Drilling mud circulating system

Abstract

The invention provides a drilling mud circulating system, and relates to the technical field of drilling engineering. The system comprises a sensor group used for collecting drilling state parameters; the fusion module is used for performing fusion calculation on the parameters based on an evidence theory discrimination model to generate a first vector representing the probability distribution of the wellbore cleaning state; the feature extraction module is used for calculating a frequency domain energy spectrum of an annulus flow field by adopting two-dimensional fast Fourier transform and vector decomposition based on the first vector and the drilling state parameters, and extracting flow field mode feature parameters from the frequency domain energy spectrum to form a second vector; and the decision module is used for determining pulse control parameters including frequency and amplitude according to a preset mapping relation in combination with the first vector and the second vector. According to the system, the pressure pulse frequency and amplitude matched with the current cleaning requirement can be stably set in real time under the conditions of multi-source uncertainty and rapid flow field change, so that the shaft cleaning efficiency is improved, and energy consumption and error control are reduced.

Description

Technical Field

[0001] This application relates to the field of drilling engineering technology, and more specifically, to a drilling mud circulation system. Background Technology

[0002] During drilling operations, mud circulation plays multiple roles, including carrying drill cuttings, cooling the drill bit, and maintaining wellbore pressure balance. Wellbore cleaning effectiveness directly impacts drilling efficiency and downhole accident risks. However, most existing mud circulation controls rely on threshold judgments or empirical adjustments of single parameters, such as inferring wellbore cleaning status through changes in riser pressure and flow rate, and then manually or statically setting pulse frequency and amplitude. These methods have significant limitations: firstly, sensor measurements during drilling inevitably contain noise and uncertainty, and the coupling relationships between different parameters are difficult to characterize using simple thresholds; secondly, the annular flow field exhibits complex transient characteristics, especially under drill string rotation, cuttings deposition, and non-Newtonian rheological effects of mud, causing the flow field pattern to change rapidly with operating conditions, making fixed pulse parameters often unsuitable for actual cleaning requirements.

[0003] This leads to an urgent engineering problem: how to determine the pressure pulse frequency and amplitude in real time and robustly to match the current cleaning requirements under conditions of multi-source uncertainty and rapid flow field changes, so as to improve wellbore cleaning efficiency and reduce miscontrol and energy consumption. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a drilling mud circulation system, comprising:

[0005] The sensor group is configured to collect drilling status parameters during the operation of the drilling rig, including at least one of drilling pressure, standpipe pressure, flow rate and mud density;

[0006] The fusion module is configured as a clean state discrimination model based on evidence theory, which performs fusion calculation on the drilling state parameters to generate a first vector to characterize the probability distribution of the wellbore clean state.

[0007] The feature extraction module is configured to calculate the frequency domain energy spectrum of the annular flow field based on the first vector and the drilling state parameters using two-dimensional fast Fourier transform and vector decomposition, and extract flow field mode feature parameters from the energy spectrum to form a second vector;

[0008] The decision module is configured to determine pulse control parameters, including frequency and amplitude, based on a preset parameter mapping relationship and by combining the first vector and the second vector.

[0009] A pulse generator is arranged in the mud circulation pipeline and configured to output pressure pulses with corresponding frequency and amplitude based on the pulse control parameters.

[0010] Optionally, generating the first vector for characterizing the probability distribution of the wellbore cleanliness state includes:

[0011] The real-time signal of each of the drilling state parameters is decomposed into basic confidence components corresponding to different wellbore cleanliness states;

[0012] Based on the basic confidence component and the corresponding signal quality factor, determine the coefficients used to weight the attenuation of the basic confidence component;

[0013] The Dempster combination rule is used to fuse all the basic confidence components after weighted attenuation coefficient processing to form a comprehensive confidence distribution;

[0014] In response to the comprehensive confidence distribution satisfying a preset convergence condition, stable probability values ​​corresponding to each cleanliness state are extracted from the comprehensive confidence distribution to generate the first vector.

[0015] Optionally, the signal decomposition rules include:

[0016] The real-time signal of the drilling status parameters is decomposed in the frequency domain to obtain at least one frequency component;

[0017] The frequency components are matched with the preset wellbore cleaning state characteristic frequency bands to form the basic confidence components.

[0018] Optionally, the convergence condition includes: within a consecutive number of calculation iterations, the rate of change of the comprehensive confidence distribution is less than a preset threshold.

[0019] Optionally, the calculation of the frequency domain energy spectrum of the annular flow field includes:

[0020] Multiple baseline boundary models are pre-constructed, and each of the baseline boundary models corresponds to a wellbore cleanliness state in the probability distribution of wellbore cleanliness state.

[0021] Extract the probability values ​​from the first vector that correspond to various wellbore cleaning states, and set the probability values ​​as weighting factors;

[0022] The weighting factor is weighted and superimposed with the corresponding multiple benchmark boundary models to generate an instantaneous equivalent boundary model.

[0023] Optionally, the calculation of the frequency domain energy spectrum of the annular flow field further includes:

[0024] The solid wall boundary conditions of the annular flow field computational domain are constructed based on the instantaneous equivalent boundary model, and the fluid properties and driving conditions of the annular flow field computational domain are set based on the drilling state parameters.

[0025] Under the combined constraints of the solid wall boundary conditions and fluid properties, the computational domain of the annular flow field is numerically solved to generate a two-dimensional velocity vector field;

[0026] The frequency domain energy spectrum is obtained by performing vector decomposition and two-dimensional fast Fourier transform on the two-dimensional velocity vector field.

[0027] Optionally, the solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model include:

[0028] The inner wall of the annular flow field computational domain is set as a rotatable movable solid wall, which is used to simulate the effect of drill string rotation or screw oscillation on the near-wall flow field.

[0029] The rotation direction and equivalent angular velocity of the moving solid wall are derived from the probability weights corresponding to the clean state in the instantaneous equivalent boundary model.

[0030] Optionally, the solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model further include:

[0031] Equivalent porous media boundary conditions are set in the peripheral lower region of the annular computational domain to simulate the flow and infiltration effect caused by solid deposition;

[0032] The equivalent thickness, porosity, and permeability parameters of the porous boundary are weighted according to the probability values ​​in the first vector corresponding to the solid deposition state on the lower side of the annulus, and set according to a preset mapping rule.

[0033] The interface conditions between the porous boundary and the main fluid region are set to meet the boundary requirements of continuous normal velocity and continuous tangential stress.

[0034] Optionally, the solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model include:

[0035] Based on the first vector and the drilling state parameters, identify the area covered by filter cake in the annular circumference;

[0036] An anisotropic partial slip boundary condition is applied to the area covered by the filter cake, and the tangential slip length parameter of the slip boundary is set according to the probability values ​​corresponding to the clean stable state and the disturbed resuspension state in the first vector by weighting.

[0037] Non-slip boundary conditions are applied to the remaining wall areas not identified as filter cake covered, and a circumferential segmented distribution of equivalent sand grain roughness is set according to the first vector to determine the shear stress distribution of the wall surface.

[0038] Optionally, generating the two-dimensional velocity vector field includes:

[0039] Based on the pulse control parameters, the inlet and outlet driving conditions of the annular flow field calculation domain are set so that the volumetric flow rate or pressure at the inlet / outlet changes periodically with time according to the frequency and amplitude of the pulse control parameters.

[0040] At each time step, the instantaneous equivalent boundary condition model is mapped to a solid wall boundary condition for solving. The solid wall boundary condition includes at least one or more of the following: a rotatable solid wall boundary, the equivalent porous medium boundary, and the anisotropic partial slip boundary and the circumferential segmented roughness boundary.

[0041] A non-Newtonian rheological model including the yield stress term is selected to characterize the mud fluid properties. Under the joint constraints of the solid wall boundary conditions and fluid properties, the transient pressure and velocity of the annular flow field computational domain are coupled to obtain the instantaneous two-dimensional velocity vector field on the selected well section cross section.

[0042] Compared with existing technologies, the drilling mud circulation system proposed in this application establishes a dual-vector representation of state and flow field by fusing multi-source state parameters into a probability distribution vector through evidence theory and combining it with annular flow field mode feature extraction based on two-dimensional fast Fourier transform. Based on this, the system can adaptively determine the frequency and amplitude of pressure pulses, thereby achieving targeted cleaning control under different well sections and unstable conditions. Compared with traditional methods relying on experience or single features, this application significantly improves the robustness and reliability of state discrimination, avoiding misjudgments and invalid pulses caused by measurement uncertainties. Simultaneously, the flow field mode-based control strategy can effectively stimulate cuttings resuspension and suppress deposition, ensuring long-term wellbore unobstructed flow and improving the automation level and energy efficiency of drilling operations. Attached Figure Description

[0043] Figure 1 A schematic diagram of a drilling mud circulation system provided in an embodiment of this application;

[0044] Figure 2 A flowchart illustrating a method for generating a first vector characterizing the probability distribution of wellbore cleanliness, provided in an embodiment of this application;

[0045] Figure 3 A flowchart illustrating a method for calculating the frequency domain energy spectrum of annular flow fields, provided in an embodiment of this application;

[0046] Figure 4 A flowchart illustrating a method for constructing solid wall boundary conditions for annular flow field computational domain based on the instantaneous equivalent boundary model, provided in this application embodiment. Detailed Implementation

[0047] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0048] See Figure 1 The diagram shown is a schematic of a drilling mud circulation system provided in an embodiment of this application, including:

[0049] The sensor group is configured to collect drilling status parameters during the operation of the drilling rig, including at least one of drilling pressure, standpipe pressure, flow rate and mud density;

[0050] The fusion module is configured as a clean state discrimination model based on evidence theory, which performs fusion calculation on the drilling state parameters to generate a first vector to characterize the probability distribution of the wellbore clean state.

[0051] The feature extraction module is configured to calculate the frequency domain energy spectrum of the annular flow field based on the first vector and the drilling state parameters using two-dimensional fast Fourier transform and vector decomposition, and extract flow field mode feature parameters from the energy spectrum to form a second vector;

[0052] The decision module is configured to determine pulse control parameters, including frequency and amplitude, based on a preset parameter mapping relationship and by combining the first vector and the second vector.

[0053] A pulse generator is arranged in the mud circulation pipeline and configured to output pressure pulses with corresponding frequency and amplitude based on the pulse control parameters.

[0054] Regarding the aforementioned sensor group:

[0055] The sensor array provides the raw data input required for subsequent calculations in the system described in this application.

[0056] The sensor array may include a pressure sensor for measuring riser pressure, a flow meter for measuring flow rate, a densitometer for measuring mud density, and a sensor for obtaining drilling pressure. These parameters were selected because changes in their values ​​are physically correlated with changes in the cleanliness of the wellbore.

[0057] For example, low-frequency trend changes in riser pressure can be used to characterize changes in overall annular frictional resistance; flow rate is used to determine cuttings transport capacity; differences in mud inlet and outlet densities can be used to characterize the cuttings load on the mud; and changes in drilling pressure are related to the rate of cuttings generation at the bottom of the well.

[0058] Secondly, the collected drilling status parameters are processed for time synchronization. The data for these parameters can be sourced from a surface electronic drilling record system or a downhole measurement-while-drilling system.

[0059] In addition, since the data may come from multiple subsystems with different sampling frequencies and time bases, this embodiment also includes a data acquisition and synchronization unit. This unit assigns a unified timestamp to parameter signals from different sources and can perform alignment or interpolation processing according to a preset time step to form a time-synchronized multidimensional dataset for use by the subsequent fusion module.

[0060] Next, the synchronized data undergoes preprocessing. Preprocessing steps may include:

[0061] Digital filters can be applied to parameter signals that include noise. For example, a low-pass filter can be used on the riser pressure signal to separate the low-frequency trend signal that reflects the overall annular pressure loss change.

[0062] A statistical threshold-based method can be used to identify, remove, or correct abnormal numerical points in the data sequence caused by abnormal operating conditions or signal interference.

[0063] Since the physical dimensions and numerical ranges of each parameter are different, the numerical ranges of each parameter sequence can be uniformly mapped linearly or nonlinearly so that they fall into a preset numerical range for subsequent comprehensive calculation of multidimensional parameters.

[0064] Regarding the aforementioned fusion module:

[0065] The fusion module's data input is connected to the output of the sensor array. Its core function is to receive the multi-dimensional drilling state parameter data stream from the sensor array and perform comprehensive analysis to address the uncertainties and limitations caused by single parameters or simple threshold methods. The module's final output is a first vector that quantitatively characterizes the probability distribution of the current wellbore cleanliness.

[0066] In this embodiment, the cleanliness status discrimination model built into the fusion module is preferably constructed based on Dempster-Shafer Theory. This theory is chosen because it can effectively handle uncertain and incomplete information, does not require a large amount of prior probability data for model training, and can combine sensor evidence from different physical properties within a unified mathematical framework.

[0067] In practice, each received drilling condition parameter data stream is transformed into an evidentiary form that the evidence theory model can process. Specifically, the module assesses the degree of support for various preset wellbore cleanliness states based on the changes in the values ​​of each parameter or their derived characteristics. For example, preset states could include "highly efficient cleanliness," "cuttings bed formation," and "stable cuttings bed." The purpose of this step is to translate the raw physical measurements into a "trust allocation" for different assumptions that the model can understand.

[0068] Secondly, the model employs built-in combination rules within the cleanliness state discrimination model to aggregate evidence from all drilling state parameters. This combination process is not a simple linear superposition but follows specific rules of evidence theory, integrating all evidence to arrive at a more reliable and comprehensive confidence distribution. During this combination process, the model can also introduce a reliability assessment mechanism for different evidence sources to adjust the contribution of different pieces of evidence to the final discrimination result. For example, the contribution of evidence from a sensor with temporarily degraded signal quality will be dynamically reduced.

[0069] Furthermore, this fusion calculation process can be iterative. In each calculation cycle, the module updates the confidence distribution based on the new data input until the probability distribution calculation results of each clean state tend to stabilize and meet a preset discrimination convergence condition, only then is the discrimination result of the current cycle confirmed to be valid.

[0070] Finally, the fusion module outputs the first vector.

[0071] For example, the first vector can be an array structure containing multiple numerical elements. For instance, if the preset cleaning states include the three mentioned above, the first vector can be specifically represented as [P1, P2, P3], where P1, P2, and P3 are the probability values ​​for determining whether the system is in a "high-efficiency cleaning state," a "cuttings bed formation state," and a "stable cuttings bed state" at the current moment, respectively, and the sum of the values ​​of each element is 1. This first vector is then transmitted to the feature extraction module as one of the key inputs for its subsequent calculations.

[0072] Optional, see Figure 2 The flowchart of a method for generating a first vector to characterize the probability distribution of wellbore cleanliness provided in this application embodiment includes steps S101 to S104, wherein:

[0073] S101: Decompose the real-time signal of each of the drilling state parameters into basic confidence components corresponding to different wellbore cleanliness states;

[0074] S102: Based on the basic confidence component and the corresponding signal quality factor, determine the coefficients used to weight and attenuate the basic confidence component;

[0075] S103: Using the Dempster combination rule, all the basic confidence components processed by the weighted attenuation coefficient are fused to form a comprehensive confidence distribution;

[0076] S104: In response to the comprehensive confidence distribution satisfying the preset convergence condition, extract the stable probability value corresponding to each cleanliness state from the comprehensive confidence distribution to generate the first vector.

[0077] Optionally, the signal decomposition rules include:

[0078] The real-time signal of the drilling status parameters is decomposed in the frequency domain to obtain at least one frequency component;

[0079] The frequency components are matched with the preset wellbore cleaning state characteristic frequency bands to form the basic confidence components.

[0080] Optionally, the convergence condition includes: within a consecutive number of calculation iterations, the rate of change of the comprehensive confidence distribution is less than a preset threshold.

[0081] To further clarify the working principle of the fusion module, an optional embodiment of this application provides a specific method for generating the first vector, which can be broken down into the following consecutive steps:

[0082] Step 1: Decomposition and generation of basic confidence components;

[0083] The core of this step is to transform the continuous physical signals from the sensor array into discretized "evidence," or fundamental confidence components, usable within the framework of evidence theory. In practice, firstly, the real-time signal sequence of a single drilling state parameter, such as riser pressure, is decomposed in the frequency domain within a preset time window. This decomposition can be achieved using conventional signal processing methods in the field, such as Fast Fourier Transform or Wavelet Transform, with the aim of obtaining the energy distribution of the signal at different frequencies, i.e., obtaining at least one frequency component.

[0084] Then, the obtained frequency components are matched against a pre-defined "wellbore cleaning state characteristic frequency band" knowledge base. This knowledge base pre-stores the correlation between different wellbore cleaning states and specific frequency signals. For example, the knowledge base may include the following rules:

[0085] Example 1: If the riser pressure signal shows a sustained increase in energy at extremely low frequencies (e.g., below 0.01 Hz), this is highly correlated with the physical process of increased overall frictional resistance in the annulus due to slow rock debris deposition. Therefore, when this characteristic is matched, the system generates a base confidence component with a high degree of confidence in the "rock debris bed formation state".

[0086] Example 2: If the bottom hole torque or vibration signal shows a significant energy peak in a specific high-frequency band (e.g., 5-15 Hz), this may correspond to the drill string scraping against the formed cuttings bed. Therefore, when this feature is matched, the system generates a base confidence component with a high confidence level for the "stable cuttings bed state".

[0087] Using this method, the real-time signal of each drilling condition parameter is decomposed and translated into a set of basic confidence components that can characterize its support for different wellbore cleanliness conditions.

[0088] Step 2: Weighted attenuation based on signal quality;

[0089] To address the dynamic changes in sensor signal quality under real-world operating conditions, the reliability of each basic confidence component needs to be assessed and corrected before evidence fusion. This step first calculates the signal quality factor of the original signal corresponding to each basic confidence component. This factor is a numerical value used to quantitatively evaluate the current reliability of the signal, and its calculation can be based on factors such as the signal-to-noise ratio, short-term variance, and packet loss rate of data transmission.

[0090] Subsequently, based on this signal quality factor, the system determines a weighted attenuation coefficient. This coefficient is positively correlated with signal quality; that is, the better the signal quality, the closer the coefficient is to 1, and the worse the signal quality, the closer the coefficient is to 0. Finally, this coefficient is applied to the corresponding base confidence component for weighted attenuation processing. The essence of this processing is to reduce the certainty of evidence originating from low-quality signals, transferring some of its confidence to the "uncertainty" term, thereby avoiding serious interference to the final fusion result caused by individual sensor malfunctions or distortions.

[0091] Step 3: Evidence fusion and convergence judgment;

[0092] This step involves fusing all the base confidence components, after weighted attenuation coefficient processing, using Dempster's combination rule, a core principle of evidence theory. This rule mathematically aggregates the confidence assignments from all independent sources of evidence, forming a comprehensive confidence distribution.

[0093] Because drilling conditions are continuously changing, the fusion process is performed at a fixed calculation cycle, for example, once per second. To ensure the stability of the output results, the system determines whether the comprehensive confidence distribution meets a preset convergence condition. One possible convergence condition is: determining whether the rate of change of the comprehensive confidence distribution is consistently less than a preset threshold within a consecutive, preset target number of calculation iteration cycles, for example, five consecutive calculation cycles. Here, the "rate of change" can be defined as the Euclidean distance or the maximum absolute difference between the probability distribution vectors calculated in two adjacent cycles.

[0094] Step 4: Extraction of stable probability values;

[0095] Once the overall confidence distribution meets the convergence condition, the fusion module considers the current discrimination result to be stable and reliable. At this point, the module extracts the final probability value corresponding to each preset wellbore cleaning state from the current stable overall confidence distribution, combines these probability values ​​to generate the first vector, and outputs it.

[0096] For example, to more clearly illustrate the fusion calculation process of the Dempster combination rule, a specific, but non-limiting, implementation example is provided below:

[0097] At a certain calculation moment in this embodiment, the preset wellbore cleaning status identification framework includes three basic states:

[0098] H1: High-efficiency cleaning status;

[0099] H2: Formation state of the rock debris bed;

[0100] H3: Stable cuttings bed condition;

[0101] Meanwhile, the fusion module has obtained the following two sets of basic confidence components from two different drilling state parameters after signal decomposition and weighted attenuation processing, referred to as Evidence 1 and Evidence 2:

[0102] Evidence 1 (derived from riser pressure analysis): This analysis indicates that annular friction is showing an increasing trend, but is not yet severe. Its basic confidence component (m1) is: 0.6 for the "clastic bed formation state (H2)"; the remaining part is an uncertain term with a support of 0.4. That is, m1(H2) = 0.6, m1({H1,H2,H3}) = 0.4.

[0103] Evidence 2 (from bottom-hole torque analysis): This analysis indicates that drill string rotation is slightly impeded. Its basic confidence components (m²) are: 0.3 for "cuttings bed formation state (H2)"; 0.2 for "stable cuttings bed state (H3)"; and the remaining portion is an uncertain term with a support of 0.5. That is:

[0104] m2(H2)=0.3, m2(H3)=0.2, m2({H1, H2, H3})=0.5.

[0105] Furthermore, the fusion module executes Dempster's combination rules, the core idea of ​​which is to seek "common support" among different pieces of evidence and strengthen this common support. This process can be understood computationally as follows:

[0106] First, construct a combination matrix or perform an equivalent cross-multiplication process to determine how all confidence components of the two evidence sources are combined. For example:

[0107] The combined support for H2 from Evidence 1 (0.6) and the support for H2 from Evidence 2 (0.3) resulted in this shared belief being assigned to H2.

[0108] The support of Evidence 1 for H2 (0.6) and the support of Evidence 2 for H3 (0.2) are combined. Since H2 and H3 are two mutually exclusive single states and their intersection is empty, this part of the belief is identified as a “conflict” between the evidences.

[0109] The support for H2 from Evidence 1 (0.6) is combined with the uncertainty from Evidence 2 (0.5), and this part of the belief is still assigned to H2 because the uncertainty can support any possibility.

[0110] Similarly, the uncertainty term (0.4) of Evidence 1 will be combined with each term (H2, H3, uncertainty term) of Evidence 2, and the beliefs will be assigned to H2, H3 and the uncertainty term itself, respectively.

[0111] Secondly, the system will aggregate all combination results. It will sum all belief values ​​assigned to the same state. Simultaneously, it will also sum all belief values ​​identified as "conflicting," forming a conflict coefficient.

[0112] Next, normalization is performed. The system subtracts the conflict coefficient from 1 to obtain a normalization factor. Then, the accumulated belief values ​​summarized in the previous step and assigned to each state (e.g., H2, H3) are divided by this normalization factor. The purpose of this step is to redistribute the beliefs of the conflicting parts proportionally to the non-conflicting items, so that the sum of the confidence levels of all items is finally equal to 1 again.

[0113] After the above fusion calculation, a new confidence distribution that integrates both types of evidence is obtained. Compared with the original result, the result has the following characteristics:

[0114] The confidence level for the “clastic bed formation state (H2)” will be significantly higher than the original 0.6 or 0.3, because both pieces of evidence support this state, and the belief is strengthened.

[0115] The system also assigns a non-zero confidence level to “Stable cuttings bed state (H3)”, which is based on the results of the torque analysis (Evidence 2).

[0116] The overall confidence level of the "uncertainty" will be significantly reduced because some of the uncertainty is eliminated by evidence during the fusion process and transformed into certain support for the specific state.

[0117] This specific example demonstrates that Dempster's combination rule can logically and consistently aggregate information from multiple sources, even those with inherent uncertainties, ultimately outputting a more reliable and informative comprehensive judgment result than any single information source. In practical applications, the fusion module combines all available evidence in pairs using this method until all evidence is fully fused.

[0118] Regarding the feature extraction module mentioned above:

[0119] The feature extraction module has its data input terminals connected to the output terminals of both the fusion module and the sensor group. If the fusion module solves the cognitive problem of "what is the wellbore state," then the core function of the feature extraction module is to further address the evaluation problem of "whether the current hydrodynamic conditions are effective in improving this state." This module uses a unique physical simulation and analysis method dynamically coupled with the system's cognitive results to ultimately output a second vector that quantitatively characterizes the current annular flow field's rock-carrying efficiency.

[0120] In its implementation, the module first constructs and solves an annular flow field calculation model that reflects the current real-world physical state of the wellbore, based on the received first vector and real-time drilling state parameters. This step does not employ a fixed, universal flow field model; instead, it adaptively adjusts key parameters of the calculation model, such as boundary and initial conditions, using the wellbore cleanliness probability distribution included in the first vector. Simultaneously, real-time drilling state parameters, such as flow rate, mud density, and drill string rotation speed, are used as input conditions to drive the numerical solution of the model.

[0121] In this way, each calculation performed by this module is a high-fidelity "virtual reproduction" for the current specific operating condition. The final result of this calculation is to generate a two-dimensional velocity vector field that describes the magnitude and direction distribution of fluid velocity on a key cross section of the annulus.

[0122] Secondly, this module performs a two-dimensional fast Fourier transform (2D-FFT) on the two-dimensional velocity vector field generated in the previous step, followed by vector decomposition-based post-processing. The purpose of this step is to transform the complex and difficult-to-interpret spatial velocity distribution information into the frequency domain for analysis. In the frequency domain, the flow field is decomposed into a collection of superimposed vortex structures of different scales. The output of this transformation is a frequency domain energy spectrum, which visually demonstrates how the kinetic energy of the flow field is distributed among vortices of different sizes.

[0123] Furthermore, this module extracts preset flow field mode characteristic parameters from the frequency domain energy spectrum, which can characterize the quality of the flow field mode. These parameters are key indicators for quantitatively evaluating the current flow field's rock-carrying efficiency.

[0124] For example, these characteristic parameters may include:

[0125] Low-frequency energy proportion: This parameter characterizes the proportion of energy occupied by large-scale vortex structures in the flow field. Generally, a high low-frequency energy proportion indicates the presence of an integral secondary flow structure in the flow field that can effectively entrain and transport the rock cuttings bed, which is an efficient rock-carrying mode.

[0126] Energy spectrum anisotropy: This parameter characterizes whether the energy distribution in the frequency domain is uniform. A high anisotropy value may indicate the formation of a well-directed, helical secondary flow, a flow pattern particularly effective for clearing cuttings beds on the underside of the wellbore.

[0127] Finally, the feature extraction module combines all extracted flow field mode feature parameters to form the second vector. Similar to the first vector, this second vector is also an array structure containing multiple numerical elements, each of which quantitatively describes a key dynamic feature of the current annular flow field mode. This second vector is then transmitted to the decision module as another key basis for its control decisions.

[0128] Optional, see Figure 3 The flowchart of a method for calculating the frequency domain energy spectrum of annular flow field provided in this application embodiment includes steps S201 to S203, wherein:

[0129] S201: Multiple baseline boundary models are pre-constructed, and each of the baseline boundary models corresponds to a wellbore cleanliness state in the wellbore cleanliness state probability distribution.

[0130] S202: Extract the probability values ​​from the first vector that correspond to various wellbore cleaning states, and set the probability values ​​as weighting factors;

[0131] S203: The weighting factor and the corresponding multiple benchmark boundary models are weighted and superimposed to generate an instantaneous equivalent boundary model.

[0132] In practical implementation, during a pre-implementation preparation phase, the system pre-builds and stores multiple baseline boundary models. Each baseline boundary model is an idealized physical description of a typical, single wellbore clean state; together, they form a model library, serving as the foundation for subsequent dynamic modeling. Each baseline boundary model includes a set of geometric and physical properties used for CFD calculations. For example:

[0133] The baseline boundary model corresponding to the "highly efficient and clean state" is defined geometrically as a smooth concentric or eccentric annular channel formed by the outer diameter of the standard drill pipe / drill collar and the inner diameter of the wellbore / casing. Its physical properties are defined as an impermeable solid wall with a baseline roughness value characterizing the smoothness of the casing wall.

[0134] The baseline boundary model corresponding to the "stable cuttings bed state" has the following geometric properties: based on the "highly efficient and clean state" model, it adds a solid region with a preset maximum thickness and shape (e.g., crescent shape) below the annulus to represent a fully developed cuttings bed. In terms of its physical properties, this solid region can be defined as a porous medium with specific porosity and permeability, while the remaining wall properties are the same as the previous model.

[0135] For example, a baseline boundary model corresponding to "severe wellbore collapse" can also be constructed, which has an irregular wellbore profile in geometry.

[0136] During system operation, the feature extraction module first receives the latest first vector from the fusion module in each calculation cycle. Then, it extracts the probability values ​​from the first vector and sets these probability values ​​directly as weight factors.

[0137] Next, the module performs a weighted superposition calculation on the various attributes of the multiple baseline boundary models to generate an instantaneous equivalent boundary model. This calculation process does not involve selecting from multiple models, but rather generates a completely new hybrid model that incorporates information from all possibilities. For example:

[0138] For the calculation of geometric properties: The equivalent thickness of the cuttings bed in the instantaneous equivalent boundary model is obtained by multiplying the cuttings bed thickness defined in each benchmark model (which is zero in the "highly efficient clean state" model) with the corresponding weight factor, i.e., the probability value, and then summing all the products.

[0139] For the calculation of physical properties: Similarly, physical properties such as equivalent permeability or equivalent roughness at a certain point on the boundary are also calculated by multiplying the corresponding physical property values ​​defined in each benchmark model with the corresponding weighting factors and then summing them.

[0140] In this way, the module ultimately generates an instantaneous equivalent boundary model. This model is not simply selected from a library, but is dynamically generated at each computational moment, integrating all current possibilities into a comprehensive and equivalent physical boundary description. This model provides subsequent numerical solutions to the flow field with geometric and physical inputs that closely match the current understanding of the system, and is the core component for achieving the overall system's adaptability.

[0141] Optionally, the calculation of the frequency domain energy spectrum of the annular flow field further includes:

[0142] The solid wall boundary conditions of the annular flow field computational domain are constructed based on the instantaneous equivalent boundary model, and the fluid properties and driving conditions of the annular flow field computational domain are set based on the drilling state parameters.

[0143] Under the combined constraints of the solid wall boundary conditions and fluid properties, the computational domain of the annular flow field is numerically solved to generate a two-dimensional velocity vector field;

[0144] The frequency domain energy spectrum is obtained by performing vector decomposition and two-dimensional fast Fourier transform on the two-dimensional velocity vector field.

[0145] This optional implementation constitutes a path to transform an abstract boundary model incorporating probabilistic information into concrete, quantifiable flow field characteristics.

[0146] In practice, the module first constructs the solid-wall boundary conditions of the annular flow field computational domain based on the instantaneous equivalent boundary model. This step concretizes the comprehensive model generated in the previous step into a series of boundary settings that can be recognized by the numerical solver.

[0147] For example, these solid wall boundary conditions are mixed and complex, and may include moving wall conditions for simulating drill string rotation or screw oscillation, equivalent porous media conditions for simulating the physical properties of cuttings beds, and slip or roughness conditions for simulating the interaction between the wellbore and the mud. Simultaneously, the fluid properties and driving conditions of the annular flow field computational domain are set based on the drilling state parameters.

[0148] Specifically, the module acquires mud density and rheological parameters, such as plastic viscosity and yield point, from real-time data provided by the sensor array, and selects a non-Newtonian rheological model that can characterize the shear thinning and yielding properties of drilling fluid to define the fluid properties within the computational domain. Furthermore, the real-time flow parameters are set as the inlet velocity or inlet mass flow rate of the computational domain, serving as the driving condition for fluid movement.

[0149] Secondly, under the combined constraints of the constructed solid wall boundary conditions and the set fluid properties and driving conditions, the module performs numerical solutions for the annular flow field computational domain. This solution process is executed using computational fluid dynamics (CFD) software commonly used in the field or a self-developed solver.

[0150] The solver iteratively calculates on a discretized computational grid to obtain the pressure and velocity distributions at every discrete point or cell within the computational domain that satisfy the governing equations of fluid dynamics, such as the Navier-Stokes equations. The solution process terminates when the computational residuals fall below a preset convergence criterion, and the module outputs a stable or transient two-dimensional velocity vector field. This vector field is a dataset that precisely describes the magnitude and direction of the fluid velocity at each point on a selected cross-section of the annulus.

[0151] Finally, vector decomposition and two-dimensional fast Fourier transform are performed on the two-dimensional velocity vector field. This step is central to the flow field mode analysis. Vector decomposition aims to process the velocity vector into component forms suitable for Fourier transform. The subsequent two-dimensional fast Fourier transform decomposes the spatial velocity field into a superposition of modes with different spatial frequencies, i.e., wavenumbers, ultimately yielding the frequency domain energy spectrum. This frequency domain energy spectrum can quantitatively describe how the kinetic energy of the flow field is distributed among vortex structures of different scales. This frequency domain energy spectrum is then used to extract the flow field mode characteristic parameters that form the second vector.

[0152] Optional, see Figure 4 The flowchart of a method for constructing solid wall boundary conditions for annular flow field computational domain based on the instantaneous equivalent boundary model provided in this application embodiment includes steps S301 to S302, wherein:

[0153] S301: Set the inner wall of the annular flow field calculation domain as a rotatable movable solid wall, which is used to simulate the effect of drill string rotation or screw oscillation on the near-wall flow field.

[0154] S302: Wherein, the rotation direction and equivalent angular velocity of the moving solid wall are derived from the probability weights corresponding to the clean state in the instantaneous equivalent boundary model.

[0155] In practical implementation, when constructing the annular flow field calculation domain, the inner wall of the annulus corresponding to the outer wall of the drill string is set as a rotatable, movable solid wall. This setting differs from the conventional approach of simplifying the inner wall as a static wall. Its purpose is to more realistically simulate the tangential driving effect exerted on the mud flow field adjacent to the wall by the rotation of the drill string or the oscillation of downhole power tools, such as screw drills, during actual drilling. The circumferential flow induced by this tangential driving effect, i.e., the circulation, plays an important physical role in entraining, stripping, and transporting cuttings that may be deposited on the lower side of the wellbore to the main fluid region.

[0156] A unique feature of this embodiment is that the rotational parameters of the moving solid wall, namely its rotational direction and equivalent angular velocity, are not simply taken from the real-time rotational speed of the drill string measured by the sensor array, but are dynamically obtained through a derivation process. This derivation process deeply couples the cognitive results of the feature extraction module (reflected in the instantaneous equivalent boundary model) with the parameter settings of the physical model.

[0157] For example, in a specific derivation process, the feature extraction module utilizes the probability weights corresponding to different wellbore cleanliness states contained within the instantaneous equivalent boundary model generated above to derive an "equivalent angular velocity" for numerical solution. Here, "equivalent" means that this angular velocity is not necessarily equal to the actual physical rotational speed of the drill string, but rather represents the most effective macroscopic momentum that the drill string rotation can transfer to the main fluid region under the current wellbore cleanliness state. For example, this derivation process can follow the following preset rules:

[0158] When the probability weight corresponding to the “highly efficient clean state” dominates in the instantaneous equivalent boundary model, the derived equivalent angular velocity is proportional to the drill string angular velocity measured by the sensor, and the proportionality coefficient is large, for example, close to 1, to simulate the physical process by which the rotational momentum of the drill string can be efficiently transferred to a clean, low-viscosity fluid.

[0159] Conversely, when the probability weight corresponding to the "stable cuttings bed state" is high, the derived equivalent angular velocity may be set to a value significantly lower than the measured drill string angular velocity. This is used to simulate the physical phenomenon that when part of the drill string is trapped or surrounded by high-concentration cuttings, most of its rotational energy is dissipated inside the viscous cuttings bed and fails to be effectively converted into momentum to drive the main fluid to rotate macroscopically.

[0160] The final equivalent angular velocity applied to the moving solid wall can be the result of a weighted average calculated from the angular velocities determined for each cleaning state according to the rules described above, based on the probability weights of each state. The rotation direction of the moving solid wall is usually set to be consistent with the actual rotation direction of the drill bit.

[0161] In this way, the inner wall rotation effect in the flow field calculation model can dynamically reflect the system's perception of the current wellbore cleanliness, making the entire simulation process closer to the real physical world.

[0162] Optionally, the solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model further include:

[0163] An equivalent porous medium boundary condition is set in the circumferential downward region of the annular computational domain to simulate the flow and permeation effect caused by solid deposition;

[0164] The equivalent thickness, porosity, and permeability parameters of the porous boundary are weighted according to a preset mapping rule based on the probability values ​​in the first vector corresponding to the solid deposition state on the lower side of the annulus.

[0165] The interface conditions between the porous boundary and the main fluid region are set to meet the boundary requirements of continuous normal velocity and continuous tangential stress.

[0166] This application further specifies additional solid wall boundary conditions for the region below the outer wall of the annular space. This setting brings a higher level of physical fidelity to the entire simulation model.

[0167] Specifically, to more accurately simulate the solid deposits, i.e., cuttings beds, formed on the lower side of the wellbore due to gravity settling in highly deviated or horizontal wells, this application sets equivalent porous media boundary conditions in the circumferential lower region of the annular computational domain. This method differs from the conventional approach of simplifying cuttings beds into newly added, impermeable solid walls.

[0168] Its technical advantage lies in the fact that real cuttings beds are not completely dense solids; they contain pores that allow for slow seepage flow of the slurry, and complex momentum exchange occurs at the interface between the cuttings bed and the main fluid. By employing a porous media model, these important micro-flow and seepage effects and their impact on the macro-flow field can be incorporated into the numerical solution.

[0169] The key physical parameters of the porous boundary, including its equivalent thickness, porosity, and permeability, are not fixed, but are dynamically and weighted according to the probability values ​​in the first vector corresponding to the solid deposition state on the lower side of the annulus, such as "rock bed formation state" and "stable rock bed state".

[0170] For example, the setup process can follow the following preset mapping rules:

[0171] Regarding the equivalent thickness: The equivalent thickness can be calculated by weighting the probability values ​​corresponding to the "stable cuttings bed state" in the first vector. The higher the probability value, the higher the certainty of the system's judgment that the cuttings bed exists, and the greater the thickness of the porous media region set in the model, and vice versa.

[0172] Regarding porosity and permeability: These two parameters can be set according to the probability ratio of "rock cuttings bed formation state" and "stable rock cuttings bed state" in the first vector.

[0173] For example, when the probability of a "clastic bed formation state" is high, a relatively high porosity can be set to simulate a loosely structured, recently deposited clast bed; when the probability of a "stable clast bed state" is high, a lower porosity can be set to simulate a clast bed that has undergone long-term compaction and has a denser structure. The permeability parameter can then be calculated based on the set porosity using empirical or semi-empirical formulas known in the art (e.g., the Kozeny-Carman equation).

[0174] Furthermore, to ensure proper coupling of the flow field between the porous medium boundary and the main fluid region, and to enable stable convergence of the numerical calculation, this embodiment also sets specific interface conditions between the two. Preferably, the interface conditions satisfy the boundary requirements of continuous normal velocity and continuous tangential stress.

[0175] Among them, the continuity of normal velocity ensures the conservation of mass of the fluid when crossing the interface; while the continuity of tangential stress (often called the Brinkman condition in fluid mechanics) ensures that the momentum transfer and frictional shear effects between the main fluid and the fluid in the porous medium region can be accurately simulated in accordance with physical laws.

[0176] Optionally, the solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model include:

[0177] Based on the first vector and the drilling state parameters, identify the area covered by filter cake in the annular circumference;

[0178] An anisotropic partial slip boundary condition is applied to the filter cake coverage area, and the tangential slip length parameter of the slip boundary is set according to the probability values ​​corresponding to the clean stable state and the disturbed resuspension state in the first vector by weighting.

[0179] Non-slip boundary conditions are applied to the remaining wall areas not identified as filter cake covered, and a circumferential segmented distribution of equivalent sand grain roughness is set according to the first vector to determine the shear stress distribution of the wall surface.

[0180] The focus of this alternative implementation is to simulate the frictional characteristics of different regions of the annular well wall, i.e., the outer wall, in a differentiated and dynamic manner.

[0181] In practical implementation, after setting boundary conditions for the porous media region on the inner and lower sides of the annulus wall, an optional embodiment of this application provides refined boundary condition settings for the remaining annulus wall regions to further improve the physical fidelity of the entire flow field calculation model. This setting can accurately simulate the complex frictional effects caused by differences in wellbore lithology and dynamic changes in the filter cake.

[0182] The setup begins with an identification step. The feature extraction module dynamically identifies areas with filter cake coverage in the annular circumference based on a first vector and drilling state parameters, such as formation permeability information provided in geological steering data. For example, the system can first identify the well sections and circumferential ranges where filter cake may form based on formation permeability data. Then, it determines whether the filter cake in these areas is intact based on the state probabilities associated with "wellbore stability" or "circulation patency" in the first vector.

[0183] For areas identified as having filter cake coverage, this embodiment creatively applies anisotropic partial slip boundary conditions.

[0184] Here, "partial slip" means that the model no longer uses the conventional "no-slip" condition that assumes the fluid velocity is zero at the wall, but instead allows the fluid to have a non-zero tangential velocity at the wall. This more realistically simulates the physical phenomena of mud flowing over a smooth, lubricating filter cake surface. "Anisotropy" further defines that this slip effect is different in different directions. For example, the degree of slip in the axial direction, i.e., along the wellbore direction, may be greater than the degree of slip in the circumferential direction, making the drag reduction effect in the mainstream scour direction more significant.

[0185] The key feature of this embodiment is that the tangential slip length parameter used to define the degree of slip is dynamically changing. The value of this parameter is weighted according to the probability values ​​corresponding to the "clean and stable state" and the "disturbed resuspended state" in the first vector.

[0186] For example, when the system determines that the wellbore is in a "clean and stable state", it will set a larger slip length, which means that the filter cake is intact and the lubrication effect is good; while when it determines that it is in a "disturbed re-suspension state", for example, after applying a pressure pulse, it will reduce the slip length to simulate that the filter cake surface is partially damaged, becomes rough, and friction increases.

[0187] For the remaining wall areas not identified as filter cake covered, such as well walls corresponding to dense, impermeable rock formations, this embodiment applies a non-slip boundary condition.

[0188] The shear stress distribution on the non-slip wall is not uniform, but is achieved by setting a circumferential segmented distribution of the equivalent sand grain roughness. This means that the model can simulate the roughness differences in different orientations of the well wall.

[0189] For example, the roughness of the upper side of the wellbore can be mainly determined by the lithology, while the area on the lower side of the wellbore near the cuttings bed can be given a larger roughness value to simulate the additional friction caused by a small number of solid particles adhering to the wall.

[0190] Similarly, this roughness distribution is also dynamically set by the first vector. The system can dynamically correct the reference roughness value of each segment based on the wellbore cleanliness reflected in the first vector, so that the shear stress distribution of the entire annulus wall can reflect the system's comprehensive understanding of the wellbore condition in real time.

[0191] By setting differentiated boundary conditions for different wall regions, all of which are dynamically coupled with system cognition (first vector), the flow field calculation model of the present invention can further improve physical realism.

[0192] Optionally, generating the two-dimensional velocity vector field includes:

[0193] Based on the pulse control parameters, the inlet and outlet driving conditions of the annular flow field calculation domain are set so that the volumetric flow rate or pressure at the inlet / outlet changes periodically with time according to the frequency and amplitude of the pulse control parameters.

[0194] At each time step, the instantaneous equivalent boundary condition model is mapped to a solid wall boundary condition for solving, wherein the solid wall boundary condition is at least one or more of the rotatable solid wall boundary, the equivalent porous medium boundary, and the anisotropic partial slip boundary and the circumferential segmented roughness boundary.

[0195] A non-Newtonian rheological model including the yield stress term is selected to characterize the mud fluid properties. Under the joint constraints of the solid wall boundary conditions and fluid properties, the transient pressure and velocity of the annular flow field computational domain are coupled to obtain the instantaneous two-dimensional velocity vector field on the selected well section cross section.

[0196] In a specific implementation, after constructing complete and highly complex solid wall boundary conditions, an optional embodiment of this application provides a specific method for numerically solving the annular flow field computational domain. This method is a transient solution method, the core purpose of which is to accurately simulate and analyze the actual effect of the pressure pulse generated by the pulse generator on the annular flow field.

[0197] In practical implementation, the first step is to set driving conditions that change dynamically over time. Unlike steady-state calculations using constant flow or pressure, a unique feature of this embodiment is that its driving conditions are set based on the pulse control parameters output by the decision module in the previous calculation cycle.

[0198] For example, the inlet volumetric flow rate of the annular flow field computational domain can be set as a function that fluctuates periodically over time. The baseline flow rate value of this function is determined by the real-time flow rate measured by the sensor array, while the frequency and amplitude of its fluctuations precisely correspond to the frequency and amplitude given by the pulse control parameters.

[0199] In this way, the numerical solution process can directly simulate the impulse control behavior applied by the system, forming a closed loop from "control output" to "physical simulation input".

[0200] Secondly, within each time step of the transient solution, the system dynamically maps the latest instantaneous equivalent boundary model to the solid wall boundary conditions used for that calculation. This means that all the complex boundary conditions mentioned above, including rotatable solid wall boundaries, equivalent porous media boundaries, and anisotropic partial slip boundaries and circumferential segmented roughness boundaries, are refreshed and applied at each time point based on the latest first vector. This ensures that the simulation model can capture the rapid changes in the wellbore state that may occur under pulse action.

[0201] Furthermore, to accurately characterize the fluid properties of drilling mud, this embodiment selects a non-Newtonian rheological model that includes a yield stress term. For example, the Bingham model or the more universal Herschel-Bulkley model can be used. The necessity of choosing such models lies in their ability to accurately simulate a key physical property of drilling fluids: fluid only begins to flow when the applied shear stress exceeds a specific "yield stress" threshold. This characteristic is crucial for accurately predicting "dead water zones" in the annulus, i.e., low-velocity or stagnant areas where cuttings easily accumulate.

[0202] Under the combined constraints of the aforementioned dynamically changing boundary and driving conditions, as well as precise fluid properties, the module employs a transient solver capable of handling pressure-velocity coupling problems to perform transient pressure and velocity coupling solutions on the annular flow field computational domain.

[0203] The solver synchronously calculates the pressure and velocity field distributions across the entire computational domain at each time step. Upon completion of the calculation, the module outputs an instantaneous two-dimensional velocity vector field, reflecting the flow field state at a specific characteristic moment under pulse action, such as the pressure peak or trough, on a pre-selected, representative well section cross-section. This vector field serves as the direct data basis for the feature extraction module to perform final frequency domain analysis and feature parameter extraction.

[0204] For example, a solver based on the Pressure-Implicit with Splitting of Operators (PISO) algorithm can be used. The advantage of choosing this algorithm is that it is a non-iterative transient calculation method, which can achieve high computational efficiency and stability while ensuring accuracy when dealing with time-dependent problems driven by pressure pulses.

[0205] Within a specific time step, the computation process of this PISO algorithm-based solver can be understood and executed as follows:

[0206] At the beginning of each transient computation time step, the solver first executes a prediction step. In this step, the solver temporarily ignores the influence of the pressure gradient term or directly uses the pressure field data from the previous time step to solve the momentum equation of the fluid dynamics. The result of this step is a temporary, intermediate velocity field. Because the coupling relationship between the pressure field and the velocity field is not fully satisfied in this step, this temporary velocity field usually does not satisfy the law of conservation of mass (i.e., the continuity equation).

[0207] To correct the problem that the velocity field does not satisfy mass conservation in the prediction step, the solver then performs one or more correction steps.

[0208] In a correction cycle, the solver first constructs and solves a pressure Poisson equation based on the mass conservation equation, resulting in a "pressure correction field." The physical meaning of this pressure correction field is that it represents what adjustments need to be made to the current pressure field so that the corresponding velocity field satisfies mass conservation.

[0209] The solver then uses the calculated pressure correction field to correct both the temporary velocity and pressure fields obtained in the prediction step. After the first correction, a new velocity field that satisfies the mass conservation constraint within the current time step is obtained. To further improve the accuracy and stability of the calculation, the essence of the PISO algorithm lies in its ability to continue executing one or more additional correction loops, i.e., repeatedly performing pressure and velocity correction calculations based on the updated velocity field.

[0210] Finally, after completing all prediction and correction calculations for the current time step, an accurate pressure and velocity field that satisfies all physical constraints (including momentum conservation, mass conservation, and all complex solid wall boundary conditions) is determined. The system then uses the pressure and velocity fields at the current moment as the initial conditions for calculations in the next time step, thus advancing one step forward on the time axis and repeating the entire "prediction-correction" process described above.

[0211] By advancing the time steps in this way, the dynamic evolution of the annular flow field over one or more pulse cycles can be obtained. Using a solver based on the PISO algorithm, the transient non-Newtonian fluid flow problem with complex dynamic boundary conditions driven by pulses, as described in this application, can be solved accurately with high computational efficiency and stability.

[0212] Furthermore, after the feature extraction module obtains the frequency domain energy spectrum of the annular flow field through transient solution and Fourier transform, the next step is to extract the flow field mode feature parameters from this energy spectrum and combine them to form a second vector. The purpose of this step is to reduce the dimensionality of the complex, high-dimensional frequency domain energy spectrum data and extract a few core indicators that can intuitively and quantitatively evaluate the quality of the current flow field mode.

[0213] For example, these characteristic parameters may include:

[0214] Low-frequency energy proportion: This parameter is obtained by calculating the ratio of the total energy within a predetermined low wavenumber range (corresponding to large-scale flow structures) in the frequency domain energy spectrum to the total energy of the entire spectrum. Physically, a high low-frequency energy proportion usually indicates the presence of dominant, large-scale vortices or secondary flow structures in the flow field. Such structures are advantageous for achieving overall stripping and efficient transport of cuttings beds.

[0215] Energy spectral anisotropy: This parameter can be obtained by calculating the dispersion or principal axis ratio of the energy distribution in different directions of the frequency domain energy spectrum. Physically, a high anisotropy value indicates that the vortex structure in the flow field has a clear directionality, such as a regular spiral progression rather than chaotic turbulence. This regular structure is more efficient at cleaning specific areas of the wellbore, such as the bottom side of a horizontal well.

[0216] Peak wavenumber: This parameter refers to the wavenumber corresponding to the point where the energy in the frequency domain energy spectrum reaches its maximum value. Physically, the reciprocal of the peak wavenumber is related to the characteristic scale of the most energetic vortex in the flow field. By comparing this characteristic scale with the average grain size of the rock cuttings, the effectiveness of the current flow model for transporting rock cuttings of a specific size can be evaluated.

[0217] After calculating one or more flow field mode feature parameters, the feature extraction module combines their values ​​into an array structure to form the second vector.

[0218] For example, the second vector can be specifically represented in the form of [low-frequency energy proportion, energy spectrum anisotropy, peak wavenumber].

[0219] Furthermore, the decision-making module is responsible for combining "perceptual cognition" (first vector) with "physical evaluation" (second vector) to make the final control action. The input of this module is the first vector generated by the fusion module and the second vector generated by the feature extraction module, and the output is pulse control parameters including frequency and amplitude.

[0220] This decision-making process is executed based on a pre-defined parameter mapping. For example, this mapping could be a rule base built upon expert knowledge.

[0221] Example of rule 1: If the first vector indicates a high probability of "cuttings bed formation state" and the second vector indicates a low "low-frequency energy proportion", this indicates that the wellbore condition is deteriorating and the current flow field pattern is very unfavorable for cleaning. In this case, the parameter mapping relationship will output a set of "high amplitude, low frequency" pulse control parameters, aiming to use high-energy pulses to impact and lift cuttings, and use low-frequency disturbances to induce large-scale vortices.

[0222] Example Rule 2: If the first vector indicates a high probability of "highly efficient and clean state," and the second vector indicates high "low-frequency energy proportion" and "energy spectrum anisotropy," this collectively indicates that the wellbore is in good condition and the current flow field mode is very effective. In this case, the parameter mapping relationship will output a set of pulse control parameters that "appropriately reduce amplitude or frequency," aiming to find the lowest energy consumption point that can maintain the current good state and achieve energy saving.

[0223] Besides being based on a rule base, the parameter mapping relationship can also be implemented in other ways, such as a multidimensional lookup table or an artificial neural network model trained on a large amount of offline simulation data.

[0224] Furthermore, the pulse generator is the final actuator of the system. It is located in the high-pressure section of the ground mud circulation pipeline, and its control terminal is electrically connected to the decision module. After receiving the pulse control parameters determined by the decision module, the device will precisely modulate the high-pressure mud flow based on these parameters to output a pressure pulse with corresponding frequency and amplitude.

[0225] In one specific embodiment, the pulse generator can be a rotary valve driven by a servo motor. The rotational speed of the servo motor is precisely controlled by a frequency converter according to input frequency parameters; the valve opening or the flow rate of a bypass circuit is adjusted according to input amplitude parameters. In this way, a stable high-pressure mud flow, after passing through the device, is modulated into a pulse flow carrying specific pressure fluctuations. This pressure pulse then propagates throughout the wellbore, acting on the bottom and annulus, achieving dynamic and optimized control of the wellbore cleanliness.

[0226] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A drilling mud circulation system, characterized in that, include: The sensor group is configured to collect drilling status parameters during the operation of the drilling rig, including at least one of drilling pressure, standpipe pressure, flow rate and mud density; The fusion module is configured as a clean state discrimination model based on evidence theory, which performs fusion calculation on the drilling state parameters to generate a first vector to characterize the probability distribution of the wellbore clean state. The feature extraction module is configured to calculate the frequency domain energy spectrum of the annular flow field based on the first vector and the drilling state parameters using two-dimensional fast Fourier transform and vector decomposition, and extract flow field mode feature parameters from the energy spectrum to form a second vector; The decision module is configured to determine pulse control parameters, including frequency and amplitude, based on a preset parameter mapping relationship and by combining the first vector and the second vector. A pulse generator is arranged in the mud circulation pipeline and configured to output pressure pulses with corresponding frequency and amplitude based on the pulse control parameters.

2. The drilling mud circulation system according to claim 1, characterized in that, The first vector generated to characterize the probability distribution of the wellbore cleanliness state includes: The real-time signal of each of the drilling state parameters is decomposed into basic confidence components corresponding to different wellbore cleanliness states; Based on the basic confidence component and the corresponding signal quality factor, determine the coefficients used to weight the attenuation of the basic confidence component; The Dempster combination rule is used to fuse all the basic confidence components after weighted attenuation coefficient processing to form a comprehensive confidence distribution; In response to the comprehensive confidence distribution satisfying a preset convergence condition, stable probability values ​​corresponding to each cleanliness state are extracted from the comprehensive confidence distribution to generate the first vector.

3. A drilling mud circulation system according to claim 2, characterized in that, Signal decomposition rules include: The real-time signal of the drilling status parameters is decomposed in the frequency domain to obtain at least one frequency component; The frequency components are matched with the preset wellbore cleaning state characteristic frequency bands to form the basic confidence components.

4. A drilling mud circulation system according to claim 2, characterized in that, The convergence condition includes: within a consecutive number of calculation iterations, the rate of change of the comprehensive confidence distribution is less than a preset threshold.

5. A drilling mud circulation system according to claim 1, characterized in that, The frequency domain energy spectrum of the calculated annular flow field includes: Multiple baseline boundary models are pre-constructed, and each of the baseline boundary models corresponds to a wellbore cleanliness state in the probability distribution of wellbore cleanliness state. Extract the probability values ​​from the first vector that correspond to various wellbore cleaning states, and set the probability values ​​as weighting factors; The weighting factor is weighted and superimposed with the corresponding multiple benchmark boundary models to generate an instantaneous equivalent boundary model.

6. A drilling mud circulation system according to claim 5, characterized in that, The frequency domain energy spectrum for calculating the annular flow field also includes: The solid wall boundary conditions of the annular flow field computational domain are constructed based on the instantaneous equivalent boundary model, and the fluid properties and driving conditions of the annular flow field computational domain are set based on the drilling state parameters. Under the combined constraints of the solid wall boundary conditions and fluid properties, the computational domain of the annular flow field is numerically solved to generate a two-dimensional velocity vector field; The frequency domain energy spectrum is obtained by performing vector decomposition and two-dimensional fast Fourier transform on the two-dimensional velocity vector field.

7. A drilling mud circulation system according to claim 6, characterized in that, The solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model include: The inner wall of the annular flow field computational domain is set as a rotatable movable solid wall, which is used to simulate the effect of drill string rotation or screw oscillation on the near-wall flow field. The rotation direction and equivalent angular velocity of the moving solid wall are derived from the probability weights corresponding to the clean state in the instantaneous equivalent boundary model.

8. A drilling mud circulation system according to claim 7, characterized in that, The solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model also include: Equivalent porous media boundary conditions are set in the peripheral lower region of the annular computational domain to simulate the flow and infiltration effect caused by solid deposition; The equivalent thickness, porosity, and permeability parameters of the porous boundary are weighted according to the probability values ​​in the first vector corresponding to the solid deposition state on the lower side of the annulus, and set according to a preset mapping rule. The interface conditions between the porous boundary and the main fluid region are set to meet the boundary requirements of continuous normal velocity and continuous tangential stress.

9. A drilling mud circulation system according to claim 8, characterized in that, The solid wall boundary conditions for constructing the annular flow field computational domain based on the instantaneous equivalent boundary model include: Based on the first vector and the drilling state parameters, identify the area covered by filter cake in the annular circumference; An anisotropic partial slip boundary condition is applied to the area covered by the filter cake, and the tangential slip length parameter of the slip boundary is set according to the probability values ​​corresponding to the clean stable state and the disturbed resuspension state in the first vector by weighting. Non-slip boundary conditions are applied to the remaining wall areas not identified as filter cake covered, and a circumferential segmented distribution of equivalent sand grain roughness is set according to the first vector to determine the shear stress distribution of the wall surface.

10. A drilling mud circulation system according to claim 9, characterized in that, The generation of the two-dimensional velocity vector field includes: Based on the pulse control parameters, the inlet and outlet driving conditions of the annular flow field calculation domain are set so that the volumetric flow rate or pressure at the inlet / outlet changes periodically with time according to the frequency and amplitude of the pulse control parameters. At each time step, the instantaneous equivalent boundary condition model is mapped to a solid wall boundary condition for solving. The solid wall boundary condition includes at least one or more of the following: a rotatable solid wall boundary, the equivalent porous medium boundary, and the anisotropic partial slip boundary and the circumferential segmented roughness boundary. A non-Newtonian rheological model including the yield stress term is selected to characterize the mud fluid properties. Under the joint constraints of the solid wall boundary conditions and fluid properties, the transient pressure and velocity of the annular flow field computational domain are coupled to obtain the instantaneous two-dimensional velocity vector field on the selected well section cross section.

Images