39 results about "Fluid modeling" patented technology

The invention belongs to the field of oil and gas field exploration and development, and discloses a numerical reservoir model optimization and history fitting method and related device.The method comprises the steps that firstly, geological exploration data are utilized, a spatial clustering algorithm is adopted to automatically determine the local updating range of a structural model, and structural layers and stratums are optimized through spatial interpolation; outputting an optimized construction model and uncertainty measurement; based on the model and logging lithofacies attribute data, adaptively identifying a lithofacies updating key area and variation function parameters by applying a machine learning algorithm, and performing lithofacies simulation to output an optimized lithofacies model; combining the porosity, permeability and saturation attribute data, adaptively determining an attribute optimization range and variation function parameters by adopting a probabilistic neural network, and outputting an optimized attribute model through phase control attribute modeling; a three-dimensional oil-water two-phase black oil seepage model is constructed, a global optimization algorithm is adopted to automatically adjust parameters based on a target function, simulation data is calculated until convergence, optimal parameters are output to complete historical fitting, and efficient and reliable intelligent support is provided for complex oil reservoir development.

The present application belongs to the field of image denoising, and in particular to a three-dimensional point cloud denoising method. The device and method are based on Laplace regularization subspace non-local low rank learning, and extend the previously proposed low-dimensional flow model of pixel blocks to surface blocks in point clouds. The graph Laplace regularizer of the surface is used to obtain the manifold dimension of the 3D point cloud image. Then, based on the principle that the high-dimensional information of the image point cloud is located in a low-dimensional subspace, the subspace non-local low rank factor is used to study the non-local self-similarity of the subspace, estimate the three-dimensional tensor generated by the non-local similar three-dimensional surface, and finally use the three-dimensional tensor to construct a denoising model to obtain a denoised image with better visual and quantitative indicators. The denoised image point cloud obtained by the method can better preserve the visually significant structural features, and the quantitative indicators are also good.

The application discloses a simulation and dispatching control method and system for urban flood process. The application belongs to the technical field of urban water conservancy system control. The method comprises the following steps: collecting and constructing a specific type of basic data set, respectively constructing a hydrological runoff model, a one-dimensional water power model of a drainage pipe network, a river network mixed water power model and an engineering dispatching model; performing three-dimensional dynamic coupling to form a water resource full-space dynamic model; using at least one historical flood event in historical flood data to calibrate parameters and verify accuracy of the water resource full-space dynamic model; after verification, using the water resource full-space dynamic model to build a digital simulation platform to simulate flood and determine a dispatching scheme. The technical scheme can realize multi-scenario dynamic simulation, automatic comparison and selection after generation of a dispatching scheme, enhance system adaptability and response timeliness under extreme weather, and provide an efficient and accurate flood control and drainage scheme generation mechanism for urban area flood control and drainage.

The present application provides a kind of gas film cooling efficiency calculation method based on k-epsilon turbulent flow model and mixing mass transfer function, first, establish gas film cooling efficiency calculation model;Then using the CFD solver embedded with k-epsilon turbulent flow model and mixing mass transfer function calculation module to the gas film cooling efficiency calculation model established in step 1 Iterative solution is carried out, the mass fraction distribution of secondary flow gas component is obtained;Finally, using post-processing software, according to the mass fraction distribution of secondary flow gas component obtained, the three-dimensional model wall surface adiabatic cooling efficiency distribution is calculated and obtained.The method introduces Gaussian three-parameter control mixing mass transfer function, with local secondary flow gas mass fraction as independent variable, and dynamically fine corrects the turbulent viscosity of mixing area through three empirical parameters, so as to significantly improve the simulation accuracy of complex mixing process, especially the simulation accuracy of gas film cooling flow field of special-shaped hole under the premise of not excessively increasing the calculation resources and complexity.

This invention discloses a two-way fluid simulation system for side-mounted inlets / outlets based on the RSM turbulence model. Belonging to the field of energy and power engineering, the system includes: a global geometric modeling module for constructing a coupled geometric model encompassing the tailrace tunnel and side-mounted inlets / outlets; a physical parameter configuration module for setting flow and pressure boundaries under two-way flow conditions; a mesh generation module for structured mesh generation and local refinement of the model; an RSM solution module for closed-loop solution of the two-way flow field using the Reynolds stress model to obtain state data; and a multi-index evaluation module for extracting and analyzing the flow regime within the diffuser section, the velocity distribution of the trash rack, flow distribution, and turbulence parameters. This invention can systematically evaluate the differences in hydraulic characteristics of different vertical expansion methods in the diffuser section, providing a reliable scientific basis for the design of the inlet / outlet shape of pumped storage power stations.

The present application relates to a kind of solid-liquid two-phase flow erosion life prediction method of submersible electric pump wide flow condition, it includes: constructing the full parameterization fluid domain grid model of submersible electric pump impeller guide vane;Set wide flow discrete working condition sequence, obtain the unsteady flow field data under the multiple discrete flow working condition section covered to;By constructing the two-way coupling discrete phase-turbulent two-phase flow model coupled Saffman lift correction, extract the transient parameter when particle impact impeller guide vane wall;Flow correlation dynamic hardness index function is built;Determine the instantaneous erosion rate of each working condition section in combination with Saffman lift correction factor;Obtain the total erosion damage degree of each node on impeller guide vane wall after experiencing total flow working condition section;The erosion life of impeller guide vane and each sub-region is predicted.The present application introduces flow correlation dynamic hardness index and dynamic velocity index, and reduces the simulation prediction error of erosion rate under variable condition to 12% or less from more than 30% of traditional method.

The application discloses a kind of gas-liquid two-phase flow metering method based on pipeline internal waveguide radar liquid level measurement, is collected uniform speed pipe flowmeter monitored gas phase instantaneous flow and waveguide radar liquid level instrument monitored instantaneous liquid level high point;According to instantaneous liquid level high point, instantaneous liquid holdup is obtained by calculating through circle defect area formula;Based on the average liquid holdup in T period 、 Maximum liquid holdup 、 Liquid holdup fluctuation amplitude and average gas phase apparent flow velocity, determine wet gas flow regime as slug flow, stratified flow, wave flow, annular flow or transition flow;Based on the drift flow model, calculate liquid instantaneous flow, the distribution coefficient and drift velocity in the drift flow model are valued according to different flow regimes;Respectively accumulate liquid instantaneous flow and gas instantaneous flow, obtain liquid total amount and gas total amount.It can effectively improve the detection precision and stability, the equipment cost is low and easy to install, the requirement for straight pipe section is low, especially suitable for wellhead, unattended station and other space limited scenes.

The present application relates to the technical field of virtual simulation and fluid modeling, in particular to a sea current prediction and visualization method based on digital twinning, comprising the following steps: S1, setting a boundary water level and a terrain elevation to calculate an initial flow velocity field, and generating a projection velocity field satisfying an incompressible condition through projection correction; S2, generating a disturbance mapping by combining a difference value of the projection field and the initial field, a terrain gradient and a boundary change rate, and marking a potential visual discontinuous area; S3, constructing a terrain-sensitive disturbance factor to modulate a velocity input, driving a wave and particle system, and realizing smooth transition and consistent evolution of a physical field and a visual rendering system. The present application realizes the fusion effect of high consistency between physical precision and visual performance of sea current simulation under complex terrain by constructing a disturbance-modulated physically consistent velocity field and driving the wave and particle system to evolve cooperatively.

This invention discloses an adaptive turbulence simulation method suitable for calculating the flow separation of underwater vehicles, the steps of which are as follows: k‑e The turbulence model stress is remodeled, and the subgrid-scale turbulence stress tensor is obtained through Reynolds stress attenuation. The resolution control function is improved based on the turbulence energy spectrum. The filtering scale coefficients in the resolution control function are calculated, and the magnitude of the resolution control function is further calculated. An adaptive transition between unsteady RANS, LES, and DNS turbulence models is achieved. The adaptive turbulence simulation modeling process is completed. Unsteady calculations are performed on the flow around underwater vehicles. The method of this invention can accurately simulate and predict the flow around underwater vehicles at large angles of attack and flow separation phenomena under relatively coarse computational grids.

This application discloses a method and system for real-time inference and parameter inversion of gravitational wave events based on deep learning. The method includes: constructing a physics-instrument dual-manifold decoupled network; extracting physics signal features and instrument noise features from time-frequency spectra; introducing physical manifold constraints and orthogonally decoupling the features in the latent space; performing parameter inversion on the decoupled physics features using a conditionally normalized flow model; and verifying the consistency of the inference results by regenerating the waveform in reverse and analyzing the residuals. This application distinguishes between high-eccentricity gravitational wave signals and detector transient noise, improving recognition accuracy and reducing false alarm rate.

The application provides a blood flow model generation method, device and equipment and a storage medium. The method comprises the following steps: obtaining three-dimensional angiography data, constructing a three-dimensional blood vessel model based on the three-dimensional angiography data; obtaining two-dimensional angiography data, simulating the flow of contrast agent in the three-dimensional blood vessel model based on the two-dimensional angiography data, and obtaining a simulation blood flow model; projecting the simulation blood flow model in the projection direction of the two-dimensional angiography data to obtain two-dimensional projection data; calculating the difference between the two-dimensional projection data and the two-dimensional angiography data, iteratively adjusting the model parameters of the simulation blood flow model based on the difference until the difference meets a preset convergence condition; determining the contrast agent concentration distribution weight of each position at different times in the iteratively adjusted simulation blood flow model; and based on the concentration distribution weight, projecting the gray scale change of the contrast agent at different times in the two-dimensional angiography data to each position of the three-dimensional blood vessel model to generate a target blood flow model.

This invention proposes a simulation optimization method and system for gas-liquid two-phase flow fields in fixed dredging module pipelines, belonging to the field of simulation optimization technology for gas-liquid two-phase flow fields in dredging. Addressing the problems in existing technologies where the determination of aeration parameters largely relies on empirical values, leading to unstable dredging effects, high energy consumption, and a lack of dynamic control over aeration parameters under different flow conditions, this invention conducts multi-condition numerical simulations using a gas-liquid two-phase flow model. With the objectives of maximizing the disturbance range and minimizing unit energy consumption, multi-dimensional disturbance indices are extracted and combined with unit energy consumption indices to construct a multi-objective optimization evaluation function. A comprehensive performance score is calculated using a combination of Pareto front screening and congestion distance scoring. Closed-loop iterative correction is performed through physical model verification experiments, achieving dynamic adjustment of aeration parameters based on real-time flow velocity monitoring. This invention significantly improves the energy efficiency ratio, accuracy, and reliability of aeration parameters in dredging operations, while reducing reliance on manual operation and maintenance.

This application discloses a method and system for visualizing flood disaster situation projection, belonging to the field of computer technology. The method includes: acquiring multi-source environmental parameters and micro-topographic data of a substation; classifying and identifying areas such as cement floors, gravel paved areas, cable trench covers, and equipment foundations based on lidar point clouds; constructing a physical model of water accumulation in the substation's micro-topography; and building a phased dynamic inflow model based on head difference for cable trench covers, including a contact immersion stage based on the weir flow formula, a transitional orifice flow stage based on a weighted average of weir and orifice flows, and a pressure full-flow stage based on the pressurized pipe flow formula; performing drainage pump power verification and equipment matching based on the projection results to generate a flood drainage resource deployment plan; and generating a dynamic path planning scheme based on real-time water accumulation depth and vehicle wading limits, and visually annotating it in a 3D digital twin scene, thereby achieving refined water accumulation calculation and dynamic resource scheduling for flood disasters in substations.

The application discloses a kind of gas-solid two-phase flow self-adapting simulation prediction method, system, medium and product, belong to the field of digital simulation of fluid dynamics, the method is: obtaining industrial equipment geometry, particle and fluid data, constructs gas-solid two-phase flow CMPPIC model;Wherein, continuous phase is with Euler method, discrete phase is with Lagrange particle bag representation;Perform time advance iteration to termination condition, output result, wherein each round of iteration includes mapping particle bag data to Euler cell, obtaining particle statistical information and marking fixed state particle according to rules;Freeze fixed cell particle volume fraction, calculate porous medium parameter, reconstruct its continuous phase momentum conservation equation;Based on reconstruction and unreconstructed equation, update flow field, while updating the particle motion of unmarked fixed state cell. By implementing the application, the problem that calculation efficiency and physical precision are difficult to consider in simulating fixed particle structure in the prior art can be solved.

This invention discloses a method for optimizing the structure of a heating furnace burner based on a fluid-thermal coupling model, belonging to the field of burner structure optimization technology. To provide an analysis and optimization method that can accurately quantify the comprehensive influence of burner structural parameters on the internal flow field, temperature field, and component field of the furnace, the method specifically includes the following steps: Establishing a three-dimensional physical model of the heating furnace burner; using a symmetrical modeling method, regularly arranging and connecting furnace tubes and U-shaped bends to the radiant manifold to form a radiant coil model; this method uses turbulence models, combustion models, and NOx generation models to simulate the internal flow field, temperature field, and component field. Through simulation analysis of the excess air coefficient, air preheating temperature, and heat load, the influence of different parameters on the combustion process and NOx emissions can be obtained, making the burner control more precise and effectively reducing NOx emissions while ensuring the safety of the tube wall temperature.

The present application belongs to the technical field of oil and gas field development, and particularly relates to a gas-liquid two-phase nozzle flow simulation method based on lattice Boltzmann, which comprises the following steps: S1, constructing a gas-liquid two-phase nozzle flow model, wherein the gas-liquid two-phase nozzle flow model adopts a single-component two-phase pseudo-potential model; S2, simulating the real-time fluid dynamics of the gas phase and the liquid phase in the space before and after the oil nozzle through the gas production rate, the liquid production rate, the production pressure difference and the oil nozzle size, and outputting the pressure, the density and the velocity data at any position in the space. The lattice Boltzmann method has excellent mobility and wide applicability, and by constructing the model in equal proportions and introducing the multiphase lattice Boltzmann method and the boundary processing mode, the real-time dynamic changes of the gas-liquid two-phase before and after the oil nozzle can be simulated.

The application discloses a k-epsilon turbulent flow model enhanced mixing correction method suitable for air film cooling, provides a mixing function C for describing two different state fluids for the k-epsilon turbulent flow model m , so as to improve the fluid mixing prediction situation of the mixing area of the two fluids, and corrects the mixing function C m of the turbulent flow viscosity coefficient μ of the different mixing degree areas t , realizes the correction of the different mixing intensity areas of the two fluids, and improves the numerical prediction accuracy of the air film cooling efficiency mixing area.

This invention discloses a numerical simulation method and apparatus for Microcystis blooms coupled with multiple hydrodynamic effects, comprising the following steps: S1, collecting basic data required for constructing a Microcystis bloom model; S2, establishing a fluid dynamics model based on a preset turbulence model, and solving the fluid dynamics model based on preset boundary conditions; S3, constructing a Microcystis dynamics model based on the response relationship between hydrodynamic elements and Microcystis population migration, buoyancy transport, growth, death, population size, and cell density, with governing equations including convection-diffusion equations, Microcystis vertical migration equations, Microcystis growth rate equations, Microcystis death rate equations, population diameter dynamic equations, and cell density dynamic equations; S4, coupling the fluid dynamics model and the Microcystis dynamics model, and obtaining the simulated concentration distribution results of Microcystis through parameter calibration and verification; This invention can fully reflect the key role of hydrodynamic elements in the bloom process, providing reliable support for bloom prediction and forecasting.

The application discloses an impeller inlet edge optimization method for improving the dead point head of a centrifugal pump, and comprises the following steps: S1, determining a basic size of the impeller; S2, making an axial plane projection distribution map according to the basic size; S3, determining a fluid dynamics control equation; S4, constructing a model for calculating a fluid domain, which comprises an inlet section, the impeller and a volute; S5, adopting corresponding mesh division modes for different components and locally encrypting; S6, adopting CFX-Pre software to perform steady-state calculation, and selecting a Shear Stress Transport turbulent flow model; S7, performing mesh independence verification; and S8, judging whether the design requirement is met or not, and if not, modifying the blade inlet edge and repeating the relevant steps, so that the dead point head is effectively improved, the flow field distribution in the flow passage is improved, the cavitation phenomenon is relieved, the stable operation of the centrifugal pump under full working conditions is ensured, and a reliable research scheme reference for improving the dead point head of the centrifugal pump is provided.

This invention discloses a hydraulic analysis method for the start-up and shutdown process of an impulse turbine in the field of hydropower station energy dispatching technology. By acquiring the unit's structural parameters and operating conditions, a three-dimensional fluid computational domain is constructed, including nozzles, deflectors, jet regions, and runner regions, and basic flow control equations are established. Based on this, a turbulence model is introduced to correct the flow characteristics, a gas-liquid two-phase flow model is introduced to regulate the phase interface evolution, and wall functions are combined to correct near-wall flow. Furthermore, the influence of the weak compressibility of water on pressure propagation is considered through density functions, thereby constructing a numerical analysis model coupled with multiple physical factors. This invention can improve the accuracy of hydraulic characteristic analysis, making the analysis results closer to actual operating conditions, and providing a basis for unit operation regulation and performance evaluation.

A method for predicting cavitation intensity within the runner of a large-scale hydro-turbine unit includes the following steps: simplifying the gas-liquid two-phase flow within the entire runner channel under typical operating conditions; extracting a complete three-dimensional geometric model of the runner channel, generating a computational mesh, and verifying mesh independence; configuring the solver and computational mode, enabling the multiphase flow model and the turbulence model, and setting boundary conditions, flow field initialization methods, and computational convergence criteria; extracting impurity parameters and arranging discrete phase particles representing impurities in the water; simulating the movement of particles within the runner channel using the discrete phase model to obtain the particle distribution and mass concentration within the channel; calculating the partial pressure of insoluble gases to quantitatively characterize the influence of water quality; calculating the corrected vaporization pressure; obtaining the cavitation occurrence location within the channel and quantitatively evaluating the cavitation intensity; and comparing the cavitation intensity prediction results considering and not considering water quality. This application can improve the prediction accuracy of hydro-turbine cavitation intensity.

The application discloses a simulation analysis method for aerodynamic characteristics of a high-speed rocket sled-track system, and establishes a mathematical model and a finite element model of the rocket sled-track system, obtains the dynamic response of the rocket sled through numerical solution, and performs comparison and verification; a fluid domain is constructed according to the geometric shape of the rocket sled and three-dimensional characteristics of a flow field, global grid division of the flow field of the rocket sled is performed, and a finite element model is constructed; an RNG k-ɛ turbulent flow model is adopted, an inlet flow velocity and an outlet pressure boundary are set, aerodynamic drag and its coefficient acting on the rocket sled are analyzed and obtained, and aerodynamic load input is provided; a coupled dynamic model of the rocket sled-track system is constructed, a rigid-flexible coupled model of the sled-track system of a flexible track structure is established; track irregularity numerical excitation is reconstructed, and data of track irregularity of the rocket sled-track system are obtained; initial conditions and boundary conditions are set, different working conditions of the rocket sled under different Mach numbers with or without aerodynamic drag load are simulated, and the motion parameters of the mass center of the rocket sled are obtained. The application provides an analysis method for studying the influence of aerodynamic drag on the dynamic response of the rocket sled-track system.

A computational method for investigating the gas compression and temperature rise phenomenon in the main steam pipe of a nuclear reactor is proposed. This method is applicable to simulating the abnormal temperature and pressure rise in the secondary steam pipe of the steam generator caused by the loss of secondary feedwater and the closure of the main steam isolation valve in a full-pressure small reactor. The method mainly includes the following parts: 1. Simplifying the geometry of the secondary steam main pipe and constructing a 3D model; 2. Using SpaceClaim software to perform 3D modeling of the simplified steam main pipe structure; 3. Using the mesh function of Fluent software to generate a computational mesh for the model; 4. Establishing suitable compressible flow and turbulence models; 5. Defining boundary conditions and initial conditions; 6. Setting the solver and solution method, and setting the time and spatial discretization formats; 7. Adjusting the sub-relaxation factor and setting the time step that conforms to the computational time scale; 8. Setting the convergence criterion and monitoring physical quantities such as cross-sectional temperature and inlet pressure until the pressurization process in the 3D model ends and the container reaches a steady state.

This invention relates to the field of virtual simulation and fluid modeling technology, specifically to a digital twin-based method for ocean current forecasting and visualization, comprising the following steps: S1, setting boundary water levels and terrain elevations to calculate an initial velocity field, and generating a projected velocity field that satisfies the incompressibility condition through projection correction; S2, combining the difference between the projected field and the initial field, terrain gradient, and boundary change rate to generate a perturbation map, marking potential visual discontinuities; S3, constructing a terrain-sensitive perturbation factor to modulate the velocity input, driving the wave and particle systems, and achieving a smooth transition and consistent evolution between the physical field and the visual rendering system. This invention, by constructing a perturbation-modulated physically consistent velocity field and driving the co-evolution of the wave and particle systems, achieves a highly consistent fusion effect between the physical accuracy and visual representation of ocean current simulation under complex terrain.

This invention proposes a simulation optimization method and system for gas-liquid two-phase flow fields in fixed dredging module pipelines, belonging to the field of simulation optimization technology for gas-liquid two-phase flow fields in dredging. Addressing the problems in existing technologies where the determination of aeration parameters largely relies on empirical values, leading to unstable dredging effects, high energy consumption, and a lack of dynamic control over aeration parameters under different flow conditions, this invention conducts multi-condition numerical simulations using a gas-liquid two-phase flow model. With the objectives of maximizing the disturbance range and minimizing unit energy consumption, multi-dimensional disturbance indices are extracted and combined with unit energy consumption indices to construct a multi-objective optimization evaluation function. A comprehensive performance score is calculated using a combination of Pareto front screening and congestion distance scoring. Closed-loop iterative correction is performed through physical model verification experiments, achieving dynamic adjustment of aeration parameters based on real-time flow velocity monitoring. This invention significantly improves the energy efficiency ratio, accuracy, and reliability of aeration parameters in dredging operations, while reducing reliance on manual operation and maintenance.

The application relates to a turbocharger flow field prediction method based on an improved turbulence model, which comprises the following steps: selecting a target turbulence model, screening the closure coefficients of the target turbulence model, and obtaining a design space; in the design space, obtaining the closure coefficient combination samples of the target turbulence model and corresponding macroscopic aerodynamic performance parameters and microscopic flow field parameters, and training a proxy model; using the trained proxy model, adopting a multi-objective optimization algorithm to perform optimization in the design space, obtaining an optimized closure coefficient combination, improving the target turbulence model, and obtaining an improved turbulence model; and using the improved turbulence model to perform numerical simulation on the internal flow field of a target turbocharger, outputting aerodynamic performance parameters and flow field structures, and realizing the prediction of the flow field of the target turbocharger. The application realizes high-precision prediction of the complex flow field of the turbocharger, the prediction method takes into account the overall performance and local flow field precision, and has certain generalization ability.

This invention provides an early warning method for slag runoff during steel ladle casting, comprising: 1) acquiring the structural and dimensional parameters of the steel ladle, establishing a three-dimensional model of the steel ladle, and performing mesh generation; 2) acquiring the casting speed, the initial liquid level of the molten steel in the ladle, and the trajectory of the sliding gate opening during casting under the given casting speed, and setting the dynamic mesh for the sliding gate using a dynamic layering method; 3) selecting the RNG k-ε turbulence model and the VOF model, and setting boundary conditions; 4) tracking the interface between the molten steel, slag, and air in the ladle; 5) obtaining slag runoff early warning conditions based on the tracking results and issuing a slag runoff early warning. This invention can realistically simulate the changing process of the sliding gate, improve the accuracy of model calculations, and significantly improve early warning efficiency.

The application provides an ultrasonic ranging interference error prediction method based on an uncertainty perception double-flow model. The spatial position of an ultrasonic emission device and an ultrasonic ranging sequence are extracted as space-time features to construct a double-flow space-time network. The model simultaneously outputs predicted results and two kinds of uncertainties by combining a Bayesian neural network, thereby achieving high robustness and ensuring that the method is more reliable when migrated to different external ultrasonic interference environments. The double-flow neural network simultaneously captures space-time features and fuses predictions by using a long short-term memory network and a double-separated attention graph neural network, thereby comprehensively learning the influence of space-time attributes of the physical world on the prediction accuracy of ultrasonic ranging. The Bayesian network variational inference part introduces a mixed prior, thereby accelerating the convergence speed of the variational distribution and the real distribution in the model training process and realizing faster and more efficient training.

This invention discloses a method for simulating leakage in oil pump gaps based on a Bayesian deep neural network, belonging to the field of oil extraction equipment simulation technology. It pre-constructs a leakage mechanism model for oil pump gaps, calibrating the total leakage as a linear superposition of pressure differential-driven flow and shear-draggled flow, and considering the coupled effects of reduced mixing viscosity and interfacial shear lifting under high gas-liquid ratio conditions. This invention constructs a simulation framework integrating an equivalent homogeneous flow model and a VOF multiphase flow model, clarifying the lifting and suppression mechanism of the annular flow pattern within the gap and the gas-liquid phase slip under high gas-liquid ratios, explaining the core reason why traditional models overestimate leakage, and providing a solid physical basis for leakage prediction. The proposed Bayesian deep neural network surrogate model, through large-scale sample training and hyperparameter optimization, achieves high-precision prediction under medium-low gas-liquid ratio conditions, with a computational speed far superior to CFD numerical simulation, solving the problem of balancing accuracy and efficiency in existing methods.