River water and sediment transportation process numerical simulation method based on multi-GPU parallel framework

Abstract

The invention discloses a numerical simulation method for a river water and sediment transportation process based on a multi-GPU parallel framework. The method specifically comprises the following steps: dividing a calculation region into triangular grid calculation regions with the same number as GPUs; the variables are initialized; creating two CUDA flows to control parallel calculation of the two GPUs, and storing data of each river channel into the corresponding GPU; directly carrying out data communication between the GPUs; in each GPU, calculating a source item and flux of each triangular grid to obtain a water depth value, and calculating a water level value of each grid according to the water depth; according to the obtained water level value, water level result data calculated in each GPU are stored in a CPU to complete water level data merging; and calculating to obtain the water level values of each river reach and the outlet section of the river at each moment. According to the river sediment transportation process simulation method based on triangular grid multi-GPU parallel computing, the water and sediment transportation process of the river area can be efficiently simulated, and the water level value of the section can be rapidly predicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of numerical simulation of river water and sediment transport processes, and relates to a numerical simulation method for river water and sediment transport processes based on a multi-GPU parallel framework. Background Art

[0002] Regarding the process of river sediment transport, on the one hand, the process of river sediment transport involves the processes of suspended sediment and bed load transport, and the movement process includes the processes of sediment scour and deposition changes; on the other hand, the movement of river sediment transport will change the topography of the river channel and riverbed, and the complex topographic changes of the river channel and riverbed will cause strong undulating oscillatory evolution movements of the water flow in some reaches of the river. It is necessary to focus on studying the sediment transport processes in these areas. Therefore, by studying the process of river sediment transport, the laws of river sediment transport movement and flood risks can be accurately evaluated. Especially in the Yellow River Basin, the processes of sediment scour and deposition changes will change the topographic conditions of the riverbed, and then affect the changes in the water levels of the rivers in the lower reaches of the Yellow River, which may lead to a high risk of flood disasters in the lower reaches of the river; at the same time, it may threaten the living and housing property safety of the citizens in the entire downstream area, and also damage the water ecological environment of the cities in the lower reaches of the river. In summary, it is very important to study the process of river sediment transport, which can not only analyze the mechanism of the river sediment transport process in detail, but also provide a reliable basis and scientific guidance for river flood planning and design and flood response management plans.

[0003] Regarding the methods for studying the process of river sediment transport, scholars at home and abroad mainly use mathematical models to simulate and analyze the process of river water and sediment transport. The numerical model of river water and sediment transport process based on hydrodynamic model has been applied by many scholars, which can accurately realize the dynamic simulation of the water and sediment transport process. By establishing a water and sediment model, the flood evolution process of river water and sediment movement is analyzed and considered, and the spatio-temporal distribution of river water and sediment transport movement is explored. In terms of simulation functions, the existing numerical models mainly simulate the sediment transport process of single river bed sediment or suspended sediment; in terms of application, the existing numerical models mostly simulate the river water and sediment transport process without considering the simulation and evaluation combining the river water and sediment transport process and flood risk; in terms of computational efficiency, although some scholars have adopted the single GPU parallel computing technology in the river water and sediment transport process, the simulation performance in large-scale river areas is low. In terms of computational grids, the existing multi-GPU numerical calculation methods are mainly based on the multi-GPU calculation method of structured grid calculation. The calculation units of structured grids cannot more accurately reflect the characteristics of key fine terrain. Therefore, the existing methods have not yet involved the multi-GPU calculation method of unstructured grids. In summary, the rapid simulation of the river water and sediment transport process is very important for the analysis of the dynamic movement process of sediment, flood risk assessment, and the change of river channel and river bed terrain environment. Therefore, it is necessary to establish a high-precision, high-performance, and multi-process coupled high-performance numerical simulation method for the river water and sediment transport process. Summary of the Invention

[0004] The purpose of the present invention is to provide a numerical simulation method for the river water and sediment transport process based on a multi-GPU parallel framework, which solves the problems of single function in the simulation process and insufficient simulation calculation performance in the existing river water and sediment transport process simulation methods.

[0005] The technical solution adopted by the present invention is a numerical simulation method for the river water and sediment transport process based on a multi-GPU parallel framework, which is specifically implemented according to the following steps:

[0006] Step 1: Divide the entire river into two river channels of the same length, and then subdivide the two calculation areas where the two river channels are located into calculation area one and calculation area two. The two adjacent columns of calculation grid areas in the middle of the two calculation areas are overlapping areas. Both calculation area one and calculation area two are composed of multiple triangular unstructured grids;

[0007] Step 2: The triangular grids on the left and right sides of the middle boundary of the overlapping area share a common edge one by one;

[0008] Step 3: Initialize the variables in the process of coupling water flow and sediment, set the total calculation duration and boundary conditions, and set the time step value of each iterative calculation to Δt;

[0009] Step 4: Create two GPUs, and store Computational Region 1 and Computational Region 2 into the corresponding GPUs respectively.

[0010] Step 5: Create two CUDA streams, allocate memory for each GPU, and copy the initialized variables from the CPU to the corresponding GPU memory respectively;

[0011] Step 6: Conduct parallel computing for the river water and sediment transport process in multiple GPUs, calculate according to the set time step Δt, complete the first time step calculation, and obtain the water level values at the downstream outlet sections of each river reach;

[0012] Step 7: Update the time step in each GPU according to the set total calculation duration, repeat Step 6 for loop calculation, until all loop calculation tasks of time steps are completed, and store the result data calculated by each GPU into the CPU for merging and then output.

[0013] Optionally, Step 2 is specifically implemented as follows: Divide the grid cells on both sides of the overlapping area into a neat column of grid cells, and the triangular grids on the left and right sides of the middle boundary of the overlapping area share one side in a one-to-one correspondence, ensuring that the interface fluxes on the shared sides of the overlapping area can be calculated correctly.

[0014] Optionally, in Step 3, the variables include: rainfall, inflow discharge, topographic elevation, Manning coefficient, bed sediment concentration, suspended sediment concentration; the boundary conditions include: time step value Δt, total duration value, upstream inlet boundary of the river, downstream outlet boundary, and boundaries on both sides of the river.

[0015] Optionally, Step 6 is specifically as follows: Conduct parallel computing in multiple GPUs, and perform data exchange and communication between GPUs for the grid calculation results on the left and right sides of the middle boundary of the overlapping area. Synchronously calculate the source terms and fluxes of each triangular grid interface on the two GPUs to obtain the water depth values, and then synchronously calculate the water depth on the two GPUs to obtain the water level values at the downstream outlet sections of each river reach.

[0016] Optionally, the parallel computing method is: Establish a coupled transport equation for water flow and sediment processes, and use the finite volume method to discretely solve the transport equation;

[0017] The transport equation is specifically expressed as:

[0018]

[0019] In the formula:

[0020] q —— is the variable vector of q x 、q y 、h, where q x 、q yThe unit-width flows in the x and y directions respectively, and h is the water depth;

[0021] f, g —— The flux vectors in the x and y directions;

[0022] S —— The source term vector;

[0023] η —— The water level value;

[0024] z b —— The topographic elevation of the river channel;

[0025] u, v —— The flow velocities in the x and y directions, where q x = uh and q y = vh;

[0026] g —— The gravity coefficient, g = 9.81 m / s 2 ;

[0027] C1, C2 —— The concentrations of suspended sediment and bed sediment;

[0028] β —— The velocity difference between the sediment deposit and the water flow;

[0029] S x1 、S x2 —— The bottom slope source terms of suspended sediment and bed sediment in the x direction;

[0030] S y1 、S y2 —— The frictional resistance source terms of suspended sediment and bed sediment in the y direction;

[0031] E1, D1 —— The deposition rate and sediment-carrying rate of suspended sediment;

[0032] E2, D2 —— The deposition rate and sediment-carrying rate of bed sediment.

[0033] Optionally, in step 7, the time step is updated and calculated on 2 GPUs, and the grid calculation results on both sides of the middle boundary of the overlapping area are exchanged and communicated between GPUs. The transport process of the coupled water flow and sediment process is synchronously and parallelly calculated on 2 GPUs until the loop calculation tasks of all time steps are completed. Then, the calculation result data of the 2 GPUs are copied and stored in the CPU. The result data includes the coordinate data, topographic elevation data, and water level data of each grid in the calculation area. Then, these data are merged, and finally, the water level value at each moment is extracted to calculate the water level values at the downstream outlet sections of each river reach for multiple durations.

[0034] The beneficial effects of the present invention are as follows: The present invention adopts a method for simulating the river water and sediment transport process based on multi-GPU parallel computing with a triangular grid, which can be applied to the high-precision simulation of the river water and sediment transport process with complex terrain, can effectively accelerate the simulation of rivers with more refined terrain, and has strong advantages in terms of calculation accuracy and acceleration performance; it can also analyze the change in river water level after the change in riverbed topography caused by the water and sediment transport process, providing technical support for the flood control planning and construction of the region. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flowchart of the numerical simulation method for the river water and sediment transport process based on the multi-GPU parallel framework of the present invention;

[0036] Figure 2 is a schematic diagram of the multi-GPU parallel computing process based on the division of the computational region with a triangular grid in the numerical simulation method for the river water and sediment transport process based on the multi-GPU parallel framework of the present invention;

[0037] Figure 3 is a schematic diagram of the river channel region division and the upstream and downstream boundary data communication process of adjacent river reaches based on multi-GPU parallel computing in the numerical simulation method for the river water and sediment transport process based on the multi-GPU parallel framework of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] The present invention provides a numerical simulation method for the river water and sediment transport process based on a multi-GPU parallel framework, as Figure 1 shown, and is specifically implemented according to the following steps:

[0040] Step 1, first, manually divide the entire river region into two river channels with the same length, and then subdivide the two computational regions where the two river channels are located into triangular unstructured grids. The specific method is as follows:

[0041] Step 1.1, divide the entire river region into two river channels with the same length. The distances of the river channels are the same, and then subdivide the two computational regions where the two river channels are located into computational region one and computational region two. The two adjacent columns of computational grid regions in the middle of the two computational regions are overlapping regions, as Figure 2 shown;

[0042] Step 1.2, subdivide the two computational regions where the two river channels of the river are located into multiple triangular unstructured grids;

[0043] Step 2: Make the triangular meshes on the left and right sides of the middle boundary of the overlapping area share a common edge one by one. Specifically, divide the grid cells on both sides of the overlapping area into a neat column of grid cells, and make the triangular meshes on the left and right sides of the middle boundary of the overlapping area share a common edge one by one to ensure that the interface fluxes on the common edges in the overlapping area can be calculated correctly.

[0044] Step 3: Initialize the variables in the coupled process of water flow and sediment, set the total calculation duration and boundary conditions, and set the time step value for each iterative calculation as Δt.

[0045] The variables include: rainfall, inflow discharge, topographic elevation, Manning coefficient, bed sediment concentration, and suspended sediment concentration. The boundary conditions include: time step value Δt, total duration value, upstream inlet boundary of the river, downstream outlet boundary, and boundaries on both sides of the river.

[0046] Step 4: Create two GPUs, and store calculation area one and calculation area two into the corresponding GPUs respectively. Specifically, create two GPUs, GPU0 and GPU1, read the topographic file with the drawn meshes. The total number of meshes is m. Store calculation area one in GPU0 with the total number of meshes m1, and store calculation area two in GPU1 with the total number of meshes m - m1.

[0047] Step 5: Create 2 CUDA streams, and call and allocate the calculation tasks for the 2 GPUs through the 2 CUDA streams respectively. Allocate memory for the 2 GPUs, and then copy and store the initialized variables of each river calculation area from the CPU to the corresponding GPUs. The data includes: rainfall, inflow discharge, topographic elevation, Manning coefficient, bed sediment concentration, and suspended sediment concentration.

[0048] Step 6: As Figure 3 shown, perform parallel calculation of the river water and sediment transport process in multiple GPUs. Calculate according to the set time step Δt to complete the first time step calculation. When performing the first time step Δt calculation, perform data exchange and communication between GPUs for the calculation results of the meshes on the left and right sides of the middle boundary of the overlapping area. Synchronously calculate the source terms and fluxes of each triangular mesh interface on the two river reaches on the two GPUs to obtain the water depth values, and then synchronously calculate the water depth on the two GPUs to obtain the water level values of each river reach and the downstream outlet section of the river.

[0049] Use the P2P data communication method to achieve data communication between adjacent two GPUs, as Figure 3As shown in the figure, the numerical value of the downstream outlet boundary of the computational river reach 0 on GPU0 is copied into GPU1 and used as the upstream inlet condition of river reach 1 for calculating the water level of river reach 1. Meanwhile, the numerical value of the upstream inlet of river reach 1 in the current GPU1 is copied into GPU0 and used as the downstream outlet information of river reach 0 to participate in the calculation. The calculation of the coupled transport process of water flow and sediment in the above two river reaches is realized in parallel on 2 GPUs, which improves the calculation efficiency.

[0050] The calculation method of the coupled transport process of water flow and sediment is as follows: A coupled transport equation for water flow and sediment is established, and the finite volume method is used to discretize and solve the transport equation.

[0051] The transport equation is specifically expressed as:

[0052]

[0053]

[0054] In the formula:

[0055] q——is the variable vector of q x , q y , h, where q x , q y are the unit-width discharges in the x and y directions respectively, and h is the water depth;

[0056] f, g——are the flux vectors in the x and y directions;

[0057] S——is the source term vector;

[0058] η——is the water level value;

[0059] z b ——is the topographic elevation of the river channel;

[0060] u, v——are the flow velocities in the x and y directions, where q x = uh and q y = vh;

[0061] g——is the gravity coefficient, g = 9.81 m / s 2 ;

[0062] C1, C2——are the concentrations of suspended sediment and bed sediment;

[0063] β——is the velocity difference between the sediment deposit and the water flow;

[0064] S x1 、S x2 ——are the bottom slope source terms of suspended sediment and bed sediment in the x direction;

[0065] S y1, S y2 —— is the frictional resistance source term of suspended sediment and bed sediment in the y - direction;

[0066] E1, D1—— are the deposition rate and sediment - carrying rate of suspended sediment;

[0067] E2, D2—— are the deposition rate and sediment - carrying rate of bed sediment.

[0068] The calculation method of the water level value η is as follows: The water level value of each cross - section at the outlet of each river reach and the water level value of the downstream outlet cross - section of the whole river reach can be obtained by adding the water depth value and elevation of each grid cell on the cross - section. The water level value η is specifically expressed as:

[0069] η = h + z b (3)

[0070] In the formula:

[0071] η—— is the water level value of each grid cell at each cross - section;

[0072] h—— is the water depth value of each grid cell at each cross - section;

[0073] z b —— is the topographic elevation value of each grid cell at each cross - section;

[0074] In step 7, according to the set total calculation duration, update the time step in each GPU, repeat step 6 for loop calculation. After all the loop calculation tasks of all time steps are completed, store the result data calculated by each GPU into the CPU for merging and then output.

[0075] Specifically: Update and calculate the time step in 2 GPUs, perform data exchange and communication between GPUs for the grid calculation results on both sides of the middle boundary of the overlapping area, synchronously perform parallel calculation of the coupled transport process of water flow and sediment on 2 GPUs. After all the loop calculation tasks of all time steps are completed, copy and store the calculation result data of the two GPUs into the CPU. The result data includes the coordinate data, topographic elevation data, and water level data of each grid in the calculation area. Then merge these data. The total number of grids in the merged calculation area is m. Finally, extract the water level value at each moment, and use MATLAB data analysis to extract the water level value at each moment, and calculate the water level value of the downstream outlet cross - section of each river reach for multiple durations.

[0076] The present invention proposes a method for dividing the computational domain of a multi-GPU based on unstructured grids, and creates a numerical model of the river water and sediment transport process for a multi-GPU parallel computing framework, which has more advantages in accurately simulating multiple processes and efficient simulation compared with the traditional simulation of sediment transport movement. It can quickly and reliably predict the water level changes at different cross-sections, providing support for quickly formulating emergency plans for flood disasters.

Claims

1. A numerical simulation method for the river water and sediment transport process based on a multi-GPU parallel framework, characterized in that, The implementation is specifically carried out according to the following steps: Step 1: Divide the entire river into two river channels of the same length, and then subdivide the two calculation regions where the two river channels are located into calculation region 1 and calculation region 2. The two adjacent columns of calculation grid regions in the middle of the two calculation regions are the overlapping regions. Both calculation region 1 and calculation region 2 are composed of multiple triangular unstructured grids; Step 2: Make the triangular grids on the left and right sides of the middle boundary of the overlapping region share one side in one-to-one correspondence; Step 3: Initialize the variables in the coupled process of water flow and sediment, set the total calculation duration and boundary conditions, and set the time step value of each iterative calculation to Δt; Step 4: Create two GPUs, and store calculation region 1 and calculation region 2 into the corresponding GPUs respectively; Step 5: Create two CUDA streams, allocate memory for each GPU, and copy the initialized variables from the CPU to the corresponding GPU memory respectively; Step 6: Perform parallel calculation of the river water and sediment transport process in multiple GPUs. Calculate according to the set time step Δt, complete the first time step calculation, and obtain the water level values at the downstream outlet sections of each river reach; Step 7: According to the set total calculation duration, update the time step in each GPU, repeat Step 6 for loop calculation. After all the loop calculation tasks of all time steps are completed, store the result data calculated by each GPU into the CPU for merging and then output.

2. The numerical simulation method for river water and sediment transport process based on a multi-GPU parallel framework according to claim 1, characterized in that The specific implementation of Step 2 is as follows: Divide the grid cells on both sides of the overlapping region into a neat column of grid cells. The triangular grids on the left and right sides of the middle boundary of the overlapping region share one side in one-to-one correspondence to ensure that the interface fluxes on the shared side of the overlapping region can be calculated correctly.

3. The numerical simulation method for river water and sediment transport process based on a multi-GPU parallel framework according to claim 1, characterized in that In Step 3, the variables include: rainfall, inflow discharge, terrain elevation, Manning coefficient, bed sediment concentration, suspended sediment concentration; the boundary conditions include: time step value Δt, total duration value, upstream inlet boundary of the river channel, downstream outlet boundary, and boundaries on both sides of the river channel.

4. The numerical simulation method for river water and sediment transport process based on a multi-GPU parallel framework according to claim 1, characterized in that Step 6 is specifically: Perform parallel calculation in multiple GPUs, and conduct data exchange and communication between GPUs for the grid calculation results on the left and right sides of the middle boundary of the overlapping region. Synchronously calculate the source terms and fluxes of each triangular grid interface on the two GPUs to obtain the water depth values, and then synchronously calculate the water depth on the two GPUs to obtain the water level values at the downstream outlet sections of each river reach.

5. The numerical simulation method for river water and sediment transport process based on the multi-GPU parallel framework according to claim 4, characterized in that The parallel calculation method is: Establish a coupled transport equation for the water flow and sediment processes, and use the finite volume method to discretely solve the transport equation; The transport equation is specifically expressed as: In the formula: q——is q x 、q y 、variable vectors of h, where q x 、q y are the unit-width discharges in the x and y directions respectively, and h is the water depth; f, g——are the flux vectors in the x and y directions; S——is the source term vector; η——is the water level value; z b —— is the topographic elevation of the river channel; u, v -- flow velocities in the x and y directions, where q x = uh and q y = vh; g——is the gravitational coefficient, g = 9.81 m / s 2 ; C1, C2——are the concentrations of suspended sediment and bed sediment; β——is the velocity difference between the sediment deposit and the water flow; S x1 、S x2 —— Source term of bottom slope for suspended sediment and bed sediment in the x direction; S y1 、S y2 —— Source terms of frictional resistance for suspended sediment and bed sediment in the y direction; E1, D1——are the deposition rate and sediment-carrying rate of suspended sediment; E2, D2——are the deposition rate and sediment-carrying rate of bed sediment.

6. The numerical simulation method for the river water and sediment transport process based on the multi-GPU parallel framework according to claim 1, wherein, In step 7, the time step is updated and calculated on two GPUs, and data exchange and communication between GPUs are performed for the grid calculation results on both sides of the middle boundary of the overlapping area. The coupled transport process of water flow and sediment is synchronously and parallelly calculated on the two GPUs until the loop calculation tasks for all time steps are completed. Then, the calculation result data of the two GPUs are copied and stored in the CPU. The result data includes the coordinate data, terrain elevation data, and water level data of each grid in the calculation area. These data are then merged, and finally the water level values at each moment are extracted to calculate the water level values at the downstream outlet section of each river reach for multiple durations.

Images