Method and system for three-dimensional dynamic deduction of pollutant diffusion for complex hydrology of a canal

Abstract

The application provides a kind of method and system for canal complex hydrology pollutant diffusion three-dimensional dynamic deduction, relating to canal pollution diffusion deduction technical field.For the problems that two-dimensional model in prior art cannot depict the influence of three-dimensional turbulence on vertical mixing of pollutants, vertical diffusion coefficient lacks physical correlation, model state is disconnected with measured data, and calculation efficiency is insufficient, the application adopts the architecture of data assimilation and three-dimensional CFD model nesting, assimilates real-time monitoring data into three-dimensional CFD model configured with large eddy simulation and dynamic mesh technology through ensemble Kalman filter algorithm, and physically parameterizes vertical turbulent diffusion coefficient from turbulent kinetic energy and dissipation rate, combines GPU parallel acceleration and grid dynamic encryption technology, realizes pollutant three-dimensional diffusion trajectory deduction and layered water intake decision support in digital twin platform.The application improves the deduction accuracy and emergency timeliness.

Description

Technical Field

[0001] This invention relates to the field of pollution diffusion simulation technology in canals, and in particular to a three-dimensional dynamic simulation method and system for pollutant diffusion in complex hydrological conditions of canals. Background Technology

[0002] As a core control node for canal navigation, the strong shear turbulence generated during the discharge process of locks, combined with the backflow mixing effect at the confluence of tributaries, results in highly three-dimensional and transient diffusion characteristics of pollutants in key navigation sections. Simultaneously, water intakes at sensitive water sources are extremely sensitive to concentration differences of pollutants at different depths, and the formulation of stratified water intake strategies relies on accurate prediction of vertical concentration distribution. Therefore, establishing a method capable of real-time, dynamic, and three-dimensional simulation of pollutant migration and diffusion processes under complex hydrological conditions is of great significance for improving the deep protection capabilities and emergency decision support levels of sensitive water sources along canals.

[0003] Currently, most canal pollutant diffusion simulations employ two-dimensional hydrodynamic-water quality models based on the static pressure assumption. These models simplify the flow and concentration fields into two-dimensional distributions through vertical averaging, failing to characterize the crucial impacts of strong three-dimensional turbulence, such as lock discharge jets and tributary inflows, on the vertical mixing of pollutants. While some three-dimensional models incorporate turbulence simulation capabilities, the vertical turbulent diffusion coefficient is typically given by empirical formulas, failing to establish a physical correlation with the turbulent characteristics of the flow field, resulting in insufficient simulation accuracy of the vertical mixing process. Furthermore, existing models largely rely on preset boundary conditions and initial parameters, lacking a dynamic fusion mechanism with real-time monitoring data. Model state variables cannot be continuously updated and corrected based on measured data, leading to significant error accumulation issues in long-term simulations. Simultaneously, in terms of computational efficiency, there is a lack of adaptive mesh optimization and GPU parallel acceleration methods for pollutant concentration fronts, making it difficult to meet the timeliness requirements of simulations under sudden pollution incidents.

[0004] Therefore, this invention is proposed. Summary of the Invention

[0005] To address these issues, this invention provides a three-dimensional dynamic simulation method and system for pollutant diffusion in complex canal hydrology. This system overcomes the problems in existing technologies, such as the inability of two-dimensional models to characterize the impact of three-dimensional turbulence caused by lock discharge and tributary inflow on vertical mixing of pollutants, the lack of physical correlation mechanisms in the use of empirical formulas for vertical turbulent diffusion coefficients, the inability of model state variables to dynamically integrate with real-time monitoring data leading to error accumulation, and the lack of adaptive grid optimization and parallel acceleration, which makes it difficult to meet emergency response requirements.

[0006] To achieve the above objectives, this invention provides a three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology, comprising: S1, acquire multi-source data of the target flight segment, and construct a three-dimensional digital model base plate based on the multi-source data; S2, establish a three-dimensional CFD model of the pollutant diffusion sensitive section based on the three-dimensional digital model base plate; S3 employs an ensemble Kalman filter algorithm to assimilate the acquired multi-source data into a three-dimensional CFD model, update the flow velocity, water level, and pollutant concentration state variables of all computational grid nodes, and use the measured upstream flow and downstream water level as time-varying boundary conditions to drive the three-dimensional CFD model to perform real-time rolling calculations. S4 uses GPU parallel acceleration technology to perform simulation calculations on the assimilated three-dimensional CFD model. In areas where the pollutant concentration front gradient exceeds the preset concentration threshold, the computational grid is dynamically densified, and the measured values ​​and calculated values ​​of the monitoring points are compared to dynamically calibrate the roughness coefficient and the vertical turbulent diffusion coefficient. S5 visualizes the simulation results in three dimensions on the digital twin platform, renders the trajectory of pollutants, and outputs the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

[0007] Furthermore, the multi-source data includes underwater topographic data, geometric data of locks and energy dissipation structures, and real-time hydrological and water quality data collected from monitoring points. The real-time hydrological and water quality data includes at least vertical velocity stratification data, water level, and pollutant concentration.

[0008] Furthermore, the three-dimensional CFD model includes a hydrodynamic module and a water quality module; The hydrodynamic module is built on the Navier-Stokes equations and is equipped with a large eddy simulation turbulence model to capture three-dimensional turbulence, the VOF method to track the free surface, and dynamic mesh technology to simulate gate opening and closing and ship movement. The water quality module is constructed based on the convection-diffusion equation. The vertical turbulent diffusion coefficient in the equation is parameterized by the turbulent kinetic energy and turbulent kinetic energy dissipation rate obtained by the hydrodynamic module to characterize the influence of three-dimensional turbulence caused by lock discharge and tributary inflow on the vertical mixing of pollutants.

[0009] Furthermore, the pollutant diffusion sensitive navigation section includes the turbulent mixing zone downstream of the lock spillway, the backflow mixing zone at the tributary confluence, the main channel deep pool area, and the intake area of ​​the downstream sensitive water source.

[0010] Furthermore, the acquisition of multi-source data for the target flight segment and the construction of a three-dimensional digital model base plate based on the multi-source data include: S11 uses multibeam bathymetry to conduct underwater topographic surveys of the target section and obtain underwater topographic data; S12 uses three-dimensional laser scanning to scan the lock chamber and stilling sill of the ship lock, obtains the geometric data of the lock and stilling structure, and performs noise reduction processing on the metal structure. S13 utilizes real-time dynamic carrier phase difference technology to provide a spatial positioning reference for multibeam bathymetry and three-dimensional laser scanning, unifies the underwater topographic data and the geometric data of the lock and energy dissipation structure into the same coordinate system and performs registration, generating a three-dimensional digital model base plate containing underwater topography and shore-side structures.

[0011] Furthermore, the ensemble Kalman filter algorithm is employed to assimilate the acquired multi-source data into the three-dimensional CFD model, updating the flow velocity, water level, and pollutant concentration state variables of all computational grid nodes, including: S31, based on the flow velocity, water level and pollutant concentration analysis field of all grid nodes calculated by the three-dimensional CFD model at the previous moment, a set of state variables is generated by the Monte Carlo method as the background field at the current moment; S32, each set member in the background field is input into the three-dimensional CFD model for one-step prediction calculation to obtain the set of predicted values ​​of flow velocity, water level and pollutant concentration on all grid nodes at the current moment; S33: Obtain the vertical flow velocity stratification data, water level and pollutant concentration collected from the monitoring points in the multi-source data at the current moment, form the observation field, and set the observation error covariance matrix; S34, calculate the background field error covariance matrix based on the predicted value set, and calculate the Kalman gain matrix in combination with the observation error covariance matrix; S35, using the Kalman gain matrix, the information of the observation field is fused into the predicted value of each set member to obtain the updated set of analytical values ​​of flow velocity, water level and pollutant concentration on all grid nodes; S36, take the mean of the set of analyzed values ​​as the assimilated state variable at the current moment, and update the flow velocity, water level and pollutant concentration of all computational grid nodes in the three-dimensional CFD model.

[0012] Furthermore, the method uses measured upstream flow and downstream water level as time-varying boundary conditions to drive the three-dimensional CFD model to perform real-time rolling calculations, including: S37, Obtain the measured upstream flow rate and downstream water level value at the current moment from the multi-source data; and convert the upstream flow rate value into the velocity distribution boundary condition of the inlet section of the three-dimensional CFD model, and convert the downstream water level value into the water level boundary condition of the outlet section of the three-dimensional CFD model. S38 uses the assimilated state variables as the initial field and the transformed velocity distribution and water level boundary conditions as time-varying boundary conditions to drive the three-dimensional CFD model to calculate one time step forward, and obtain the predicted fields of velocity, water level and pollutant concentration on all computational grid nodes at the next moment. S39, the predicted field is used as the background field for a new round of ensemble Kalman filter assimilation, and S37 to S38 are repeated to achieve real-time rolling calculation of the three-dimensional CFD model.

[0013] Furthermore, the process of using GPU parallel acceleration technology to perform inference calculations on the assimilated 3D CFD model, and dynamically refining the computational grid in regions where the pollutant concentration front gradient exceeds a preset concentration threshold, includes: S41 distributes the computational tasks of the assimilated and updated 3D CFD model to multiple GPU computing cores, and solves the Navier-Stokes equations, the large eddy simulation turbulence model and the convection-diffusion equations in parallel to obtain the flow velocity, turbulent kinetic energy, turbulent kinetic energy dissipation rate and pollutant concentration at each computing grid node. S42, traverse all computational grid nodes, calculate the pollutant concentration gradient value between adjacent nodes, compare it with a preset concentration threshold, and mark the grid area whose gradient value exceeds the preset concentration threshold as the area to be encrypted; S43, refine the grid within the area to be encrypted, increase the number of grid nodes, and interpolate and map the flow field variables and pollutant concentration field variables on the original grid nodes to the newly generated grid nodes to generate the encrypted computational grid; S44, continue to perform the deduction calculation on the encrypted computing grid, and repeat S42 and S43 at each computing time step until the deduction calculation ends.

[0014] This invention also provides a three-dimensional dynamic simulation system for pollutant diffusion in complex canal hydrology, the system being used to implement any of the aforementioned three-dimensional dynamic simulation methods for pollutant diffusion in complex canal hydrology, the system comprising: The data acquisition module is used to acquire multi-source data of the target flight segment and construct a three-dimensional digital model base plate based on the multi-source data; The 3D model building module is used to build 3D CFD models of pollutant diffusion-sensitive flight segments based on a 3D digital model base. The rolling drive module is used to assimilate the acquired multi-source data into the three-dimensional CFD model using the ensemble Kalman filter algorithm, update the flow velocity, water level and pollutant concentration state variables of all computational grid nodes, and drive the three-dimensional CFD model to perform real-time rolling calculations using the measured upstream flow and downstream water level as time-varying boundary conditions. The deduction and optimization module is used to perform deduction calculations on the assimilated three-dimensional CFD model through GPU parallel acceleration technology. In areas where the pollutant concentration front gradient exceeds the preset concentration threshold, the calculation grid is dynamically densified, and the measured values ​​and calculated values ​​of monitoring points are compared to dynamically calibrate the roughness coefficient and vertical turbulent diffusion coefficient. The visualization module is used to visualize the results of the simulation calculation in three dimensions on the digital twin platform, render the movement trajectory of pollutants, and output the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: Firstly, this invention employs an ensemble Kalman filter algorithm to assimilate real-time vertical flow velocity stratification data, water level, and pollutant concentration into a three-dimensional CFD model, updates the state variables of all computational grid nodes, and uses measured upstream flow and downstream water level as time-varying boundary conditions to drive the model to perform real-time rolling calculations of forecast-assimilation-forecast. This solves the problems in existing technologies where model state variables cannot be dynamically integrated with real-time monitoring data and long-term prediction errors continue to accumulate, improves the consistency between the model prediction trajectory and measured data, and reduces the uncertainty of pollutant diffusion path and arrival time prediction. Secondly, this invention, by real-time parameterizing the vertical turbulent diffusion coefficient obtained from the hydrodynamic module in the convection-diffusion equation of the water quality module, and the turbulent kinetic energy and turbulent kinetic energy dissipation rate, replaces the empirical constants or empirical formulas set in the traditional model. This solves the problems in the prior art where the vertical turbulent diffusion coefficient lacks a physical correlation with the actual turbulent characteristics of the flow field, and cannot reflect the influence of strong three-dimensional turbulence such as lock discharge jets and tributary inflow backflow on the vertical mixing of pollutants. It establishes a direct physical correlation between the vertical mixing process of pollutants and the turbulent characteristics of the flow field, improving the physical realism and prediction accuracy of the simulation of the vertical concentration distribution of pollutants in the turbulent zone below the lock and the tributary inflow zone. Thirdly, this invention constructs a three-dimensional digital model base plate, establishes a three-dimensional CFD model based on the base plate, and uses GPU parallel acceleration and dynamic mesh encryption technology based on concentration front gradient for inference calculation. Finally, it realizes three-dimensional visualization rendering of pollutant movement trajectory and output of layered water intake decision information in the digital twin platform. It constructs a complete technical link from data acquisition, model building, data assimilation, accelerated inference to decision support, and solves the problems of rough geometric boundaries, insufficient turbulence capture capability, low computational efficiency and unintuitive decision information in the existing technology. It realizes high-precision simulation and rapid emergency response of three-dimensional dynamic diffusion of pollutants under complex hydrological conditions of canals, and provides accurate, intuitive and operable decision support for the in-depth protection of sensitive water sources. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology provided in this embodiment of the invention; Figure 2 The structural block diagram of the three-dimensional dynamic simulation system for pollutant diffusion in complex canal hydrology provided in this embodiment of the invention. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0021] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] Example 1 like Figure 1 As shown, this invention provides a three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology, comprising: S1, acquiring multi-source data for the target flight segment, and constructing a three-dimensional digital model base plate based on the multi-source data, including: S11 uses multibeam bathymetry to conduct underwater topographic surveys of the target section and obtain underwater topographic data; S12 uses three-dimensional laser scanning to scan the lock chamber and stilling sill of the ship lock, obtains the geometric data of the lock and stilling structure, and performs noise reduction processing on the metal structure. S13 utilizes real-time dynamic carrier phase difference technology to provide a spatial positioning reference for multibeam bathymetry and three-dimensional laser scanning, unifies the underwater topographic data and the geometric data of the lock and energy dissipation structure into the same coordinate system and performs registration, generating a three-dimensional digital model base plate containing underwater topography and shore-side structures.

[0023] The multi-source data includes underwater topographic data, geometric data of locks and energy dissipation structures, and real-time hydrological and water quality data collected from monitoring points. The real-time hydrological and water quality data includes at least vertical velocity stratification data, water level, and pollutant concentration.

[0024] In one possible implementation, in the target canal section, a survey vessel equipped with a multibeam echo sounder performs reciprocating scans along survey lines perpendicular to the channel direction to acquire raw point cloud data of the underwater topography within the section. In the lock area, a 3D laser scanner is deployed to perform substation scans of the lock chamber sidewalls, gates, and stilling sills to acquire point cloud data of the structural surfaces; specular reflection noise generated by metal structures such as gates is removed using a reflection intensity filtering method.

[0025] During the measurement process, a real-time dynamic carrier phase differential technology reference station was set up at a known control point on the shore. Simultaneously, rover receivers were installed on the transducer of the multibeam echo sounder and the 3D laser scanner to receive differential correction signals in real time, providing centimeter-level spatial positioning references for all point cloud data. The acquired underwater topographic point cloud and the denoised structural point cloud were imported into the same engineering coordinate system, and registration and fusion were performed using an iterative nearest-point algorithm. The registered point cloud underwent surface reconstruction and meshing to generate a 3D digital model base plate containing both underwater topography and shore-side structures.

[0026] Meanwhile, multiple sets of online monitoring stations were deployed at the upstream boundary of the target navigation section, the lock spillway, the tributary confluence, and the downstream water intake area. Each station is equipped with an acoustic Doppler current profiler to collect velocity data in a vertical stratified manner; a pressure level gauge to obtain water level; and a multi-parameter water quality probe to obtain water quality parameters such as turbidity and conductivity. The turbidity data was then converted into pollutant concentration values ​​through calibration conversion relationships.

[0027] This invention utilizes multibeam bathymetry to perform underwater topographic surveys of target navigation sections, solving the problem that traditional single-beam bathymetry or two-dimensional generalized models cannot accurately represent three-dimensional geomorphic features such as deep channels and shallow shoals. This improves the spatial resolution and three-dimensional morphological characterization accuracy of underwater topographic data, providing high-fidelity bottom boundary conditions for three-dimensional CFD models. Furthermore, it employs three-dimensional laser scanning to perform detailed scanning of structures such as lock chambers and stilling sills, and performs noise reduction processing on metal structures. This solves the problems of traditional measurement methods struggling to accurately acquire the geometry of complex hydraulic structures and the introduction of noise from specular reflections on metal surfaces, improving the authenticity and completeness of the geometric data of shore-side structures and ensuring that the boundary conditions of the simulated flow field in the downstream turbulent zone are consistent with the actual engineering conditions. Finally, it uses real-time dynamic carrier phase difference technology to provide centimeter-level spatial positioning references for multibeam bathymetry and three-dimensional laser scanning, and unifies the two types of data to the same coordinate system for registration. This solves the problem of model splicing misalignment in the water-land interface area caused by inconsistencies in coordinate systems between underwater topographic data and shore-side structure data, improving the spatial integrity of the three-dimensional digital model base and the accuracy of the water-land connection.

[0028] S2, establish a three-dimensional CFD model of the pollutant diffusion sensitive section based on the three-dimensional digital model base plate; The 3D CFD model includes a hydrodynamic module and a water quality module; The hydrodynamic module is built on the Navier-Stokes equations and is equipped with a large eddy simulation turbulence model to capture three-dimensional turbulence, the VOF method to track the free surface, and dynamic mesh technology to simulate gate opening and closing and ship movement. The water quality module is constructed based on the convection-diffusion equation. The vertical turbulent diffusion coefficient in the equation is parameterized by the turbulent kinetic energy and turbulent kinetic energy dissipation rate obtained by the hydrodynamic module to characterize the influence of three-dimensional turbulence caused by lock discharge and tributary inflow on the vertical mixing of pollutants.

[0029] Sensitive sections for pollutant diffusion include the turbulent mixing zone downstream of the lock spillway, the backflow mixing zone at the confluence of tributaries, the deep pool area of ​​the main channel, and the intake area of ​​sensitive downstream water sources.

[0030] In one possible implementation, based on the 3D digital model base plate generated in step S1, a 500-meter section extending from downstream of the lock spillway to upstream of the downstream intake is selected as a pollutant diffusion-sensitive section. The underwater topographic surface of this section, along with the surface meshes of the lock chamber and stilling sill structure, are extracted from the base plate to construct the model's computational domain. The inlet of the computational domain is set upstream of the lock spillway, and the outlet is set downstream of the intake.

[0031] Mesh generation was performed within the computational domain: unstructured tetrahedral meshes were used for discretization, and prismatic layer meshes were set in the near-wall region to analyze the boundary layer flow; the mesh scale was refined to 0.2 to 0.5 meters in the jet zone of the lock spillway, the backflow zone downstream of the stilling sill, the mixing zone at the tributary confluence, and the bottom of the deep channel, and the mesh scale was 2 to 5 meters in the straight section of the channel, with a total of approximately 5 million to 8 million mesh elements.

[0032] A 3D CFD model was constructed on the divided mesh, specifically including the configuration of hydrodynamic and water quality modules. The hydrodynamic module uses the incompressible Navier-Stokes equations as the governing equations, a large eddy simulation model for turbulent closure is used, and subgrid stress is calculated using the Smagorinsky-Lilly model with model coefficients set to Cs = 0.1. Free surface tracking employs the VOF method, and the volume fraction transport equation is solved using a geometric reconstruction scheme to accurately capture the rapid changes in the free surface during gate discharge. The gate opening and closing process is implemented using a layered dynamic mesh technique, with the gate movement speed set to a uniform opening and closing speed of 0.02 m / s, and the dynamic mesh region updated in real time according to the gate position. Ship movement is handled using a sliding mesh technique, translating the mesh nodes in the ship's location area according to the actual speed and heading data provided by the Automatic Identification System (AIS). The model inlet boundary is given flow boundary conditions, the outlet boundary is given water level boundary conditions, the bottom and structure surfaces are set as non-slip wall boundaries, and the top surface is set as a pressure outlet boundary.

[0033] Based on the velocity field solved by the hydrodynamic module, the water quality module adopts the convection-diffusion equation as the control equation for pollutant transport. The velocity component in the convection term is directly taken from the three-dimensional velocity field output at each computation time step of the hydrodynamic module; the horizontal diffusion coefficient in the diffusion term is the sum of the molecular diffusion coefficient and the subgrid eddy diffusion coefficient, while the vertical turbulent diffusion coefficient is provided in real-time by the hydrodynamic module in the following manner: After obtaining the turbulent kinetic energy k and turbulent kinetic energy dissipation rate ε of each grid node at each computation time step of the hydrodynamic module, parameterization calculation is performed according to the following formula: ; Where Dz is the vertical turbulent diffusion coefficient and Cμ is an empirical constant, which is taken as 0.09 in this embodiment. This parameterization method transmits the three-dimensional turbulent characteristics generated by the lock discharge jet and the tributary confluence backflow to the water quality module in real time through turbulent kinetic energy and dissipation rate, so that the vertical mixing process of pollutants is directly determined by the actual turbulent intensity of the flow field, replacing the empirical vertical diffusion coefficient setting in the traditional model.

[0034] The hydrodynamic module and the water quality module are coupled and solved using the same computational grid and time step. In each time step, the hydrodynamic field is solved iteratively until convergence. Then, the updated velocity field, turbulent kinetic energy field and turbulent kinetic energy dissipation rate field are transferred to the water quality module to solve the pollutant concentration field. After completing the calculation of the current time step, the next time step is entered.

[0035] This invention establishes a 3D CFD model of pollutant diffusion-sensitive waterways based on a 3D digital model base, solving the problem of local flow field simulation distortion caused by rough geometric boundaries in traditional 2D or simplified 3D models. This improves the geometric fidelity and overall accuracy of the 3D CFD model in simulating flow and concentration fields under complex boundary conditions such as lock spillways and tributary inlets. Furthermore, by configuring a large eddy simulation (LES) turbulence model in the hydrodynamic module to replace the traditional Reynolds averaged (REM) model, it addresses the Reynolds averaged (REM) model's insufficient ability to capture large-scale transient vortex structures such as lock spillway jets and tributary inlet backflows. This improves the spatiotemporal analytical accuracy of turbulent kinetic energy and turbulent kinetic energy dissipation rate in the 3D turbulent field, providing a realistic turbulent parameter field for refined simulation of the vertical mixing process of pollutants. Finally, by configuring the VOF method in the hydrodynamic module to track the free surface, it solves the problem that static pressure assumption models or rigid cover assumption models cannot simulate the rapid changes in the free surface during lock spillway discharge. This improves the simulation realism of instantaneous water surface fluctuations and aeration processes during gate opening and closing; by configuring dynamic mesh technology in the hydrodynamic module to simulate gate opening and closing and ship movement, it solves the problem that fixed mesh models cannot reflect the instantaneous flow field disturbances caused by gate movement and the impact of ship waves generated by ship navigation on pollutant diffusion paths, thus improving the completeness and accuracy of flow field and pollutant transport simulation under dynamic boundary conditions; by using real-time parameterization of the turbulent kinetic energy and turbulent kinetic energy dissipation rate obtained by solving the vertical turbulent diffusion coefficient in the water quality module from the hydrodynamic module, it solves the problem that the vertical diffusion coefficient in traditional water quality models depends on empirical constants or empirical formulas and cannot reflect the differences in local turbulent intensity in the flow field. It establishes a direct physical correlation between the vertical mixing process and the actual turbulent characteristics of the flow field, reduces the simulation uncertainty caused by subjective setting of empirical parameters, and improves the physical realism and prediction accuracy of the vertical concentration distribution simulation of pollutants in the lock jet area and tributary confluence area.

[0036] S3 employs an ensemble Kalman filter algorithm to assimilate the acquired multi-source data into a 3D CFD model, updating the flow velocity, water level, and pollutant concentration state variables of all computational grid nodes. It uses measured upstream flow and downstream water level as time-varying boundary conditions to drive the 3D CFD model for real-time rolling calculations, including: S31, based on the flow velocity, water level and pollutant concentration analysis field of all grid nodes calculated by the three-dimensional CFD model at the previous moment, a set of state variables is generated by the Monte Carlo method as the background field at the current moment; S32, each set member in the background field is input into the three-dimensional CFD model for one-step prediction calculation to obtain the set of predicted values ​​of flow velocity, water level and pollutant concentration on all grid nodes at the current moment; S33: Obtain the vertical flow velocity stratification data, water level and pollutant concentration collected from the monitoring points in the multi-source data at the current moment, form the observation field, and set the observation error covariance matrix; S34, calculate the background field error covariance matrix based on the predicted value set, and calculate the Kalman gain matrix in combination with the observation error covariance matrix; S35, using the Kalman gain matrix, the information of the observation field is fused into the predicted value of each set member to obtain the updated set of analytical values ​​of flow velocity, water level and pollutant concentration on all grid nodes; S36, take the mean of the set of analyzed values ​​as the assimilated state variable at the current moment, and update the flow velocity, water level and pollutant concentration of all computational grid nodes in the three-dimensional CFD model.

[0037] S37, Obtain the measured upstream flow rate and downstream water level value at the current moment from the multi-source data; and convert the upstream flow rate value into the velocity distribution boundary condition of the inlet section of the three-dimensional CFD model, and convert the downstream water level value into the water level boundary condition of the outlet section of the three-dimensional CFD model. S38 uses the assimilated state variables as the initial field and the transformed velocity distribution and water level boundary conditions as time-varying boundary conditions to drive the three-dimensional CFD model to calculate one time step forward, and obtain the predicted fields of velocity, water level and pollutant concentration on all computational grid nodes at the next moment. S39, the predicted field is used as the background field for a new round of ensemble Kalman filter assimilation, and S37 to S38 are repeated to achieve real-time rolling calculation of the three-dimensional CFD model.

[0038] In one possible implementation, after the 3D CFD model is constructed and the initial field is set, a real-time rolling calculation phase begins, executing cyclically with an assimilation cycle of 15 minutes. Based on the analysis field output by the 3D CFD model after assimilation and update at the previous time, denoted as t-1, five state variables are extracted from all computational grid nodes: velocity components (u, v, w), water level, and pollutant concentration. A Gaussian perturbation with zero mean and variance determined by the model prediction error is applied to this analysis field. Fifty ensemble members are generated using the Monte Carlo method, constituting the background field at the current time, i.e., time t. Each ensemble member is a column vector containing all state variables from all grid nodes.

[0039] The 50 members of the background field are input one by one as initial conditions into the 3D CFD model. The boundary conditions of the previous time step drive the model to integrate forward by one time step, which is 15 minutes in this embodiment, to obtain the predicted values ​​of flow velocity, water level, and pollutant concentration at all grid nodes at time t, totaling 50 sets of predicted fields. Vertical stratified flow velocity data from acoustic Doppler current profilers, water level values ​​from pressure level gauges, and turbidity data from multi-parameter water quality probes are obtained from each monitoring station at time t and converted into pollutant concentrations. The above measured data are arranged into observation vectors according to the spatial location of the monitoring points. At the same time, the observation error covariance matrix R is set according to the nominal accuracy of each sensor: the variance of flow velocity observation error is 0.01 m² / s², the variance of water level observation error is 0.0025 m², and the variance of concentration observation error is 0.04.

[0040] Based on the predicted values ​​of the 50 ensemble members, calculate the background field error covariance matrix among the state variables of all grid nodes. Combined with the observation error covariance matrix R set in step S33, calculate the Kalman gain matrix K using the following formula: K = Pb·H T · (H · Pb · H T + R) -1 ; Where Pb is the background field error covariance matrix, and H is the observation operator that maps the grid node state variables to the monitoring point locations. T This is the transpose of H. The observation operator H uses linear interpolation to interpolate the state variables of the four grid nodes closest to the monitoring point to the monitoring point location using inverse distance weighted interpolation.

[0041] Using the Kalman gain matrix K, the observed field information is fused into the prediction value of each ensemble member. For the i-th ensemble member, it is updated as follows: xi a = xi f + K · (y obs - H · xi f ); Among them, xi f Let y be the predicted value vector of the i-th set member. obs Let xi be the observation vector. a Let be the updated analysis value vector for the i-th set member. Iterate through all 50 set members to obtain the updated analysis value set.

[0042] The mean value of the 50 analytical values ​​obtained in step S35 is calculated, and this mean vector is taken as the assimilated state variable at time t. The flow velocity, water level, and pollutant concentration of all computational grid nodes in the 3D CFD model are updated to the corresponding values ​​of this mean vector, completing the state assimilation at time t. The measured upstream cross-sectional flow rate and downstream cross-sectional water level at time t are extracted from the multi-source data. In this embodiment, the upstream cross-sectional flow rate is 120 cubic meters per second, and the downstream cross-sectional water level is 32.50 meters. The upstream flow rate is weighted according to the grid area of ​​the inlet cross-section and converted into the normal velocity distribution of each grid surface on the inlet cross-section, completing the flow-velocity boundary transformation. The downstream water level is directly assigned as a pressure boundary condition to all grid nodes below the water surface on the outlet cross-section, completing the water level-pressure boundary transformation.

[0043] Using the updated state variables as the initial field, and the transformed inlet velocity distribution and outlet water level as time-varying boundary conditions, the 3D CFD model is driven to perform one step forward with a time step of 15 minutes, obtaining the predicted fields of velocity, water level, and pollutant concentration at all computational grid nodes at time t+1. The predicted field at time t+1 is then used as the background field for the next round of ensemble Kalman filter assimilation. Using the measured upstream flow and downstream water level at time t+1 as new time-varying boundary conditions, steps S37 to S38 are repeated to form a continuous rolling loop, achieving real-time rolling calculation of the 3D CFD model.

[0044] This invention employs an ensemble Kalman filter algorithm to assimilate real-time multi-source data into a 3D CFD model and updates the state variables of flow velocity, water level, and pollutant concentration at all computational grid nodes. This solves the problem of long-term accumulation of extrapolation errors caused by the inability of traditional model state variables to dynamically integrate with measured data, thus improving the consistency between model extrapolation trajectories and measured data. Furthermore, by generating a set of state variables as a background field using the Monte Carlo method and performing forward prediction calculations on each set member, this invention addresses the difficulty in analytically obtaining the background field error covariance in the high-dimensional nonlinear state space of 3D CFD models. This improves the accuracy of error statistics in Kalman gain matrix calculation and the constraint capability of the assimilation process on complex flow field states. Finally, by constructing a model containing vertical velocity stratification, water level, and pollutant concentration... The Kalman gain matrix is ​​calculated by combining the background field error covariance matrix and the observation error covariance matrix with the pollutant concentration observation field. This solves the problem that traditional interpolation or correction methods cannot perform optimal weight allocation based on the confidence levels of the model and observations, improves the physical coordination of observation information propagation to the entire computational domain, and reduces unreasonable disturbances introduced by local observation updates to the global flow field and concentration field. By converting measured upstream flow and downstream water level into time-varying boundary conditions and performing a rolling calculation of forecast-assimilation-forecast in each assimilation cycle, the problem of the inlet and outlet conditions of the model being distorted due to the inability of preset fixed boundary conditions to respond to actual hydrological changes is solved. This improves the model's real-time response capability to external flow and water level fluctuations and ensures the authenticity and timeliness of boundary conditions during emergency simulations.

[0045] S4 utilizes GPU parallel acceleration technology to perform simulation calculations on the assimilated 3D CFD model. In regions where the pollutant concentration front gradient exceeds a preset concentration threshold, the computational grid is dynamically refined. The measured and calculated values ​​from monitoring points are compared to dynamically calibrate the roughness coefficient and vertical turbulent diffusion coefficient, including: S41 distributes the computational tasks of the assimilated and updated 3D CFD model to multiple GPU computing cores, and solves the Navier-Stokes equations, the large eddy simulation turbulence model and the convection-diffusion equations in parallel to obtain the flow velocity, turbulent kinetic energy, turbulent kinetic energy dissipation rate and pollutant concentration at each computing grid node. S42, traverse all computational grid nodes, calculate the pollutant concentration gradient value between adjacent nodes, compare it with a preset concentration threshold, and mark the grid area whose gradient value exceeds the preset concentration threshold as the area to be encrypted; S43, refine the grid within the area to be encrypted, increase the number of grid nodes, and interpolate and map the flow field variables and pollutant concentration field variables on the original grid nodes to the newly generated grid nodes to generate the encrypted computational grid; S44, continue to perform the deduction calculation on the encrypted computing grid, and repeat S42 and S43 at each computing time step until the deduction calculation ends.

[0046] In one possible implementation, the computational task of the assimilated and updated 3D CFD model is deployed on a computing node equipped with 8 GPUs. Using the CUDA parallel programming framework, the 8 million grid cells within the computational domain are spatially decomposed into 8 subdomains, each assigned to a GPU core. Each GPU core is independently responsible for solving the pressure-velocity coupled Navier-Stokes equations within its subdomain, calculating the subgrid eddy viscosity coefficient in the large eddy simulation turbulence model, and updating the contaminant concentration in the convection-diffusion equations. Data exchange at the subdomain boundaries is accomplished through the NVLink high-speed interconnect channel between GPUs, with data synchronization performed once per iteration step. After parallel solving, the three components of velocity (u, v, w), turbulent kinetic energy k, turbulent kinetic energy dissipation rate ε, and contaminant concentration C at all computational grid nodes in the current time step are obtained.

[0047] The process iterates through all computational grid nodes, calculating the pollutant concentration difference between each node and all its neighboring nodes. This difference is then divided by the node spacing to obtain the concentration gradient. A preset concentration threshold is set at 0.1 mg / L / m, representing the ratio of the concentration difference between adjacent nodes to the node spacing. The calculated gradient value for each node is compared to the preset concentration threshold. If the gradient value between a node and any neighboring node exceeds 0.1 mg / L / m, that node and the grid cells formed by it and its neighboring nodes are marked as regions to be refined. The tetrahedral grid cells within the marked regions to be refined are further subdivided using the edge midpoint subdivision method, dividing each tetrahedral cell into 8 sub-tetrahedral cells. The total number of grid nodes increases with each refinement operation. On the newly generated grid nodes, a second-order precision least-squares interpolation method is used to interpolate and map the flow field variables (u, v, w, k, ε) and pollutant concentration field variables (C) from the original grid nodes to the newly generated nodes. This ensures the continuity and local conservation of the variable field distribution before and after refinement, generating the refined computational grid.

[0048] On the encrypted computational grid generated in step S43, the GPU parallel solution process from step S41 is invoked again, using the variable field at the encryption time as the initial field for forward extrapolation calculations. After the extrapolation is completed at each subsequent computation time step, the gradient calculation and encryption marking in step S42, and the mesh refinement and interpolation mapping operations in step S43 are repeated. When the pollutant concentration front moves to a new region, the encrypted region dynamically expands accordingly; when the concentration gradient in the region falls below a preset threshold after the front passes, the encrypted grid is restored to its original resolution, releasing computational resources. The above steps are executed cyclically until the extrapolation calculation ends or the pollutant concentration field reaches a steady state.

[0049] During the simulation calculations from S41 to S44, dynamic parameter calibration was performed simultaneously: after acquiring the measured values ​​at the monitoring points in each assimilation cycle, the calculated values ​​of the 3D CFD model at the corresponding monitoring point locations were extracted. The following objective function was constructed: J = Σ[(h cal - h obs )² / σ h ² + (C cal - C obs )² / σ C ²] Among them, h cal and h obs These are the calculated water level and the measured water level, respectively. cal and C obs σ represents the calculated concentration and the measured concentration, respectively. h ² and σ C ² represents the variance of water level and concentration observation errors, respectively. The Nelder-Mead simplex optimization algorithm is used, with the empirical constants Cμ of the roughness coefficient n and the vertical turbulent diffusion coefficient Dz as optimization variables, to minimize the objective function J. After optimization convergence, the updated values ​​of n and Cμ are applied to the extrapolation calculations for the next time period, completing the dynamic calibration of the roughness and vertical turbulent diffusion coefficients.

[0050] This invention addresses the problems of long simulation times for high-resolution 3D CFD models in traditional serial computing modes, such as the inability of fixed meshes to focus on high-gradient regions of pollutant fronts, and the inability of model parameters to adaptively adjust based on empirical values, by combining GPU parallel acceleration technology with dynamic mesh refinement and parameter dynamic calibration. It significantly reduces the computation time for emergency simulations while maintaining simulation accuracy, improving the model's rapid response to sudden pollution incidents and the accuracy of adaptive simulations of pollutant transport processes under complex hydrological conditions. By traversing the computational mesh nodes, calculating the pollutant concentration front gradient and comparing it with a preset threshold, and refining the mesh and interpolating variables for regions exceeding the threshold, this invention repeats the operation at each computation time step, solving the problem of fixed-resolution meshes being insufficient for pollutant... This approach addresses the issues of wasted computational resources in areas not reached by the front and insufficient grid resolution in frontal regions to accurately capture abrupt concentration changes. It improves the resolution of the spatial location of the pollutant diffusion front and the attenuation process of concentration peaks, while reducing computational resource consumption in non-critical areas, achieving a dynamic balance between computational accuracy and computational cost. By comparing measured values ​​from monitoring points with model-calculated values, an objective function is constructed, and optimization algorithms are used to dynamically calibrate the roughness coefficient and vertical turbulent diffusion coefficient. This solves the problem of increased model-actual deviation caused by fixed parameters being unable to adapt to time-varying conditions such as changes in lock discharge and fluctuations in tributary inflows. It reduces the subjective influence of empirical parameter values ​​on simulation results and improves the model's adaptability to changing environments and the reliability of concentration prediction results during long-duration simulations.

[0051] S5 visualizes the simulation results in three dimensions on the digital twin platform, renders the trajectory of pollutants, and outputs the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

[0052] In one possible implementation, the digital twin platform is built on the Unity engine, pre-loading a 3D digital model base plate generated by S1 as the scene foundation, including underwater topographic surfaces, lock chambers, and stilling sill structures. The platform receives the time-step data streams from the simulation calculations via the WebSocket protocol, parses them, and performs 3D visualization rendering. In the 3D scene, the water area is divided into semi-transparent rendering layers. The pollutant concentration field is displayed using volume rendering technology, with a blue-to-red gradient band mapping the concentration value range, where blue represents the background concentration and red represents the peak concentration. Users can freely rotate the viewpoint and section the 3D cloud map in the platform interface to observe the 3D spatial distribution of pollutant clumps in the deep channel area and the turbulent area below the lock.

[0053] The trajectory of pollutants is visualized using the Lagrange particle tracing method. Virtual tracer particles are released at a rate of 50 per second at the leak source location. Each particle undergoes advection based on the instantaneous flow velocity of its grid node, and the particle paths accumulate over time to form a trajectory line. The trajectory line is rendered with a gradient from cool to warm colors according to time sequence, visually demonstrating the three-dimensional transport path of pollutants migrating from the leak source to the water intake. Simultaneously, the measured concentration values ​​of each monitoring station and the location markers of each water source intake are labeled in the scene. Flow field vector arrow diagrams are overlaid at key cross-sections, with the arrow direction and magnitude representing the flow direction and velocity, respectively, to help determine whether pollutants are likely to enter the water intake.

[0054] Upon receiving each frame of simulation results, the platform automatically performs threshold judgments on the pollutant concentration of each grid node containing the water intake. The warning trigger concentration is set at 0.005 mg / L. When the pollutant concentration of a grid node containing a water intake first exceeds this threshold, the platform records the current simulation time and calculates the countdown from the current real time to the warning time, displaying it on the interface's warning panel. For example, the estimated arrival time of the pollution front at water intake No. 1 is 1 hour and 23 minutes later. Simultaneously, the platform issues tiered warnings to management personnel via sound alarms and push notifications.

[0055] At the water intake cross-section, concentration sampling points are set at 0.5-meter intervals along the vertical direction. The platform extracts the pollutant concentration values ​​of the corresponding grid nodes for each sampling point and plots a concentration-depth profile curve of the water intake cross-section on the interface, with the vertical axis representing water depth and the horizontal axis representing pollutant concentration. Simultaneously, the concentration values ​​for each layer are listed in tabular form in the data panel. This profile curve is updated in real-time with simulation calculations, clearly showing the differences in pollutant concentrations among the surface, middle, and bottom layers of the water. Based on this, management personnel determine whether the water intake gate should be adjusted to the water depth range with the lowest pollutant concentration. For example, in this case, the lowest concentration is achieved when the water intake depth is set at 3.0 meters. The platform displays in the decision suggestion panel: Recommended water intake depth: 3.0 to 3.5 meters, expected contact concentration below 0.01 mg / L.

[0056] After the simulation calculations are completed, the platform supports exporting the full-time 3D visualization results as video files and automatically generating pollution incident handling reports, including early warning timelines, concentration change curves at each water intake, and stratified water intake recommendations, providing support for post-incident assessment and management documentation.

[0057] This invention solves the problem that traditional two-dimensional displays or tabular data cannot intuitively present the three-dimensional diffusion path of pollutants among complex underwater terrain and structures by performing three-dimensional visualization rendering of the simulation calculation results on a digital twin platform and simultaneously outputting the early warning time of pollutants reaching the water intake and the concentration difference at different water depths. This improves the intuitiveness and accuracy of decision-makers' perception of the pollution situation and reduces the risk of decision delays and misjudgments caused by the lack of spatial information.

[0058] To verify the accuracy and timeliness of the proposed method under complex hydrological conditions in a canal, a 5.2km section from a canal lock to the downstream water source intake was selected as the test area. This section includes the lock spillway, a tributary inlet, and two water source intakes. Intake No. 1 is 3.8km from the spillway, and Intake No. 2 is 4.9km from the spillway. A comparative experiment was conducted using a simulated pollutant leakage event in March 2025. The leakage source was located 200m downstream of the lock spillway, with a leakage volume of 50kg of conservative pollutant. The leakage lasted for 10 minutes, during which the lock discharged water normally at a flow rate of 85m³ / s, and the tributary inflow was 12m³ / s. Real-time monitoring data was obtained from five deployed online monitoring stations, including vertical velocity stratification, water level, and pollutant concentration.

[0059] Using the method of this invention, namely 3D CFD, ensemble Kalman filter assimilation, GPU parallel acceleration, dynamic mesh refinement, and turbulent parameterized vertical diffusion coefficient as the experimental group, two control groups were set up: Control group 1 uses a traditional two-dimensional hydrodynamic-water quality model with an empirical constant of 0.01 m² / s for the vertical diffusion coefficient. Data assimilation is not possible, the grid is fixed, and a single CPU is used for computation.

[0060] Control group 2 uses a three-dimensional CFD model, but the vertical diffusion coefficient is calculated using the empirical formula D. z =0.067u×h, where u is the frictional velocity and h is the water depth. No data assimilation is used, the grid is fixed, and single-GPU computation is used.

[0061] The experimental group initially calculated a grid of approximately 6.2 million elements, using eight A100 GPUs, an ensemble Kalman filter with 50 members, and an assimilation period of 15 minutes. Control group 1 had approximately 12,000 grid elements, and control group 2 had approximately 5.8 million grid elements.

[0062] The accuracy of pollutant arrival time prediction was compared. The time when the actual pollutant concentration at water intake No. 1 first exceeded 0.005 mg / L was taken as the reference time and recorded as 0 min. The arrival time deviations predicted by the three methods were statistically analyzed, and the results are shown in Table 1.

[0063] Table 1: Comparison of Predictive Accuracy of Pollutant Arrival Time at Water Intake As shown in Table 1, the prediction error of the pollutant front arrival time by the method of the present invention is only 1.2 minutes, which is far better than -12.7 minutes of control group 1 and -6.5 minutes of control group 2; the relative error of the concentration peak is also significantly reduced from 46.1% and 22.4% to 8.3%, indicating that data assimilation and turbulence parameterization effectively suppress error accumulation and improve the reliability of the inference.

[0064] At the No. 1 water intake section, peak pollutant concentrations were measured at the surface (0.5m depth), middle layer (3.0m depth), and bottom layer (5.5m depth) during the passage of the pollution plume. Table 2 shows the comparison between simulated and measured values ​​for each method. Table 2: Simulation Accuracy of Vertical Concentration Distribution at Water Intake Point The method of this invention accurately reproduces the vertical stratification characteristics of high surface concentration and low bottom concentration, with errors in each layer within ±7%. Control group 1, due to its two-dimensional vertical averaging assumption and constant diffusion coefficient, completely fails to reflect the low-concentration region at the bottom, resulting in a relative error of 171.9% in the bottom layer prediction. Although control group 2 is a three-dimensional model, its empirical diffusion coefficient cannot reflect the strong turbulence generated by the lock jet, leading to an overestimation of bottom-layer diffusion and an error of 65.6%. This verifies the advantages of the physical mechanism of this invention, which parameterizes the vertical turbulent diffusion coefficient through turbulent kinetic energy and dissipation rate.

[0065] Example 2 like Figure 2 As shown, the present invention also provides a three-dimensional dynamic simulation system for pollutant diffusion in complex canal hydrology. The system is used to implement the three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology described in any of Embodiment 1. The system includes: The data acquisition module is used to acquire multi-source data of the target flight segment and construct a three-dimensional digital model base plate based on the multi-source data; The 3D model building module is used to build 3D CFD models of pollutant diffusion-sensitive flight segments based on a 3D digital model base. The rolling drive module is used to assimilate the acquired multi-source data into the three-dimensional CFD model using the ensemble Kalman filter algorithm, update the flow velocity, water level and pollutant concentration state variables of all computational grid nodes, and drive the three-dimensional CFD model to perform real-time rolling calculations using the measured upstream flow and downstream water level as time-varying boundary conditions. The deduction and optimization module is used to perform deduction calculations on the assimilated three-dimensional CFD model through GPU parallel acceleration technology. In areas where the pollutant concentration front gradient exceeds the preset concentration threshold, the calculation grid is dynamically densified, and the measured values ​​and calculated values ​​of monitoring points are compared to dynamically calibrate the roughness coefficient and vertical turbulent diffusion coefficient. The visualization module is used to visualize the results of the simulation calculation in three dimensions on the digital twin platform, render the movement trajectory of pollutants, and output the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

[0066] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology, characterized in that, include: S1, acquire multi-source data of the target flight segment, and construct a three-dimensional digital model base plate based on the multi-source data; S2, establish a three-dimensional CFD model of the pollutant diffusion sensitive section based on the three-dimensional digital model base plate; S3 employs an ensemble Kalman filter algorithm to assimilate the acquired multi-source data into a three-dimensional CFD model, update the flow velocity, water level, and pollutant concentration state variables of all computational grid nodes, and use the measured upstream flow and downstream water level as time-varying boundary conditions to drive the three-dimensional CFD model to perform real-time rolling calculations. S4 uses GPU parallel acceleration technology to perform simulation calculations on the assimilated three-dimensional CFD model. In areas where the pollutant concentration front gradient exceeds the preset concentration threshold, the computational grid is dynamically densified, and the measured values ​​and calculated values ​​of the monitoring points are compared to dynamically calibrate the roughness coefficient and the vertical turbulent diffusion coefficient. S5 visualizes the simulation results in three dimensions on the digital twin platform, renders the trajectory of pollutants, and outputs the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

2. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 1, characterized in that, The multi-source data includes underwater topographic data, geometric data of locks and energy dissipation structures, and real-time hydrological and water quality data collected from monitoring points. The real-time hydrological and water quality data includes at least vertical velocity stratification data, water level, and pollutant concentration.

3. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 1, characterized in that, The three-dimensional CFD model includes a hydrodynamic module and a water quality module; The hydrodynamic module is built on the Navier-Stokes equations and is equipped with a large eddy simulation turbulence model to capture three-dimensional turbulence, the VOF method to track the free surface, and dynamic mesh technology to simulate gate opening and closing and ship movement. The water quality module is constructed based on the convection-diffusion equation. The vertical turbulent diffusion coefficient in the equation is parameterized by the turbulent kinetic energy and turbulent kinetic energy dissipation rate obtained by the hydrodynamic module to characterize the influence of three-dimensional turbulence caused by lock discharge and tributary inflow on the vertical mixing of pollutants.

4. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 1, characterized in that, The pollutant diffusion sensitive sections include the turbulent mixing zone downstream of the lock spillway, the backflow mixing zone at the tributary confluence, the deep pool area of ​​the main channel, and the intake area of ​​the downstream sensitive water source.

5. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 1, characterized in that, The acquisition of multi-source data for the target flight segment and the construction of a three-dimensional digital model base plate based on the multi-source data include: S11 uses multibeam bathymetry to conduct underwater topographic surveys of the target section and obtain underwater topographic data; S12 uses three-dimensional laser scanning to scan the lock chamber and stilling sill of the ship lock, obtains the geometric data of the lock and stilling structure, and performs noise reduction processing on the metal structure. S13 utilizes real-time dynamic carrier phase difference technology to provide a spatial positioning reference for multibeam bathymetry and three-dimensional laser scanning, unifies the underwater topographic data and the geometric data of the lock and energy dissipation structure into the same coordinate system and performs registration, generating a three-dimensional digital model base plate containing underwater topography and shore-side structures.

6. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 2, characterized in that, The ensemble Kalman filter algorithm is used to assimilate the acquired multi-source data into the three-dimensional CFD model, updating the flow velocity, water level, and pollutant concentration state variables of all computational grid nodes, including: S31, based on the flow velocity, water level and pollutant concentration analysis field of all grid nodes calculated by the three-dimensional CFD model at the previous moment, a set of state variables is generated by the Monte Carlo method as the background field at the current moment; S32, each set member in the background field is input into the three-dimensional CFD model for one-step prediction calculation to obtain the set of predicted values ​​of flow velocity, water level and pollutant concentration on all grid nodes at the current moment; S33: Obtain the vertical flow velocity stratification data, water level and pollutant concentration collected from the monitoring points in the multi-source data at the current moment, form the observation field, and set the observation error covariance matrix; S34, calculate the background field error covariance matrix based on the predicted value set, and calculate the Kalman gain matrix in combination with the observation error covariance matrix; S35, using the Kalman gain matrix, the information of the observation field is fused into the predicted value of each set member to obtain the updated set of analytical values ​​of flow velocity, water level and pollutant concentration on all grid nodes; S36, take the mean of the set of analyzed values ​​as the assimilated state variable at the current moment, and update the flow velocity, water level and pollutant concentration of all computational grid nodes in the three-dimensional CFD model.

7. The method for three-dimensional dynamic simulation of pollutant diffusion in complex canal hydrology according to claim 6, characterized in that, The method of using measured upstream flow and downstream water level as time-varying boundary conditions to drive the three-dimensional CFD model to perform real-time rolling calculations includes: S37, Obtain the measured upstream flow rate and downstream water level value at the current moment from the multi-source data; and convert the upstream flow rate value into the velocity distribution boundary condition of the inlet section of the three-dimensional CFD model, and convert the downstream water level value into the water level boundary condition of the outlet section of the three-dimensional CFD model. S38 uses the assimilated state variables as the initial field and the transformed velocity distribution and water level boundary conditions as time-varying boundary conditions to drive the three-dimensional CFD model to calculate one time step forward, and obtain the predicted fields of velocity, water level and pollutant concentration on all computational grid nodes at the next moment. S39, the predicted field is used as the background field for a new round of ensemble Kalman filter assimilation, and S37 to S38 are repeated to achieve real-time rolling calculation of the three-dimensional CFD model.

8. The three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology according to claim 3, characterized in that, The process involves using GPU parallel acceleration technology to perform extrapolation calculations on the assimilated 3D CFD model, and dynamically refining the computational grid in regions where the pollutant concentration front gradient exceeds a preset concentration threshold, including: S41 distributes the computational tasks of the assimilated and updated 3D CFD model to multiple GPU computing cores, and solves the Navier-Stokes equations, the large eddy simulation turbulence model and the convection-diffusion equations in parallel to obtain the flow velocity, turbulent kinetic energy, turbulent kinetic energy dissipation rate and pollutant concentration at each computing grid node. S42, traverse all computational grid nodes, calculate the pollutant concentration gradient value between adjacent nodes, compare it with a preset concentration threshold, and mark the grid area whose gradient value exceeds the preset concentration threshold as the area to be encrypted; S43, refine the grid within the area to be encrypted, increase the number of grid nodes, and interpolate and map the flow field variables and pollutant concentration field variables on the original grid nodes to the newly generated grid nodes to generate the encrypted computational grid; S44, continue to perform the deduction calculation on the encrypted computing grid, and repeat S42 and S43 at each computing time step until the deduction calculation ends.

9. A three-dimensional dynamic simulation system for pollutant diffusion in complex canal hydrology, characterized in that, The system is used to implement the three-dimensional dynamic simulation method for pollutant diffusion in complex canal hydrology as described in any one of claims 1 to 8, and the system comprises: The data acquisition module is used to acquire multi-source data of the target flight segment and construct a three-dimensional digital model base plate based on the multi-source data; The 3D model building module is used to build 3D CFD models of pollutant diffusion-sensitive flight segments based on a 3D digital model base. The rolling drive module is used to assimilate the acquired multi-source data into the three-dimensional CFD model using the ensemble Kalman filter algorithm, update the flow velocity, water level and pollutant concentration state variables of all computational grid nodes, and drive the three-dimensional CFD model to perform real-time rolling calculations using the measured upstream flow and downstream water level as time-varying boundary conditions. The deduction and optimization module is used to perform deduction calculations on the assimilated three-dimensional CFD model through GPU parallel acceleration technology. In areas where the pollutant concentration front gradient exceeds the preset concentration threshold, the calculation grid is dynamically densified, and the measured values ​​and calculated values ​​of monitoring points are compared to dynamically calibrate the roughness coefficient and vertical turbulent diffusion coefficient. The visualization module is used to visualize the results of the simulation calculation in three dimensions on the digital twin platform, render the movement trajectory of pollutants, and output the early warning time of pollutants reaching the water intake of each water source and the difference in pollutant concentration at different water depths at the water intake, so as to provide decision support for stratified water intake.

Images