logic filter has the following constraints: Constraint 1: If the wellhead circulation inflow and outflow are balanced and the mud level in the mud pit remains stable, then the forced gas-liquid two-phase flow distribution coefficient correction is 0. Constraint 2: The gas-liquid two-phase flow distribution coefficient is between [0.8, 1.5].

9. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data, as described in any one of claims 1-8, is characterized in that... In step S6, the physical parameters are corrected using the following formula: （12） （13） In the formula: and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient at the current moment, respectively. and These are the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient of the previous time step, respectively. and All are correction gain coefficients; and These are the correction values ​​for the gas-liquid two-phase flow distribution coefficient and the multiphase flow mixing friction coefficient.

10. The method for real-time assessment of bottom hole pressure in annular multiphase flow driven by both physical model and data as described in claim 9, characterized in that, The corrected gain coefficient is greater than 0 and less than 1.

Images