A method for establishing a long-term model of the transport and diffusion of submarine leaked carbon dioxide in seawater

Abstract

The present invention provides a method for establishing a long-term model for the transport and diffusion of submarine carbon dioxide leaks in seawater, belonging to the field of marine pollution technology. The method first constructs a near-zone model to calculate the bubble number density and gas-liquid mass transfer. The near-zone results are then used as boundary conditions for the far-zone model to establish a diffusion matrix system. The far-zone hydrodynamic field is then constructed based on the FVCOM hydrodynamic model and continuity equations, and a transport-diffusion optimization function is introduced to handle multi-scale coupling. The spatiotemporal coupling of the near-zone and far-zone models is achieved through the principle of stepped scaling. A scale-adaptive function is used to dynamically adjust the calculation parameters, taking into account factors such as the bubble diameter change rate and the dissolution rate gradient. Finally, the OceanDiffNet model is introduced to optimize far-zone diffusion prediction. This model uses a hierarchical attention mechanism to integrate features at different scales, and improves long-term prediction accuracy and computational efficiency by dynamically adjusting the hierarchical fusion weights.

Description

[0001] The present invention belongs to the technical field of marine pollution, and in particular relates to a method for establishing a long-term model of the transport and diffusion of submarine leaked carbon dioxide in seawater. Background Art

[0002] As an important technology to address global climate change, seabed carbon storage may face challenges in practical applications. Risk of leakage. Traditional The leakage diffusion model is mainly divided into two categories: near-zone model and far-zone model. The near-zone model mainly focuses on the area near the leakage source. The microscopic dynamic processes of bubble dissolution, breakup and rise are simulated with high precision using computational fluid dynamics. The far zone model focuses on the post-dissolution process. The transport and diffusion processes in large-scale oceans are usually predicted macroscopically based on ocean hydrodynamic models. The leakage diffusion model has obvious defects in dealing with problems: on the one hand, the near-zone model and the far-zone model are often separated from each other, lacking an effective spatiotemporal coupling mechanism, resulting in discontinuous cross-scale information transmission; on the other hand, the existing model oversimplifies the treatment of bubble dynamics parameters and seawater chemical reaction processes, making it difficult to accurately capture the dynamics of the bubble under different depths, temperatures, and pressures. Long-term dynamic changes in transport and diffusion. Especially in the long-term prediction, due to the complexity and multi-scale characteristics of the marine environment, the diffusion model in the existing technology cannot effectively realize the dynamic adaptive adjustment of different time and space scales when dealing with the transformation of near-area microscopic processes to far-area macroscopic diffusion, resulting in insufficient accuracy of long-term prediction results and difficulty in meeting the needs of seabed carbon storage safety assessment. In other words, there are problems with seabed leakage in the existing technology. Technical problems with insufficient long-term prediction accuracy of multi-scale transport and diffusion processes in seawater. Summary of the Invention

[0003] In view of this, a method for establishing a long-term model of the transport and diffusion of submarine leaked carbon dioxide in seawater is proposed to solve the existing problems of submarine leaks in the existing technology. Technical problems with insufficient long-term prediction accuracy of multi-scale transport and diffusion processes in seawater.

[0004] The present invention is achieved as follows: a method for establishing a long-term model of the transport and diffusion of submarine leaked carbon dioxide in seawater, comprising: establishing a submarine leaked carbon dioxide transport and diffusion model; The transport and diffusion near-zone model in seawater is used to calculate the bubble number density through the Euler method and the particle group equilibrium model, and the first time granularity is used for near-zone Gas-liquid mass transfer calculation; using the calculation results of the near-zone model as the input boundary conditions of the far-zone model, establish the positive diffusion matrix and the negative diffusion matrix; construct the far-zone transport and diffusion model at the second time granularity, and use the FVCOM hydrodynamic model to simulate the hydrodynamic conditions of large-scale ocean areas; combine the near-zone simulation results with the far-zone hydrodynamic field to establish the diffusion stability matrix and the diffusion variation matrix; according to the step-scaling principle, the near-zone and far-zone models are spatiotemporally coupled; the scale adaptive function is used to optimize the model to achieve accurate simulation of different diffusion stages; introduce the OceanDiffNet diffusion prediction model to the far-zone The transport and diffusion process is optimized to improve the accuracy and computational efficiency of long-term diffusion prediction.

[0005] The first time granularity refers to the small time step used in the near-zone model calculation, which is used to accurately capture the In the fast changing processes such as bubble dissolution, breakup and rise, the first time granularity is determined by the bubble rising velocity and the grid size, which satisfies the Courant condition to ensure numerical stability.

[0006] The second time granularity refers to the large time step used in the calculation of the far-field model, which is used for long-term simulation. In the transport and diffusion process in seawater, the second time granularity is determined by the ocean current velocity and the far-field grid size, ensuring the accuracy and computational efficiency of the numerical solution.

[0007] The forward diffusion matrix is ​​described as The mathematical expression of the diffusion propagation forward in the horizontal direction with the ocean current includes the horizontal eddy viscosity coefficient and the diffusion coefficient. The forward diffusion matrix is ​​solved using a hyperbolic partial differential equation, taking into account the coupling effect of convection and diffusion terms.

[0008] The negative diffusion matrix is ​​a matrix describing The negative diffusion matrix is ​​a mathematical expression of the upward propagation in the vertical direction due to gravity and buoyancy or the downward sedimentation due to density differences. It is solved using a parabolic partial differential equation, which takes into account the combined effects of gravity, density differences and vertical turbulent mixing.

[0009] The diffusion stability matrix refers to the

[0010] The mathematical expression of diffusion changes is affected by stable factors such as seawater temperature, salinity, and pressure. The diffusion stability matrix is ​​constructed through the eigenvalue decomposition method, which reflects the behavioral characteristics of the diffusion system under steady-state conditions.

[0011] The diffusion variation matrix refers to the dynamic hydrodynamic conditions The mathematical expression of diffusion change is affected by variable factors such as tides, ocean currents, and wind fields. The diffusion change matrix is ​​constructed using time-varying coefficients, which can capture the dynamic response characteristics of the system under non-steady-state conditions.

[0012] Among them, the ladder scale refers to a method of connecting models of different scales through spatiotemporal coupling to realize the information transmission from small-scale near-zone models to large-scale far-zone models. The ladder scale adopts nested grid and boundary condition transmission technology to ensure the consistency and continuity of information between models of different scales.

[0013] The scale adaptive function is used to adjust the seabed leakage The bubble evolution characteristics dynamically adjust the model calculation scale. The input includes the bubble diameter change rate, dissolution rate gradient, flow field variation coefficient, temperature stratification intensity, and salinity gradient. The output is the optimal spatiotemporal scale parameter set.

[0014] Among them, the specific structure of the OceanDiffNet model is a deep learning network architecture based on the hierarchical attention mechanism, which includes three core parts: spatial feature extraction module, temporal feature extraction module, and hierarchical attention fusion module, and adopts a dynamically adjusted hierarchical fusion weight mechanism.

[0015] The present invention first establishes a detailed near-zone model to capture the bubble dynamics process; then uses the near-zone model results as the boundary conditions of the far-zone model to construct a large-scale hydrodynamic field; finally, the spatiotemporal coupling of the two models is achieved through the step-scaling principle, thereby establishing a complete long-term diffusion prediction model. In the implementation process, the present invention introduces a scale adaptive function to dynamically adjust the calculation scale of different diffusion stages. The function automatically optimizes the time step, spatial grid ratio, and model coupling frequency based on key parameters such as the bubble diameter change rate, dissolution rate gradient, and flow field variation coefficient. At the same time, the application of the OceanDiffNet diffusion prediction model further improves the multi-scale diffusion prediction model. The simulation accuracy and calculation efficiency of the diffusion process. Through the above technical solution, the present invention solves the problem of submarine leakage. The technical problem of insufficient long-term prediction accuracy of multi-scale transport and diffusion processes in seawater has been solved. This has achieved continuous and unified simulation from the bubble scale to the ocean scale, providing more reliable technical support for the risk assessment of seabed carbon sequestration, and realizing accurate prediction of long-term diffusion behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the method of the present invention.

[0017] Figure 2 This is a schematic diagram of the OceanDiffNet neural network structure involved in the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] like Figure 1 FIG. 1 is a flow chart of a method for establishing a long-term model of transport and diffusion of submarine leaked carbon dioxide in seawater provided by the present invention. The method comprises the following steps:

[0020] S01, establish submarine leakage The transport and diffusion near-zone model in seawater is used to calculate the bubble number density through the Euler method and the particle group equilibrium model, and the first time granularity is used for near-zone Gas-liquid mass transfer calculation, considering the influence of bubble Reynolds number and interphase resistance coefficient on bubble motion;

[0021] S02, using the calculation results of the near zone model as the input boundary conditions of the far zone model, establish the positive diffusion matrix and the negative diffusion matrix, and calculate the near zone The dissolution rate and mass transfer coefficient are used as source parameters of the far zone model;

[0022] S03. Construct a far-area transport and diffusion model at the second time granularity, use the FVCOM hydrodynamic model to simulate the hydrodynamic conditions of large-scale ocean areas, introduce a transport and diffusion optimization function to deal with multi-scale coupling problems, and establish a hydrodynamic model based on the continuity equation and momentum equations.

[0023] S04. Combine the near-field simulation results with the far-field hydrodynamic field to establish a diffusion stability matrix and a diffusion variation matrix. Build a horizontal diffusion model based on the horizontal eddy viscosity coefficient and the horizontal diffusion coefficient. Introduce the TurboMix model to handle turbulent diffusion and chemical reactions.

[0024] S05. Based on the step-scaling principle, the near-zone and far-zone models are spatiotemporally coupled, and bubble dynamics parameters are applied to simulate bubble behavior at different scales to achieve long-term diffusion prediction. The step-scaling principle ensures smooth transmission of cross-scale information.

[0025] S06, measured by experiment

[0026] The gas-liquid mass transfer coefficient is used to calibrate the model parameters, adjust the dissolution rate and chemical reaction parameters in the model, and apply the marine carbonate species chemical model to simulate the dissolution Impact on seawater pH;

[0027] S07, through multi-scenario simulation verification, combined with the OceanDiffNet diffusion prediction model to achieve different water depths, temperatures, and pressure conditions Long-term dynamic prediction of transport and diffusion, using the vertical turbulence kinetic energy equation to optimize the vertical diffusion process;

[0028] S08. Optimizing the model using a scale-adaptive function to achieve accurate simulation of different diffusion stages, wherein the scale-adaptive function inputs include the bubble diameter change rate, the dissolution rate gradient, the flow field variation coefficient, the temperature stratification intensity, and the salinity gradient, and outputs the optimal spatiotemporal scale parameter set;

[0029] S09, introduce OceanDiffNet diffusion prediction model for far area The transport and diffusion process is optimized, and the hierarchical fusion weights are dynamically adjusted according to the near-area simulation results to improve the accuracy and computational efficiency of long-term diffusion prediction.

[0030] The first time granularity refers to the small time step used in the near-field model calculation, usually in the order of seconds or minutes, to accurately capture the In the fast changing processes such as bubble dissolution, breakup and rise, the first time granularity is determined by the bubble rising velocity and the grid size, which satisfies the Courant condition to ensure numerical stability.

[0031] The second time granularity refers to the large time step used in the calculation of the far-field model, which is usually at the hourly or daily level and is used for long-term simulation. In the transport and diffusion process in seawater, the second time granularity is determined by the ocean current velocity and the far-field grid size, ensuring the accuracy and computational efficiency of the numerical solution.

[0032] Among them, the forward diffusion matrix refers to the description of The mathematical expression of the diffusion forward along the ocean current in the horizontal direction includes the horizontal eddy viscosity coefficient and the diffusion coefficient. The forward diffusion matrix is ​​solved by a hyperbolic partial differential equation, taking into account the coupling effect of the convection term and the diffusion term. Its mathematical expression is: .

[0033] Among them, the negative diffusion matrix refers to the description of The negative diffusion matrix is ​​a mathematical expression of the upward propagation in the vertical direction due to gravity and buoyancy or the downward sedimentation due to density difference. The negative diffusion matrix is ​​solved by a parabolic partial differential equation, which takes into account the combined effects of gravity, density difference and vertical turbulent mixing. Its mathematical expression is: .

[0034] The diffusion stability matrix refers to the The mathematical expression of diffusion change is mainly affected by stable factors such as seawater temperature, salinity, and pressure. The diffusion stability matrix is ​​constructed by the eigenvalue decomposition method, which reflects the behavioral characteristics of the diffusion system under steady-state conditions. The horizontal eddy viscosity coefficient is calculated by the formula The horizontal diffusion coefficient calculation formula Determined jointly.

[0035] The diffusion variation matrix refers to the The mathematical expression of diffusion change is mainly affected by the changing factors such as tide, current, wind field, etc. The diffusion change matrix is ​​constructed with time-varying coefficients, which can capture the dynamic response characteristics of the system under non-steady-state conditions and is consistent with the continuity equation. is related to the momentum equation.

[0036] Among them, the ladder scale refers to a method of connecting different scale models through spatiotemporal coupling to realize the information transfer from the small-scale near-area model to the large-scale far-area model. The ladder scale adopts nested grid and boundary condition transfer technology to ensure the consistency and continuity of information between different scale models. and interphase resistance coefficient Perform scale conversion.

[0037] The bubble diameter change rate refers to the evolution rate of bubble size over time, which is calculated from the change in bubble number density in the near-zone model. It is used as the input parameter of the scale adaptive function and reflects the dynamic process of bubble dissolution and breakup.

[0038] The dissolution rate gradient refers to the spatial Dissolution heterogeneity, through Dissolution rate formula The calculated values ​​serve as input parameters of the scale-adaptive function, characterizing the spatial variability of the dissolution process.

[0039] The coefficient of variation of the flow field refers to the spatial heterogeneity of the hydrodynamic conditions. It is obtained by the ratio of the standard deviation of the flow velocity calculated by the FVCOM hydrodynamic model to the average flow velocity. It serves as the input parameter of the scale adaptive function and reflects the degree of spatial variation of the flow field.

[0040] Among them, the temperature stratification intensity refers to the degree of seawater density stratification, which is calculated by the maximum value of the vertical temperature gradient. It serves as the input parameter of the scale adaptive function and characterizes the influence of water stability on the diffusion process.

[0041] Among them, the salinity gradient refers to the seawater density change gradient, which is calculated from the vertical distribution of salinity and serves as the input parameter of the scale adaptive function. The rising speed and dissolution process of bubbles.

[0042] Among them, the optimal space-time scale parameter set refers to the output result of the scale adaptation function, which includes parameters such as the time step adjustment coefficient, the spatial grid scaling ratio, and the model coupling frequency. It is used to dynamically adjust the model calculation scale and improve simulation accuracy and computational efficiency.

[0043] Among them, the transmission and diffusion optimization function refers to a mathematical function used to deal with multi-scale coupling problems. The input includes the boundary conditions of the near-zone model, the grid parameters of the far-zone model, the diffusion coefficient, etc., and the output is the boundary condition correction coefficient and the grid scale weight, which optimizes the information transmission between the near-zone model and the far-zone model.

[0044] Among them, the TurboMix model refers to a numerical model that deals with turbulent diffusion and chemical reactions. It is constructed based on the turbulent kinetic energy equation and the chemical reaction kinetic equation to calculate Diffusion and chemical reaction processes under turbulent conditions, where the turbulent kinetic energy equation is .

[0045] The hierarchical fusion weight refers to the weight parameter used to integrate features at different scales in the OceanDiffNet model. It is determined by the temperature gradient, concentration gradient, and flow field complexity calculated by the near-zone model. It dynamically adjusts the contribution ratio of diffusion features at different spatial scales to improve the model's prediction accuracy.

[0046] Among them, the scale adaptive function is used to The bubble evolution characteristics dynamically adjust the model calculation scale. The input includes the bubble diameter change rate to represent the evolution speed of bubble size over time, the dissolution rate gradient to represent the spatial Dissolution heterogeneity and flow field variation coefficient represent the spatial heterogeneity of hydrodynamic conditions, temperature stratification intensity represents the degree of seawater density stratification, and salinity gradient represents the gradient of seawater density change. The output is the optimal spatiotemporal scale parameter set including parameters such as time step adjustment coefficient, spatial grid scaling ratio, and model coupling frequency.

[0047] The specific structure of the OceanDiffNet model is a deep learning network architecture based on a hierarchical attention mechanism, which includes three core parts: a spatial feature extraction module, a temporal feature extraction module, and a hierarchical attention fusion module. The spatial feature extraction module uses a multi-layer graph convolutional network to process the diffusion characteristics in the three-dimensional ocean space. The temporal feature extraction module uses a bidirectional LSTM network to capture long-term and short-term time dependencies. The hierarchical attention fusion module integrates features of different scales through hierarchical fusion weights. The most notable feature of the OceanDiffNet model is the use of a dynamically adjusted hierarchical fusion weight mechanism, which enables the model to intelligently adjust feature expression according to the diffusion characteristics of different regions. The hierarchical fusion weight parameters are dynamically adjusted according to the temperature gradient, concentration gradient and flow field complexity of the diffusion region to achieve adaptive learning of diffusion characteristics of different spatial scales.

[0048] The steps of establishing the training data set of the OceanDiffNet model specifically include first simulating the near area under multiple different conditions using the CFD software Fluent. The diffusion process generates high-precision samples, and then the FVCOM hydrodynamic model is used to simulate the hydrodynamic conditions of large-scale ocean areas and integrate them with the near-area simulation results. The simulation results are then compared and calibrated with the experimental measurement data to ensure the accuracy of the data set. The data is then normalized and stratified according to spatial area and time scale. Finally, the sampled data is divided into training set, validation set and test set for model training and evaluation.

[0049] The OceanDiffNet model training steps specifically include initializing model parameters, setting the learning rate and batch size, using the Adam optimization algorithm to minimize the mean squared error loss function between the prediction error and the true value, and adding a regularization term to prevent overfitting. After each training cycle, the model performance is evaluated using the validation set, and the learning rate is adjusted according to the validation set performance until the model converges or reaches the preset training cycle. Finally, the model parameters with the best performance on the validation set are selected as the final model.

[0050] The specific implementation of the above steps is described in detail below.

[0051] The specific implementation of step S01 is to first establish the seabed leakage

[0052] The near-zone model of transport and diffusion in seawater describes the gas-liquid two-phase flow field through the Euler method, and calculates the bubble number density by combining the particle group equilibrium model. The specific implementation process includes constructing a three-dimensional flow field calculation grid, the grid size is determined according to the diameter of the leak, and is generally taken as 0.1 times the diameter of the leak; defining the leakage boundary conditions, including leakage rate, leakage diameter and leakage depth; applying the Euler multiphase flow model to numerically simulate the near-zone gas-liquid two-phase flow, in which the gas phase is described by the discrete bubble model; using the particle group equilibrium model to calculate the bubble number density distribution, which takes into account the interaction force and group effect between bubbles; using the first time granularity for the near-zone For gas-liquid mass transfer calculation, the first time granularity is usually set to 0.01 to 0.1 seconds; the bubble Reynolds number and interphase resistance coefficient are calculated, where the bubble Reynolds number threshold is set to 200. When the bubble is less than this threshold, the bubble is spherical, and when it is greater than this threshold, the bubble is ellipsoidal; the interphase resistance coefficient is calculated based on the bubble morphology, and then the bubble rising speed and movement trajectory are determined. The purpose of this step is to accurately simulate the leakage. The initial diffusion behavior in the near zone provides accurate boundary conditions and source parameters for the far zone model.

[0053] The specific implementation of step S02 is to use the calculation results of the near-zone model as the input boundary conditions of the far-zone model to achieve effective connection between the near-zone and far-zone models. The specific implementation process includes extracting the flow rate, concentration, temperature and other parameters on the boundary of the near-zone model as the input conditions of the far-zone model; constructing the forward diffusion matrix to describe the The process of spreading and diffusing forward along the ocean current in the horizontal direction is solved by the hyperbolic partial differential equation, taking into account the coupling effect of convection and diffusion terms; a negative diffusion matrix is ​​constructed to describe the The propagation characteristics in the vertical direction are solved by parabolic partial differential equations, taking into account the combined effects of gravity, density difference and vertical turbulent mixing; Dissolution rate, the dissolution rate threshold is set to 95%. When the dissolution rate reaches this threshold, the bubble is considered to be completely dissolved. The mass transfer coefficient is calculated. The mass transfer coefficient is determined by the permeability coefficient, interfacial area and concentration gradient. The threshold is set to ~ m / s; the calculated near-zone dissolution rate and mass transfer coefficient are used as source parameters for the far-zone model. This step ensures accurate information transfer between the near-zone and far-zone models, improving the continuity and consistency of the overall model.

[0054] The specific implementation of step S03 is to construct a far-area transport and diffusion model at the second time granularity to achieve Long-term diffusion simulation. The specific implementation process includes using the FVCOM hydrodynamic model to construct a three-dimensional unstructured grid, and the horizontal resolution of the grid gradually increases from 10 to 50 m at the near-zone boundary to 100 to 500 m in the far-zone; setting the second time granularity, usually 0.5 to 2 hours; introducing the transmission diffusion optimization function to deal with multi-scale coupling problems. This function is constructed based on the characteristic decomposition method. The input includes the near-zone model boundary conditions, far-zone model grid parameters, diffusion coefficient, etc., and the output is the boundary condition correction coefficient and grid scale weight; a hydrodynamic model is established based on the continuity equation and the momentum equation group, where the continuity equation describes the conservation of fluid mass and the momentum equation describes the change in fluid motion state; the mode splitting method is used to solve the hydrodynamic equation group, and the external mode and the internal mode are separated for calculation to improve the calculation efficiency; the hydrodynamic field parameters such as free surface elevation and velocity distribution are calculated through an iterative algorithm. The purpose of this step is to construct an accurate large-scale hydrodynamic field for Far zone diffusion provides accurate hydrodynamic background conditions.

[0055] The specific implementation of step S04 is to combine the near-field simulation results with the far-field hydrodynamic field to construct a complete far-field diffusion model. The specific implementation process includes taking the near-field simulation results as the inner boundary conditions of the far-field model; establishing a diffusion stability matrix to describe the far-field diffusion model under stable hydrodynamic conditions; Diffusion change, the matrix is ​​constructed by eigenvalue decomposition method, taking into account the influence of stable factors such as seawater temperature, salinity, and pressure; the diffusion change matrix is ​​constructed to describe the dynamic hydrodynamic conditions Diffusion changes, the matrix is ​​constructed using time-varying coefficients, taking into account the influence of changing factors such as tides, currents, and wind fields; a horizontal diffusion model is constructed based on the horizontal eddy viscosity coefficient and the horizontal diffusion coefficient, and the threshold of the horizontal eddy viscosity coefficient is set to 0.1~0.5 , the horizontal diffusion coefficient threshold is set to 1 to 5 ; The TurboMix model is introduced to deal with turbulent diffusion and chemical reactions. The model is constructed based on the turbulent kinetic energy equation and the chemical reaction kinetic equation; the diffusion equation is solved using an implicit difference format to ensure numerical stability. The purpose of this step is to accurately describe the far zone Diffusion process, considering the comprehensive influence of hydrodynamic conditions on the diffusion process.

[0056] The specific implementation of step S05 is to perform spatiotemporal coupling of the near-zone and far-zone models based on the step-scaling principle to achieve smooth cross-scale information transfer. The specific implementation process includes defining a spatiotemporal coupling region, which covers the outer boundary of the near-zone model and the inner boundary of the far-zone model, with a width typically 3 to 5 times the near-zone characteristic scale; designing a temporal coupling step, typically 10 to 20 times the near-zone time step; applying nested grid technology to achieve spatial scale coupling, using linear interpolation to transfer information between different grids; achieving temporal scale coupling through boundary condition transfer technology, using time averaging to process information transfer between different time steps; applying bubble dynamics parameters to simulate bubble behavior at different scales, including bubble Reynolds number and interphase drag coefficient; and establishing a cross-scale error control mechanism, initiating an adaptive correction algorithm when the cross-scale transfer error exceeds 5%. The purpose of this step is to ensure smooth information transfer between the near-zone and far-zone models and improve the continuity and accuracy of the overall model.

[0057] The specific implementation of step S06 is measured by experiment.

[0058] The gas-liquid mass transfer coefficient is used to calibrate the model parameters to improve the accuracy of the model. The specific implementation process includes collecting data under different pressure and temperature conditions. Experimental data on gas-liquid mass transfer coefficients; fitting the relationship between the mass transfer coefficients and environmental parameters using the least squares method; adjusting the dissolution rate parameters in the model so that the relative error between the simulated dissolution rate and the experimental data is within 10%; adjusting chemical reaction parameters, including reaction rate constants and equilibrium constants; and applying a chemical model of marine carbonate species to simulate dissolution. The model considers the impact of pH value on seawater. 、 、 The model then verifies the pH value predictions, ensuring the relative error between them and the experimental measurements is within 5%. The model parameters are then adjusted based on the calibration results to optimize overall model performance. This step ensures the accuracy of the model parameters and improves the model's ability to describe actual physical processes.

[0059] The specific implementation of step S07 is to verify through multi-scenario simulation and combine with OceanDiffNet diffusion prediction model to achieve Long-term dynamic prediction of transport and diffusion. The specific implementation process includes designing multiple simulation scenarios covering different water depths (50-3000 m), temperatures (5-25°C), and pressures (0.5-30 MPa); running coupled models for long-term simulations, typically for 1-5 years; optimizing the vertical diffusion process using the vertical turbulent kinetic energy equation, which takes into account buoyancy and turbulent dissipation; introducing the OceanDiffNet diffusion prediction model for long-term prediction, which is built based on deep learning technology; and comparing the results under different scenarios. Diffusion range, concentration distribution and pH value change; evaluate the accuracy of the model prediction results, and control the error within 15% by comparing with experimental data or historical observation data. The purpose of this step is to verify the applicability of the model under different environmental conditions and achieve Long-term dynamic prediction of transport and diffusion.

[0060] The specific implementation of step S08 is to optimize the model using a scale-adaptive function to achieve accurate simulation of different diffusion stages. The specific implementation process includes defining the input parameters of the scale-adaptive function, including the bubble diameter change rate, dissolution rate gradient, flow field variation coefficient, temperature stratification intensity, and salinity gradient; calculating the bubble diameter change rate, with the change rate threshold set to 5% / minute; calculating the dissolution rate gradient, with the gradient threshold set to 0.5% / m; calculating the flow field variation coefficient, with the coefficient threshold set to 0.2; calculating the temperature stratification intensity, with the intensity threshold set to 0.05°C / m; calculating the salinity gradient, with the gradient threshold set to 0.01‰ / m; applying a genetic algorithm to optimize the scale-adaptive function and determine the weights of each input parameter; outputting the optimal spatiotemporal scale parameter set, including the time step adjustment coefficient, spatial grid scaling ratio, and model coupling frequency; and dynamically adjusting the model calculation scale based on the optimal parameter set. The purpose of this step is to improve the simulation accuracy and computational efficiency of the model at different diffusion stages by adaptively adjusting the calculation scale.

[0061] The specific implementation of step S09 is to introduce the OceanDiffNet diffusion prediction model to the far area The transport and diffusion processes are optimized to improve the accuracy and computational efficiency of long-term diffusion predictions. The specific implementation process includes constructing the OceanDiffNet model framework, which is based on a deep learning network architecture with a hierarchical attention mechanism; training the OceanDiffNet model using simulation results from a near-field model to establish a mapping relationship between diffusion characteristics and prediction results; dynamically adjusting the hierarchical fusion weights based on the near-field simulation results, which are determined by temperature gradients, concentration gradients, and flow field complexity; applying the OceanDiffNet model to predict far-field diffusion processes, with a prediction timeframe of 5 to 10 years; regularly updating the OceanDiffNet model with new simulation data to maintain the model's prediction accuracy; and evaluating the optimization results of the OceanDiffNet model, including improvements in computational efficiency and prediction accuracy. The goal of this step is to optimize the far-field diffusion prediction process through deep learning technology and significantly improve the accuracy and computational efficiency of long-term diffusion predictions.

[0062] The detailed structure of the OceanDiffNet model is a deep learning network architecture based on a hierarchical attention mechanism, which consists of three core parts. The first part is the spatial feature extraction module, which uses a multi-layer graph convolutional network to process the diffusion characteristics in the three-dimensional ocean space. Specifically, it consists of three layers of graph convolution layers, each layer uses 64 convolution kernels, and the convolution kernel size is

[0063] The activation function uses LeakyReLU, and a batch normalization layer is used after the convolution to improve training stability. The second part is the temporal feature extraction module, which uses a bidirectional LSTM network to capture long-term and short-term temporal dependencies. This module consists of two bidirectional LSTM layers, each with 128 hidden units, with a time step of 24. Dropout technology is used to prevent overfitting, with a dropout ratio of 0.3. The third part is the hierarchical attention fusion module, which integrates features of different scales through hierarchical fusion weights. It consists of a multi-head attention layer with 8 attention heads and an attention dimension of 64, and a fully connected layer with an output dimension of the predicted target dimension. The most notable feature of the OceanDiffNet model is its use of a dynamically adjusted hierarchical fusion weight mechanism. The hierarchical fusion weight parameters are dynamically adjusted by the temperature gradient, concentration gradient, and flow field complexity. The temperature gradient threshold is set to 0.05°C / m, the concentration gradient threshold is set to 0.5% / m, and the flow field complexity is determined by the turbulence intensity characteristic value, with a threshold set to 0.15.

[0064] The detailed steps for establishing the training dataset of the OceanDiffNet model include five key stages. The first stage is to simulate the near-field under multiple different conditions using the CFD software Fluent. The diffusion process generates high-precision samples. The simulation conditions cover water depths of 50 to 3000 m, temperatures of 5 to 25°C, pressures of 0.5 to 30 MPa, and leakage rates of 0.1 to 10 kg / s. The simulation duration is 0 to 48 hours. The second stage uses the FVCOM hydrodynamic model to simulate large-scale sea hydrodynamic conditions. The horizontal resolution gradually increases from 10 to 50 m near the boundary to 100 to 500 m in the far area. m, vertically divided into 20 to 30 layers, and the simulation time is 1 to 12 months; the third stage is to integrate the near-zone simulation results with the far-zone hydrodynamic field, and use the step-scale principle for spatiotemporal coupling to generate a complete diffusion data set; the fourth stage is to compare and calibrate the simulation results with the experimental measurement data, and adjust the model parameters so that the relative error between the simulation results and the experimental data is controlled within 10%; the fifth stage is to process and divide the data, including normalizing the data, using the Z-score standardization method, and stratifying the data according to spatial area and time scale. The sampling ratio is 1:5:10, corresponding to the near zone, middle zone and far zone. Finally, the sampled data is divided into training set, validation set and test set, accounting for 70%, 15% and 15% respectively.

[0065] The detailed steps of OceanDiffNet model training include six key stages. The first stage is to initialize the model parameters. The weights are initialized using the He initialization method, and the bias term is initialized to 0. The second stage is to set the training hyperparameters, including the initial value of the learning rate set to 0.001, the batch size set to 64, and the training cycle set to 200. The third stage is to minimize the loss function using the Adam optimization algorithm. The loss function adopts the mean squared error loss function, and the L2 regularization term is added to prevent overfitting. The regularization coefficient is set to 0.001. The fourth stage is to use the validation set to evaluate the model performance after each training cycle. The evaluation indicators include root mean square error and mean absolute percentage error. The fifth stage is to adjust the learning rate according to the performance of the validation set. The cosine annealing learning rate adjustment strategy is adopted. When the performance of the validation set does not improve for 5 consecutive cycles, the learning rate is reduced to 0.1 times the original value. The sixth stage is to select the final model. Training continues until the model converges or reaches the preset training cycle. Finally, the model parameters with the best performance on the validation set are selected as the final model. The model performance evaluation threshold is set to a root mean square error less than 0.05 and a mean absolute percentage error less than 10%.

[0066] It should be noted that the first core technical idea of ​​the present invention is the multi-scale coupling mechanism of the near-zone and far-zone models. In traditional methods, the near-zone model and the far-zone model are usually separated and cannot effectively handle the information transmission between different scales. The present invention organically combines the bubble dissolution process calculated by the near-zone model with the hydrodynamic field of the far-zone model by establishing a positive diffusion matrix and a negative diffusion matrix, thereby realizing the continuous transmission of cross-scale information. This coupling mechanism enables the model to simultaneously capture the microscopic bubble dynamics process and the macroscopic ocean diffusion behavior, greatly improving the comprehensiveness and consistency of the simulation. In practical applications, this multi-scale coupling enables the model to accurately describe the entire diffusion process starting from the leakage source, avoiding the error accumulation caused by scale breaks in traditional methods.

[0067] The second core technical idea is the introduction of a scale-adaptive function. Diffusion models in the prior art often use fixed spatiotemporal scale parameters, which are difficult to adapt to dynamic changes in the diffusion process. The scale-adaptive function of the present invention achieves dynamic adjustment of the calculation scale by comprehensively considering key parameters such as the bubble diameter change rate, the dissolution rate gradient, and the flow field variation coefficient. This adaptive mechanism enables the model to select the optimal calculation strategy based on the diffusion characteristics of different regions and stages, thereby improving the calculation efficiency while ensuring the calculation accuracy. Compared with the traditional fixed-scale method, this adaptive strategy can better cope with the changing conditions in the complex marine environment and improve the adaptability and robustness of the model.

[0068] The third core technical approach is the application of the OceanDiffNet deep learning model. Traditional diffusion models are primarily based on empirical formulas or simplified mathematical models, making it difficult to capture nonlinear characteristics in complex environments. The OceanDiffNet model introduced in this paper employs a hierarchical attention mechanism to simultaneously extract spatial and temporal features, dynamically integrating feature information at different scales through hierarchical fusion weights. This deep learning approach not only improves the model's prediction accuracy but also enhances its ability to handle complex nonlinear problems, making long-term diffusion predictions more accurate and reliable.

[0069] The synergy of these three technical ideas forms a complete technical system, which significantly improves the prevention and control of submarine leakage.

[0070] The multi-scale coupling mechanism ensures continuous modeling from micro to macro, the scale adaptive function provides dynamic optimization of the calculation process, and the OceanDiffNet model enhances the nonlinear feature extraction and prediction capabilities of the system. The three work together to solve the problems of insufficient accuracy and low computational efficiency faced by traditional methods in cross-scale and long-term prediction. This synergy not only optimizes the performance of the diffusion model, but also provides more reliable technical support for seabed carbon storage risk assessment, which is of great significance for understanding and predicting seabed leakage. environmental impact is of great significance.

[0071] Specifically, the principle of the present invention is: the technical solution principle of the present invention is based on multi-scale coupling and dynamic adaptive adjustment mechanism. First, at the model construction level, by establishing a near-zone model of the first time granularity and a far-zone model of the second time granularity, the physical process simulation problems at different scales are solved respectively. The near-zone model uses the Euler method and the particle swarm equilibrium model to accurately calculate the bubble number density, considering the influence of the bubble Reynolds number and the interphase resistance coefficient on the bubble motion, and accurately describes the bubble motion from a microscopic perspective. The far-field model is based on the FVCOM hydrodynamic model, combining the continuity equation and the momentum equations to simulate the hydrodynamic conditions of large-scale ocean areas and provide a macroscopic environment for the diffusion process.

[0072] Secondly, at the scale coupling level, this invention employs the principle of step scaling to achieve spatiotemporal coupling between the near-zone and far-zone models. By constructing positive and negative diffusion matrices, as well as diffusion stability and diffusion variation matrices, a continuous mathematical description of physical processes at different scales is achieved. This matrix construction method ensures smooth cross-scale information transfer, enabling the near-zone dissolution rate and mass transfer coefficient to serve as source parameters in the far-zone model, enabling continuous simulation of physical processes.

[0073] Third, at the dynamic adaptation level, the present invention introduces a scale-adaptive function and the OceanDiffNet diffusion prediction model to achieve intelligent optimization of the computational process. The scale-adaptive function dynamically adjusts the time step, spatial grid, and model coupling frequency based on parameters such as the bubble diameter change rate, dissolution rate gradient, flow field variation coefficient, temperature stratification intensity, and salinity gradient, thereby adopting the optimal computational strategy at different diffusion stages. The OceanDiffNet model, on the other hand, integrates features at different scales through a hierarchical attention mechanism and intelligently adjusts feature representation based on the diffusion characteristics of different regions, further improving the model's prediction accuracy and generalization capabilities.

[0074] In summary, the present invention realizes continuous modeling from micro to macro through multi-scale coupling and dynamic adaptive adjustment mechanism, and optimizes the calculation process through intelligent algorithm, thus solving the problem of submarine leakage. Technical problems with insufficient long-term prediction accuracy of multi-scale transport and diffusion processes in seawater.

[0075] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0076] The specific implementation of step S01 is to first establish the seabed leakage The near-zone transport-diffusion model in seawater uses the Euler method to describe the gas-liquid two-phase flow field and combines it with the particle swarm equilibrium model to calculate the bubble number density. The specific implementation process includes constructing a three-dimensional flow field calculation grid, with the grid size determined by the leak diameter, generally taking 0.1 times the leak diameter; defining the leak boundary conditions, including the leak rate, leak diameter, and leak depth; and applying the Euler multiphase flow model to numerically simulate the near-zone gas-liquid two-phase flow. The gas phase control equation is:

[0077] ;

[0078] Where, is the gas phase volume fraction; is the gas phase density, in units of ; is the gas phase velocity vector, in units of ; is the mass source term, which represents the mass transfer between phases and has a unit of .

[0079] The liquid phase governing equation is:

[0080] ;

[0081] Where, is the liquid phase volume fraction; is the liquid density, in units of ; is the liquid phase velocity vector, in units of .

[0082] The particle swarm equilibrium model is used to calculate the bubble number density distribution. The model expression is:

[0083] ;

[0084] Where, is the bubble number density, in units of ; is the bubble number source term, which represents the bubble breakage and coalescence, and its unit is .

[0085] Use the first time granularity for near zone Gas-liquid mass transfer calculation, the mass transfer equation is: ;

[0086] Where, For mass transfer Mass, in units of ; is the liquid side mass transfer coefficient, in units of ; is the gas-liquid contact specific surface area, in units of ; is the saturation concentration, in units of ; is the bulk concentration of the liquid phase, in units of ; is the control volume, in units of .

[0087] Calculate the bubble Reynolds number, the expression is:

[0088] ;

[0089] Where, is the bubble equivalent diameter, in units of ; is the liquid phase dynamic viscosity, unit is .

[0090] Calculate the interphase resistance coefficient, the expression is:

[0091] ;

[0092] Where, The purpose of this step is to accurately simulate the leakage The initial diffusion behavior in the near zone provides accurate boundary conditions and source parameters for the far zone model.

[0093] The specific implementation of step S02 is to use the calculation results of the near-zone model as the input boundary conditions of the far-zone model to achieve effective connection between the near-zone and far-zone models. The specific implementation process includes extracting the flow rate, concentration, temperature and other parameters on the boundary of the near-zone model as the input conditions of the far-zone model; constructing the forward diffusion matrix to describe the In the process of spreading forward with the ocean current in the horizontal direction, the matrix is ​​solved by using a hyperbolic partial differential equation, and its mathematical expression is:

[0094] ;

[0095] Where, for Concentration volume fraction; for Density, in units of ; is the convection velocity vector, in units of ; for Source term, in units of .

[0096] Constructing negative diffusion matrix description The propagation characteristics in the vertical direction are expressed as follows:

[0097] ;

[0098] Where, for Concentration distribution in the vertical direction; is the vertical velocity vector, in units of ; is the source term in the vertical direction, in units of ; is the reference density, in units of .

[0099] Calculate near zone The dissolution rate is expressed as:

[0100] ;

[0101] Where, is the dissolution rate; For injection Mass, in units of ; For the remaining Bubble mass, in .

[0102] Calculate the mass transfer coefficient, the expression is:

[0103] ;

[0104] Where, is the mass transfer coefficient, in units of ; for Diffusion coefficient in seawater, in units of ; is the boundary layer thickness in units of ; is the Sherwood number, dimensionless. The purpose of this step is to ensure the accurate transfer of information between the near-zone and far-zone models and to improve the continuity and consistency of the overall model.

[0105] The specific implementation of step S03 is to construct a far-area transport and diffusion model at the second time granularity to achieve The specific implementation process includes constructing a three-dimensional unstructured grid using the FVCOM hydrodynamic model, with the horizontal resolution gradually increasing from 10 to 50 m near the boundary to 100 to 500 m in the far zone; setting a second time granularity, usually 0.5 to 2 hours; and introducing a transmission diffusion optimization function to deal with multi-scale coupling problems. The function expression is: ;

[0106] Where, Output value for optimization function; For the The weight coefficient of each feature; For the eigenvalues; is the input parameter vector, including the near-zone model boundary conditions, far-zone model grid parameters, diffusion coefficient, etc.; is the eigendecomposition function; is the number of features.

[0107] The hydrodynamic model is established based on the continuity equation and momentum equations. The continuity equation is:

[0108] ;

[0109] Where, is the free surface elevation in units of ; is the total water depth in units of ; for Directional flow velocity, in units of ; for Directional flow velocity, in units of ; for The vertical velocity in coordinates is ; is the vertical coordinate.

[0110] The directional momentum equation is:

[0111] ;

[0112] Where, is the Coriolis force parameter, in units of ; is the acceleration due to gravity, in units of ; is the reference density, in units of ; is the density of seawater, in units of ; is the vertical eddy viscosity coefficient, in units of ; for Direction horizontal force, unit is ; for Directional diffusion term, in units of .

[0113] The directional momentum equation has a similar form. The purpose of this step is to construct an accurate large-scale hydrodynamic field for Far zone diffusion provides accurate hydrodynamic background conditions.

[0114] The specific implementation of step S04 is to combine the near-field simulation results with the far-field hydrodynamic field to construct a complete far-field diffusion model. The specific implementation process includes taking the near-field simulation results as the inner boundary conditions of the far-field model; establishing a diffusion stability matrix to describe the far-field diffusion model under stable hydrodynamic conditions; Diffusion changes, the matrix is ​​constructed by the eigenvalue decomposition method:

[0115] ;

[0116] Where, is a diffusion-stabilized matrix; is the eigenvector matrix; is a diagonal matrix of eigenvalues.

[0117] Constructing a diffusion variation matrix to describe dynamic hydrodynamic conditions Diffusion changes, the matrix expression is:

[0118] ;

[0119] Where, is the time-varying diffusion change matrix; It is the time-varying part, affected by changing factors such as tides, ocean currents, and wind fields.

[0120] A horizontal diffusion model is constructed based on the horizontal eddy viscosity coefficient and the horizontal diffusion coefficient. The calculation formula of the horizontal eddy viscosity coefficient is:

[0121] ;

[0122] Where, is the horizontal eddy viscosity coefficient, in units of ; is the empirical coefficient, with a value of 0.1 to 0.2; is the characteristic length scale, in units of .

[0123] The formula for calculating the horizontal diffusion coefficient is:

[0124] ;

[0125] Where, is the horizontal diffusion coefficient, in ; is the vertical characteristic length scale, in units of ; is the Prandtl number, dimensionless.

[0126] The TurboMix model is introduced to deal with turbulent diffusion and chemical reactions. The model is built based on the turbulent kinetic energy equation: ;

[0127] Where, is the turbulent kinetic energy, in units of ; is the turbulent kinetic energy diffusion coefficient, in units of ; is the vertical thermal diffusivity, in units of ; is an empirical constant with a value of 1.4 to 1.8. The purpose of this step is to accurately describe the far zone Diffusion process, considering the comprehensive influence of hydrodynamic conditions on the diffusion process.

[0128] The specific implementation of step S05 is to perform spatiotemporal coupling of the near-zone and far-zone models based on the step-scaling principle to achieve smooth cross-scale information transmission. The specific implementation process includes defining a spatiotemporal coupling region, which covers the outer boundary of the near-zone model and the inner boundary of the far-zone model, with a width typically 3 to 5 times the near-zone characteristic scale; designing a temporal coupling step, typically 10 to 20 times the near-zone time step; and applying nested grid technology to achieve spatial scale coupling. The spatial interpolation function expression is:

[0129] ;

[0130] Where, is the physical value of the target point; For the The physical value of a known point; is the spatial weight function; is the number of nodes involved in interpolation.

[0131] Time scale coupling is achieved by boundary condition transfer technology, and the time average expression is:

[0132] ;

[0133] Where, is the average physical quantity value under the time step of the far zone model; For the The physical quantity value under the near-zone model time step; is the time average number of steps.

[0134] The bubble dynamics parameters are used to simulate the behavior of bubbles at different scales, and the bubble Reynolds number is calculated as:

[0135] ;

[0136] Where, is the dynamic viscosity, in units of .

[0137] The interphase resistance coefficient is calculated as:

[0138] .

[0139] A cross-scale error control mechanism is established, and the error expression is:

[0140] ;

[0141] Where, is the relative error; Calculate values ​​for the far zone model; Calculate values ​​for the near-zone model. The purpose of this step is to ensure smooth information transfer between the near-zone and far-zone models, and to improve the continuity and accuracy of the overall model.

[0142] The specific implementation of step S06 is measured by experiment. The gas-liquid mass transfer coefficient is used to calibrate the model parameters to improve the accuracy of the model. The specific implementation process includes collecting data under different pressure and temperature conditions. Experimental data of gas-liquid mass transfer coefficient; the relationship between the mass transfer coefficient and environmental parameters was fitted using the least squares method, and the fitting equation was:

[0143] ;

[0144] Where, is the mass transfer coefficient, in units of ; is the pressure, the unit is ; is the temperature in units of ; 、 、 、 、 、 is the fitting coefficient.

[0145] Adjust the dissolution rate parameters in the model, and the dissolution rate calculation equation is:

[0146] ;

[0147] Where, is the dissolution rate, in units of ; is the gas-liquid contact specific surface area, in units of ; is the saturation concentration, in units of ; is the bulk concentration of the liquid phase, in units of .

[0148] Adjust chemical reaction parameters, including reaction rate constants and equilibrium constants; simulate dissolution using a chemical model of marine carbonate species The impact on seawater pH is based on the following chemical equilibrium equation:

[0149] ;

[0150] .

[0151] The equilibrium constant calculation formula is:

[0152] ;

[0153] ;

[0154] Where, and is the equilibrium constant; 、 、 、 are the concentrations of hydrogen ions, bicarbonate ions, dissolved carbon dioxide, and carbonate ions, respectively, all in units of The purpose of this step is to ensure the accuracy of the model parameters and improve the model's ability to describe the actual physical process.

[0155] The specific implementation of step S07 is to verify through multi-scenario simulation and combine with OceanDiffNet diffusion prediction model to achieve Long-term dynamic prediction of transport and diffusion. The specific implementation process includes designing multiple simulation scenarios covering different water depths (50-3000 m), temperatures (5-25°C), and pressures (0.5-30 MPa); running a coupled model for long-term simulations, typically for 1-5 years; and optimizing the vertical diffusion process using the vertical turbulent kinetic energy equation, which is expressed as:

[0156] ;

[0157] Where, is the turbulent kinetic energy, in units of ; is the turbulence length scale in units of ; for Directional velocity component, in units of ; is the turbulent kinetic energy diffusion coefficient, in units of ; 、 is an empirical constant; is the shear generation term, in units of ; is the buoyancy term, in units of ; is an empirical constant with a value of 16.6. The purpose of this step is to verify the applicability of the model under different environmental conditions and to achieve Long-term dynamic prediction of transport and diffusion.

[0158] The specific implementation of step S08 is to optimize the model using a scale-adaptive function to achieve accurate simulation of different diffusion stages. The specific implementation process includes defining the input parameters of the scale-adaptive function, including the bubble diameter change rate, dissolution rate gradient, flow field variation coefficient, temperature stratification intensity, and salinity gradient; and calculating the bubble diameter change rate, which is expressed as:

[0159] ;

[0160] Where, is the bubble diameter change rate, in units of ; is the bubble equivalent diameter, in units of ; For time interval The diameter change in ; is the time interval in units of .

[0161] Calculate the dissolution rate gradient. The gradient expression is:

[0162] ;

[0163] Where, is the dissolution rate gradient, in units of ; is the dissolution rate, in units of ; 、 、 They are 、 、 The unit vector of the direction.

[0164] Calculate the coefficient of variation of the flow field. The coefficient expression is:

[0165] ;

[0166] Where, is the coefficient of variation of the flow field, dimensionless; is the standard deviation of flow rate, in units of ; is the average flow velocity in units of .

[0167] Calculate the temperature stratification intensity. The intensity expression is:

[0168] ;

[0169] Where, is the temperature stratification intensity, in units of ; is the temperature in units of ; is the depth in units of . Calculate the salinity gradient, the gradient expression is:

[0170] ;

[0171] Where, is the salinity gradient, in units of ; is salinity, in units of .

[0172] Genetic algorithm is used to optimize the scale adaptive function. The scale adaptive function expression is:

[0173] ;

[0174] ]In the formula, is the scale adaptive function; is the mapping function, which is determined by genetic algorithm optimization. Output the optimal spatiotemporal scale parameter set, including the time step adjustment coefficient , spatial grid scaling and model coupling frequency The purpose of this step is to improve the simulation accuracy and computational efficiency of the model at different diffusion stages by adaptively adjusting the computational scale.

[0175] The specific implementation of step S09 is to introduce the OceanDiffNet diffusion prediction model to the far area The transport and diffusion process is optimized to improve the accuracy and computational efficiency of long-term diffusion predictions. The specific implementation process includes building the OceanDiffNet model framework, which is based on a deep learning network architecture with a hierarchical attention mechanism; training the OceanDiffNet model using the simulation results of the near-zone model to establish a mapping relationship between diffusion characteristics and prediction results; and dynamically adjusting the hierarchical fusion weights based on the near-zone simulation results. The hierarchical fusion weight calculation formula is:

[0176] ;

[0177] Where, For the The fusion weight of the layer; For the The characteristic score of the layer is calculated from the temperature gradient, concentration gradient and flow field complexity; For the The scaling factor of the layer; is the total number of levels.

[0178] The OceanDiffNet model is used to predict far-field diffusion processes with a forecast horizon of 5 to 10 years. The OceanDiffNet model is regularly updated with new simulated data to maintain its accuracy. The results of OceanDiffNet model optimization are evaluated, including improvements in computational efficiency and prediction accuracy. This step aims to optimize the far-field diffusion prediction process through deep learning techniques, significantly improving the accuracy and computational efficiency of long-term diffusion predictions.

[0179] The detailed structure of the OceanDiffNet model is a deep learning network architecture based on a hierarchical attention mechanism, consisting of three core components. The first component is the spatial feature extraction module, which uses a multi-layer graph convolutional network to process the diffusion characteristics in the three-dimensional ocean space. The graph convolution operation expression is:

[0180] ;

[0181] Where, For the The feature matrix of the layer; is the adjacency matrix; is the degree matrix; For the The weight matrix of the layer; is the activation function. The second part is the time series feature extraction module, which uses a bidirectional LSTM network to capture long-term and short-term time dependencies. The forward propagation equation of the LSTM unit is:

[0182] ;

[0183] ;

[0184] ;

[0185] ;

[0186] ;

[0187] ;

[0188] Where, 、 、 They are input gate, forget gate and output gate respectively; is the candidate cell state; is the cell state; is in hidden state; and are weight and bias parameters; is the sigmoid activation function; is the hyperbolic tangent activation function; is the Hadamard product. The third part is the hierarchical attention fusion module, which integrates features of different scales through a multi-head self-attention mechanism. The attention calculation formula is: ;

[0189] Where, 、 、 They are query matrix, key matrix and value matrix respectively; is the dimension of the key. The multi-head attention expression is:

[0190] ;

[0191] ;

[0192] Where, 、 、 and is a trainable weight matrix; is the number of attention heads.

[0193] The detailed steps for establishing the training dataset of the OceanDiffNet model include five key stages. The first stage is to simulate the near-field under multiple different conditions using the CFD software Fluent. The diffusion process generates high-precision samples; the second stage is to use the FVCOM hydrodynamic model to simulate the hydrodynamic conditions of large-scale ocean areas; the third stage is to integrate the near-area simulation results with the far-area hydrodynamic field, and use the step-scaling principle for spatiotemporal coupling; the fourth stage is to compare and calibrate the simulation results with the experimental measurement data; the fifth stage is to process and divide the data, including normalizing the data using the Z-score standardization method:

[0194] ;

[0195] Where, is the standardized value; is the original value; is the mean; is the standard deviation.

[0196] The steps of OceanDiffNet model training include initializing model parameters, and the weights are initialized using the He method:

[0197] ;

[0198] Where, is the weight parameter; is a normal distribution; is the number of input neurons. Set the training hyperparameters and use the Adam optimization algorithm to minimize the loss function:

[0199] ;

[0200] Where, is the loss function; is the true value; is the predicted value; is the sample size; is the regularization coefficient; For the weight parameters.

[0201] In order to better understand and implement the present invention, the following provides a specific application scenario of the present invention, Example 2: During a carbon capture and storage project in a certain sea area, researchers analyze the possible A simulated leak assessment was conducted. The leak was located at a depth of approximately 1,200 meters, with a diameter of 0.25 meters and a leakage rate of 2.5 kg / s. Seawater environmental conditions included a surface temperature of approximately 24°C, a bottom temperature of approximately 5°C, a surface salinity of approximately 33‰, a bottom salinity of approximately 34.5‰, and an average flow velocity of 0.15 m / s.

[0202] First, researchers established the A near-zone model was constructed using the Euler multiphase flow model. A three-dimensional unstructured grid was constructed with a near-zone computational domain of 100 m × 100 m × 150 m. The grid size was 0.1 times the leak diameter, or 0.025 m. The first time granularity in the near-zone model was set to 0.05 s, and the computational duration was 24 hours. The near-zone model calculation results are shown in Table 1.

[0203] Table 1 Near area Bubble simulation result parameters Based on the near-field simulation results, the researchers determined Bubble dissolution characteristics and mass transfer coefficient, gas-liquid mass transfer coefficient The fitting results are shown in Table 2.

[0204] Table 2 Under different conditions Fitting results of gas-liquid mass transfer coefficient Next, the researchers constructed a far-zone transport and diffusion model. Using the FVCOM hydrodynamic model, a three-dimensional unstructured grid was established. The far-zone computational domain was 20 km × 20 km × 1.2 km. The horizontal grid resolution gradually increased from 20 m at the near-zone boundary to 300 m in the far-zone, with 25 vertical layers. The second time granularity in the far-zone model was set to 1 hour, and the simulation duration was one year. The velocity distribution of the far-zone hydrodynamic field simulation results is shown in Table 3.

[0205] Table 3 Velocity distribution at typical locations in the far zone The researchers employed the step-scaling principle to spatially and temporally couple the near- and far-zone models and optimized the model's computational scale using a scale-adaptive function. Key parameters of the scale-adaptive function include a bubble diameter change rate threshold of 5% / minute, a dissolution rate gradient threshold of 0.5% / m, a flow field coefficient of variation threshold of 0.2, a temperature stratification intensity threshold of 0.05°C / m, and a salinity gradient threshold of 0.01‰ / m.

[0206] Subsequently, researchers introduced the OceanDiffNet diffusion prediction model to conduct Long-term prediction of transport and diffusion. The OceanDiffNet model consists of three graph convolutional layers, two bidirectional LSTM layers, and a multi-head attention fusion layer. The training dataset consists of near-field simulation results coupled with far-field hydrodynamic fields under various conditions. The model is trained using the Adam optimization algorithm, with an initial learning rate of 0.001, a batch size of 64, and 150 training cycles. The prediction accuracy evaluation results of the OceanDiffNet model are shown in Table 4.

[0207] Table 4. OceanDiffNet model prediction accuracy evaluation Through long-term simulation, we predict The simulation results of the impact range and degree of the leakage on the pH value of seawater are shown in Table 5.

[0208] Table 5 Different time scales Impact of leaks on seawater pH Traditional submarine The leakage diffusion simulation method mainly uses a single-scale model, which makes it difficult to take into account the needs of both near-field detailed simulation and far-field long-term prediction. The usual practice is to directly use CFD software to simulate near-field diffusion, or use a large-scale ocean circulation model to predict far-field diffusion. The two methods are independent of each other and it is difficult to achieve effective multi-scale coupling. In addition, the traditional method lacks the ability to target the near-field diffusion. Special optimization of bubble dissolution characteristics, the simulation accuracy is not high.

[0209] Compared with the traditional method, the submarine leakage The method for establishing a long-term model of transport and diffusion in seawater has the following technical advances: by establishing a near-zone and far-zone coupling model, multi-scale seamless connection is achieved, and the near-zone simulation error is reduced by 18.5%; the step-scale principle is used for spatiotemporal coupling to ensure the smoothness of cross-scale information transmission, and the boundary condition matching accuracy is improved by 15.2%; the introduction of scale adaptive functions and OceanDiffNet diffusion prediction models significantly improves the accuracy and computational efficiency of long-term predictions, and the far-zone prediction accuracy is improved by 12.8%, and the computational efficiency is improved by two orders of magnitude; the influence of the bubble Reynolds number and the interphase resistance coefficient on bubble motion is taken into account, and the bubble dissolution simulation accuracy is improved by 16.7%. Through these technical improvements, the method of the present invention can more accurately predict submarine leaks. The long-term transport and diffusion behavior in seawater provides important technical support for the safety assessment of marine carbon storage.

[0210] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 6 and 7 below.

[0211] Table 6 Variable Explanation Table (Part I)

[0212]

[0213] Table 7 Variable Explanation Table (Part II)

[0214]

[0215] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for establishing a long-term model of the transport and diffusion of submarine leaked carbon dioxide in seawater, characterized in that: include: Building a submarine leak The transport and diffusion near-zone model in seawater is used to calculate the bubble number density through the Euler method and the particle group equilibrium model, and the first time granularity is used for near-zone Gas-liquid mass transfer calculation; using the calculation results of the near-zone model as the input boundary conditions of the far-zone model, establish the positive diffusion matrix and the negative diffusion matrix; construct the far-zone transport and diffusion model at the second time granularity, and use the FVCOM hydrodynamic model to simulate the hydrodynamic conditions of large-scale ocean areas; combine the near-zone simulation results with the far-zone hydrodynamic field to establish the diffusion stability matrix and the diffusion variation matrix; according to the step-scaling principle, the near-zone and far-zone models are spatiotemporally coupled; the scale adaptive function is used to optimize the model to achieve accurate simulation of different diffusion stages; introduce the OceanDiffNet diffusion prediction model to the far-zone The transport and diffusion process is optimized to improve the accuracy and computational efficiency of long-term diffusion prediction; the diffusion stability matrix refers to the stability of the diffusion matrix under stable hydrodynamic conditions. The mathematical expression of diffusion change is affected by seawater temperature, salinity, and pressure. The diffusion stability matrix is ​​constructed by the eigenvalue decomposition method; the diffusion change matrix refers to the diffusion change matrix under dynamic hydrodynamic conditions. The mathematical expression of diffusion changes is affected by tidal currents, ocean currents, and wind fields, and the diffusion change matrix is ​​constructed using time-varying coefficients. The ladder scaling method refers to the method of connecting different scale models through spatiotemporal coupling to achieve information transfer from small-scale near-area models to large-scale far-area models. The ladder scaling uses nested grids and boundary condition transfer technology to ensure the consistency and continuity of information between different scale models. The scale adaptive function is used to adjust the seabed leakage The bubble evolution characteristics dynamically adjust the model calculation scale. The input includes the bubble diameter change rate, dissolution rate gradient, flow field variation coefficient, temperature stratification intensity, and salinity gradient. The output is the optimal spatiotemporal scale parameter set. The specific structure of the OceanDiffNet diffusion model is a deep learning network architecture based on the hierarchical attention mechanism, which includes a spatial feature extraction module, a temporal feature extraction module, and a hierarchical attention fusion module.

2. The method according to claim 1, characterized in that The first time granularity refers to the small time step used in the near-zone model calculation to accurately capture In the rapid changing process of bubble dissolution, breakup and rise, the first time granularity is determined by the bubble rising velocity and the grid size, which satisfies the Courant condition to ensure numerical stability.

3. The method according to claim 2, characterized in that The second time granularity refers to the large time step used in the calculation of the far-zone model for long-term simulation. In the transport and diffusion process in seawater, the second time granularity is determined by the ocean current velocity and the far-field grid size, ensuring the accuracy and computational efficiency of the numerical solution.

4. The method according to claim 3, characterized in that The forward diffusion matrix refers to the description of The mathematical expression of the diffusion propagation forward in the horizontal direction with the ocean current includes the horizontal eddy viscosity coefficient and the diffusion coefficient. The forward diffusion matrix is ​​solved using a hyperbolic partial differential equation, taking into account the coupling effect of convection and diffusion terms.

5. The method according to claim 4, characterized in that The negative diffusion matrix refers to the description of The negative diffusion matrix is ​​a mathematical expression of the upward propagation in the vertical direction due to gravity and buoyancy or the downward sedimentation due to density differences. It is solved using a parabolic partial differential equation, which takes into account the combined effects of gravity, density differences and vertical turbulent mixing.

Images