Deposition numerical simulation acceleration method, device and equipment

Abstract

The invention discloses a deposition numerical simulation acceleration method, device and equipment, and relates to the technical field of oil-gas exploration and development and artificial intelligence crossing. The method comprises the steps of determining a control equation of river deposition; constructing a physical information neural network; adding a low-rank adapter in a hidden layer of the physical information neural network; a training data set is obtained, wherein the training data set comprises water and sediment movement elements of different monitoring points at different time; embedding the control equation into the loss function in a residual form; according to the training data set, training the physical information neural network added with a low-rank adapter by taking the minimum loss calculated by the loss function as a target, and adjusting hidden layer parameters to obtain a trained physical information neural network; and inputting the space coordinates and the prediction time of the to-be-simulated river into the trained physical information neural network to obtain water and sediment motion elements of the to-be-simulated river in different space coordinates and different prediction time in the deposition process. According to the invention, the deposition process can be efficiently and accurately simulated.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of oil and gas exploration and development and artificial intelligence, and in particular to a method, apparatus and equipment for accelerating sediment numerical simulation. Background Technology

[0002] In numerical simulations of sedimentary processes in oil and gas exploration, traditional methods such as computational fluid dynamics can accurately simulate sedimentation processes, but their high computational complexity and long processing time make them unsuitable for real-time and large-scale computations. In recent years, with the development of deep learning technology, neural networks have shown considerable application potential in numerical simulation. Deep learning-based sedimentary numerical simulation methods, such as the Diffusion model, simulate various sedimentary environments by generating sediment parameters (such as grain size distribution and initial porosity boundaries). However, conventional neural networks often require large amounts of labeled data for training when dealing with complex physical processes and are prone to overfitting, making it difficult to accurately follow the physical laws governing sedimentation.

[0003] Therefore, how to efficiently and accurately simulate the deposition process is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, and device for accelerating numerical simulation of sedimentation, which can efficiently and accurately simulate the sedimentation process.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] In a first aspect, this application provides a method for accelerating numerical simulation of deposition, including:

[0007] Determine the governing equations for river deposition;

[0008] Construct a physical information neural network; the physical information neural network includes hidden layers;

[0009] Add a low-rank adapter to the hidden layer;

[0010] Obtain a training dataset; the training dataset includes water and sediment movement elements at different monitoring points at different times;

[0011] The control equations are embedded in the loss function of the physical information neural network in the form of residuals;

[0012] Based on the training dataset, with the goal of minimizing the loss calculated by the loss function, a physical information neural network with a low-rank adapter is trained, and the hidden layer parameters are adjusted to obtain the trained physical information neural network.

[0013] The spatial coordinates and prediction time of the river to be simulated are input into a trained physical information neural network to obtain water and sediment movement elements at different spatial coordinates and prediction times during the sedimentation process of the river to be simulated.

[0014] Secondly, this application provides a deposition numerical simulation acceleration device, including: a control equation determination module, a neural network construction module, a low-rank adapter addition module, a training data acquisition module, a control equation embedding module, a training module, and a simulation module.

[0015] The governing equation determination module is used to determine the governing equations for river deposition.

[0016] A neural network construction module is used to construct a physical information neural network; the physical information neural network includes hidden layers.

[0017] A low-rank adapter addition module is used to add a low-rank adapter in the hidden layer;

[0018] The training data acquisition module is used to acquire the training dataset; the training dataset includes water and sediment movement elements at different monitoring points at different times;

[0019] The control equation embedding module is used to embed the control equations into the loss function of the physical information neural network in the form of residuals;

[0020] The training module is used to train a physical information neural network with a low-rank adapter based on the training dataset, with the goal of minimizing the loss calculated by the loss function, and to adjust the hidden layer parameters to obtain the trained physical information neural network.

[0021] The simulation module is used to input the spatial coordinates and prediction time of the river to be simulated into a trained physical information neural network to obtain the water and sediment movement elements of the river at different spatial coordinates and different prediction times during the deposition process.

[0022] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the above-described method for accelerating deposition numerical simulation.

[0023] According to the specific embodiments provided in this application, this application has the following technical effects:

[0024] This application provides a method, apparatus, and device for accelerating sedimentation numerical simulation. By combining physical information neural networks and low-rank adapter technology, the network training parameters and computational load are significantly reduced while ensuring simulation accuracy, thus significantly improving the computational efficiency of sedimentation numerical simulation. By embedding the governing equations of the physical laws of reaction into the physical information neural network, the physical information neural network can better capture the physical laws of the sedimentation process, reduce the dependence on a large amount of labeled data, improve the generalization ability of the model, and more accurately simulate the sedimentation process under different working conditions. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic flowchart illustrating a method for accelerating numerical simulation of deposition provided in this application embodiment;

[0027] Figure 2 A schematic diagram illustrating the principle of a deposition numerical simulation acceleration method provided in this application embodiment;

[0028] Figure 3 A simplified flowchart illustrating a method for accelerating numerical simulation of deposition provided in this application embodiment;

[0029] Figure 4 A schematic diagram illustrating the construction of a network based on a physical information neural network and a low-rank adapter, provided for embodiments of this application;

[0030] Figure 5 A schematic diagram of the functional modules of a deposition numerical simulation acceleration device provided in an embodiment of this application;

[0031] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Physics-Informed Neural Networks (PINNs) embed physical laws as constraints into neural networks, enabling efficient simulation of physical processes with relatively little data and addressing some of the limitations of traditional neural networks. However, PINNs suffer from high-dimensional and complex deposition problems due to the large number of network parameters and low training efficiency.

[0035] Low-Rank Adaptation (LoRA) is a lightweight neural network parameter optimization technique that adds low-rank trainable modules to the original neural network, significantly reducing the number of training parameters and computational cost while maintaining the network's expressive power, thus accelerating the network training process.

[0036] Based on the above, in an exemplary embodiment, such as Figure 1 and Figure 2 As shown, a method for accelerating numerical simulation of deposition is provided, including steps 101 to 107. Wherein:

[0037] Step 101: Determine the governing equations for river deposition;

[0038] Step 102: Construct a physical information neural network; the physical information neural network includes hidden layers;

[0039] Step 103: Add a low-rank adapter to the hidden layer;

[0040] Step 104: Obtain the training dataset; the training dataset includes water and sediment movement elements at different monitoring points at different times;

[0041] Step 105: Embed the control equations into the loss function of the physical information neural network in the form of residuals;

[0042] Step 106: Based on the training dataset, with the goal of minimizing the loss calculated by the loss function, train the physical information neural network with a low-rank adapter, adjust the hidden layer parameters, and obtain the trained physical information neural network.

[0043] Step 107: Input the spatial coordinates and prediction time of the river to be simulated into the trained physical information neural network to obtain the water and sediment movement elements of the river at different spatial coordinates and different prediction times during the deposition process.

[0044] By implementing steps 101 to 107 above, a technical solution combining PINN and LoRA for accelerating sedimentation numerical simulation is achieved. By combining physical laws with deep learning and using LoRA technology to optimize network training, the computational efficiency of sedimentation numerical simulation is significantly improved and the computational cost is reduced while ensuring simulation accuracy.

[0045] In another exemplary embodiment of this application, a sedimentary physical model and governing equations are determined: Based on the specific sedimentary problem, a corresponding physical model is established, and governing equations describing the sedimentary process are determined. The physical model of river sedimentation is a complex system describing the coupled effects of water flow, sediment transport (suspended and bedload), and the resulting riverbed deformation (erosion and deposition). The core governing equations include the water flow continuity equation, momentum equation, and sediment transport equation. Therefore, the governing equations of river sedimentation include: the continuity equation, the momentum equation, and the sediment transport equation.

[0046] The continuity equation is:

[0047]

[0048] In the formula, ρ is the fluid density, and t is time. For gradient operators, This is the fluid velocity vector.

[0049] The momentum equation is:

[0050]

[0051] In the formula, p is the pressure, and μ is the dynamic viscosity. This is the acceleration due to gravity.

[0052] The sediment transport equation is:

[0053]

[0054] In the formula, C is the sediment concentration, D is the sediment diffusion coefficient, and S is the sediment source-sink term.

[0055] In another exemplary embodiment of this application, a physical information neural network structure is constructed: a neural network comprising an input layer, hidden layers, and an output layer is designed. The input layer contains variables related to the deposition process, such as spatial coordinates (x, y, z) and time t; the output layer contains physical quantities to be solved (water and sediment movement elements), such as fluid velocity, pressure, and sediment concentration. The hidden layers employ a multi-layer fully connected structure, with activation functions, such as ReLU (Rectified Linear Unit) functions, added between the hidden layers to enhance the network's nonlinear expressive power, enabling it to approximate complex functions. Simultaneously, the determined governing equations are embedded in the neural network's loss function in residual form to construct physical constraints.

[0056] In physical information neural networks, the connections in the hidden layers follow the hierarchical structure of a standard fully connected neural network (Feedforward Neural Network). Specifically, the hidden layers are located between the input layer and the output layer. In physical information neural networks, the hidden layers are the core structure of the neural network, responsible for transforming input data (such as spatiotemporal coordinates) into outputs (such as flow rate, concentration, etc.) that conform to physical laws.

[0057] In another exemplary embodiment of this application, a low-rank adapter module is added: a LoRA module is added to some or all of the hidden layers of the PINN. Specifically, for each fully connected layer to which a LoRA module is to be added, the original weight matrix W is decomposed into a product of low-rank matrices, i.e., W = W A W B W A and W B For a low-rank matrix, by training only W A and W B This allows for updating network parameters, thereby reducing the number of training parameters and computational load.

[0058] In another exemplary embodiment of this application, training data is prepared by acquiring a small amount of initial and boundary condition data of the depositional process, as well as physical quantity measurements at some key locations, through traditional numerical simulation methods (such as the finite element method or the finite difference method) or experimental measurements. This data is used to supervise the training of PINN to improve the network's ability to fit the actual depositional process. Specifically, water and sediment movement elements at different monitoring points at different times during the depositional process are obtained through numerical simulation methods or experimental measurements; these water and sediment movement elements include river velocity, pressure, and sediment concentration.

[0059] In another exemplary embodiment of this application, a loss function is defined: the loss function consists of two parts: data-driven loss and physical constraint loss. The loss function of the physical information neural network is:

[0060] L=αLdata +βL physics ;

[0061] In the formula, L represents the total loss of the physical information neural network. data For data-driven loss, L physics Let α be the weighting coefficient for the data-driven loss and β be the weighting coefficient for the physical constraint loss. α and β are used to balance the relative importance of the data-driven loss and the physical constraint loss.

[0062] Data-driven loss measures the error between the network output and the actual measured data. It is calculated using the mean squared error (MSE) formula:

[0063]

[0064] In the formula, N is the number of data samples in the training dataset. Let i be the network prediction value for the i-th data sample. This represents the actual measured value of the i-th data sample.

[0065] Physical constraint loss is used to measure the magnitude of the residuals of the governing equations. It is also calculated using the mean squared error formula:

[0066]

[0067] In the formula, M is the sample size for calculating the residuals of the governing equations, and R... j Let be the residual of the j-th governing equation.

[0068] Among them, R j It is a solution predicted by a physical information neural network. The degree to which the equation is not satisfied after being substituted into the partial differential equation (PDE).

[0069] The partial differential equation is:

[0070]

[0071] Residual calculation:

[0072] 1. Physical information neural network outputs predicted solution

[0073] 2. Calculate partial derivatives (using automatic differentiation):

[0074]

[0075] 3. Residual:

[0076]

[0077] 3. Optimization objective: Minimize R 2 .

[0078] The residual R of the j-th governing equation j The expression is:

[0079]

[0080] In the formula, F is the differential operator. The solution is predicted by a physical information neural network, where t is time. Here, x is the gradient operator, and x is the x-axis coordinate in spatial coordinates.

[0081] R j It is not derived from R; R is R j In the specific implementation of a particular governing equation. R j =F(...) is a general template describing how to construct residuals for any governing equation. Different governing equations have different forms of the differential operator F.

[0082] F in the equation is not arbitrary; it is directly the mathematical expression of the partial differential governing equation to be solved. The predicted solution obtained from automatic differential calculation... derivative This is substituted into the equation expression F. The result R obtained after substitution is the deviation (residual) of the predicted solution satisfying the governing equation at the current point. The norm of R over the entire domain (e.g., R0) is minimized using an optimization algorithm (such as gradient descent). 2 (Sum / integral), the neural network parameters are adjusted to improve the predicted solution. It not only fits the data, but more importantly, it increasingly and accurately satisfies the physical laws described by the governing equation F=0.

[0083] Therefore, the residual R principle is the core mechanism by which PINN seamlessly integrates physical constraints (governing equations) into the neural network training process, and the formula F(...) is the embodiment of the governing equations.

[0084] In another exemplary embodiment of this application, the network is trained by using a stochastic gradient descent algorithm or improved algorithms such as Adam, Adagrad, etc., to train the PINN network with added LoRA modules. During training, the network parameters and the parameters of the LoRA modules are continuously adjusted to minimize the total loss function.

[0085] In another exemplary embodiment of this application, when the spatial coordinates and prediction time of the river to be simulated are input into the trained physical information neural network, the initial conditions and boundary conditions of the river to be simulated are also input into the trained physical information neural network; the initial conditions are the initial water and sediment movement elements of the river to be simulated; the boundary conditions are the physical laws at the boundary of the river to be simulated.

[0086] When performing sedimentation numerical simulation, the initial and boundary conditions of the sedimentation problem to be simulated are input into the trained network. The predicted values ​​of each physical quantity in the sedimentation process at different spatial and temporal points are calculated through the forward propagation of the network, thus completing the sedimentation numerical simulation.

[0087] Figure 3 This is a simplified flowchart of the method described in this application. The simplified process includes: determining the deposition physics model and the control equations, embedding the neural network loss function; establishing a physically interpretable physical information neural network structure; introducing LoRA technology, and, based on freezing the pre-trained PINN backbone network parameters, using a trainable low-rank matrix (W... A W B It enables rapid adaptation to different geological scenarios; through a collaborative optimization mechanism of low-rank matrix decomposition and physical equation constraints, it achieves efficient and physically reliable sedimentation simulation.

[0088] Figure 4 This is a schematic diagram illustrating the network constructed based on a physical information neural network and a low-rank adapter in this application. Figure 4 In this diagram, spatial coordinates (x, y, z) and time t are the inputs to the physical information neural network, σ represents the ReLU activation function, and W represents the original weight matrix. A W B This represents a low-rank matrix. This is the solution predicted by the physical information neural network, that is, the water and sediment movement elements predicted by the physical information neural network. u represents the velocity component in the x-direction (horizontal mainstream direction), and v represents the velocity component in the y-direction (horizontal or vertical direction). It represents the partial derivative of a physical quantity n (which can correspond to flow velocity u / v, pressure p, sediment concentration C, etc.) with respect to time t, describing the rate of change of the physical quantity with time, such as how the flow velocity changes with time and the evolution trend of sediment concentration over time. The partial derivative of a physical quantity *n* with respect to the spatial coordinate *x* (which can be understood as the longitudinal direction of a river, such as from upstream to downstream) reflects the gradient of the physical quantity in the *x* direction. For example, if the longitudinal flow velocity of a river varies, this partial derivative can reflect the trend of the flow velocity along the length of the river, indicating whether it is accelerating or decelerating. This is the mathematical expression of the boundary conditions. 'I' represents the entry constraint, which is expressed in PINN using the loss condition. inConstraining the network output at the inlet physical quantities ensures the simulation starts with the "correct starting conditions," which is particularly crucial for flow problems like river deposition (where upstream flow determines downstream deposition). ε is a threshold used to determine if the minimum value of the loss function is less than the threshold.

[0089] This application constructs a dual-constraint fusion architecture by embedding sedimentary dynamics control equations, such as the continuity equation and sediment transport equation, into a neural network loss function to establish a physically interpretable PINN framework that follows the laws of mass and momentum conservation. It also introduces LoRA technology, freezing the pre-trained PINN backbone network parameters and using a trainable low-rank matrix (W... A W B This method enables rapid adaptation to different geological scenarios. It achieves efficient and physically reliable sedimentary simulation through collaborative optimization of low-rank matrix factorization and physical equation constraints.

[0090] The method of this application will be explained in detail below using a numerical simulation of a river deposition process as an example.

[0091] 1. Determine the sedimentary physics model and governing equations: For river sedimentation, the continuity equation, momentum equation and sediment transport equation mentioned above are used as governing equations to describe the flow of fluids and the transport of sediment in the river.

[0092] 2. Constructing a physical information neural network structure: The input layer is designed with spatial coordinates (x, y, z) and time t, with a total of 4 input nodes; the output layer has fluid velocities u and v, pressure p, and sediment concentration C, with a total of 4 output nodes. The hidden layers are set as 5 fully connected layers, each containing 100 neurons, and the ReLU activation function is used between the hidden layers. The governing equations are embedded into the loss function in residual form.

[0093] 3. Add a low-rank adapter module: Add a LoRA module to all 5 hidden layers of PINN to perform low-rank decomposition on the weight matrix of each hidden layer. The rank of the low-rank matrix is ​​set to 10. The network parameters are updated by training the low-rank matrix.

[0094] The LoRA module is applied only to the weight matrix of the five hidden layers. Therefore, only the network parameters of the hidden layers are updated: the weights of the hidden layers are updated using the matrix obtained from the low-rank decomposition (rank 10); other parameters are frozen: the connection parameters from the input layer to the first hidden layer and from the last hidden layer to the output layer remain unchanged. This method can adapt to specific tasks by training a small number of parameters (low-rank matrices) while keeping the main structure of the model unchanged, significantly reducing the computational cost.

[0095] 4. Prepare training data: A small-scale simulation of the river deposition process was conducted using the finite element method to obtain initial data on velocity, pressure, and sediment concentration at the river inlet, as well as boundary condition data at the river boundary. Simultaneously, monitoring points were set up at five key locations along the river to obtain measurements of fluid velocity, pressure, and sediment concentration at these points at different times, which were then used as training data.

[0096] The process of acquiring boundary condition data for river boundaries is as follows: defining the boundary type → setting constraint types based on physical (fluid dynamics) laws → calculating specific values ​​through finite element simulation → calibrating by combining measured data. These data ultimately serve as constraints for PINN training, ensuring that the neural network follows the physical laws at the river boundary during simulation (e.g., fluid does not penetrate the riverbank, and the inflow at the inlet conforms to actual dynamic characteristics).

[0097] Five key locations: the main channel, near-shore / riverbank areas, abrupt changes in riverbed topography, not far downstream of the inlet, and not far upstream of the outlet. The core function of these five key points is to cover the areas in the river with the greatest differences in "dynamic characteristics" and "depositional behavior." By monitoring data (velocity, pressure, sediment concentration) at different times, these points can capture both the strong dynamic sediment transport processes in the main channel and the weak dynamic deposition processes in the boundary areas, while also connecting the boundary effects between upstream and downstream. This provides PINN with "precise data constraints at specific points," assisting the network in learning the overall flow and deposition patterns of the river.

[0098] 5. Define the loss function: Both data-driven loss and physical constraint loss are calculated using mean squared error. Based on the actual situation, set the weighting coefficients α = 0.6 and β = 0.4, and calculate the total loss function.

[0099] 6. Network Training: The Adam optimization algorithm was used to train the network, with a learning rate of 0.001 and 10,000 training iterations. During training, the loss function was monitored in real time, and training was stopped when the loss function converged.

[0100] 7. Perform sedimentation numerical simulation: Input the new initial and boundary conditions of the river into the trained network, and calculate the distribution of fluid velocity, pressure and sediment concentration at different locations of the river in the future through forward propagation of the network, thus completing the river sedimentation numerical simulation.

[0101] The "new initial and boundary conditions for rivers" emphasize that these conditions can cover a wider range of scenarios compared to the "small-scale simulation data" used in the training phase. For example, the training data may come from a simulation of a 100-meter section of a river, while the new conditions can be used to simulate a 1000-meter section of the same river; or the training data may be based on parameters during the dry season, while the new conditions can be used to simulate flood season (with higher flow velocity and sediment concentration). By inputting these "new initial and boundary conditions for rivers" into the trained network, the fluid movement and sediment deposition processes in the corresponding scenarios can be quickly predicted, demonstrating the model's generalization ability.

[0102] Through the above implementation methods, this application can quickly and accurately accelerate sedimentation numerical simulation, providing an efficient technical means for research and application in related fields.

[0103] The beneficial effects of this application are as follows:

[0104] 1. Improve simulation efficiency: By combining PINN and LoRA techniques, the network training parameters and computational load are significantly reduced while ensuring simulation accuracy, which greatly improves the computational efficiency of sedimentation numerical simulation and saves a lot of computation time and resources compared with traditional methods.

[0105] 2. Enhanced generalization ability: PINN embeds physical laws into the network, enabling the network to better capture the physical laws of the deposition process, reducing the dependence on a large amount of labeled data, improving the model's generalization ability, and enabling it to more accurately simulate the deposition process under different working conditions.

[0106] 3. Reduced computational costs: The application of LoRA technology reduces the computational resources required for network training, lowers hardware costs and energy consumption, and is suitable for large-scale, long-term depositional numerical simulation tasks.

[0107] Based on the same inventive concept, this application also provides a sedimentation numerical simulation acceleration apparatus for implementing the aforementioned sedimentation numerical simulation acceleration method. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more sedimentation numerical simulation acceleration apparatus embodiments provided below can be found in the limitations of the sedimentation numerical simulation acceleration method described above, and will not be repeated here.

[0108] In one exemplary embodiment, such as Figure 5 As shown, a deposition numerical simulation acceleration device is provided, comprising: a control equation determination module, a neural network construction module, a low-rank adapter addition module, a training data acquisition module, a control equation embedding module, a training module, and a simulation module.

[0109] The system comprises the following modules: a governing equation determination module for determining the governing equations for river deposition; a neural network construction module for constructing a physical information neural network (PEN), which includes hidden layers; a low-rank adapter addition module for adding low-rank adapters to the hidden layers; a training data acquisition module for acquiring a training dataset, which includes water and sediment movement elements at different monitoring points at different times; a governing equation embedding module for embedding the governing equations as residuals into the loss function of the PSN; a training module for training the PSN with added low-rank adapters based on the training dataset, aiming to minimize the loss calculated by the loss function, adjusting the hidden layer parameters, and obtaining a trained PSN; and a simulation module for inputting the spatial coordinates and prediction time of the simulated river into the trained PSN to obtain water and sediment movement elements at different spatial coordinates and prediction times during the deposition process.

[0110] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores water and sediment movement elements at different spatial coordinates and prediction times during the sedimentation process of the simulated river. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for accelerating numerical sedimentation simulation.

[0111] Those skilled in the art will understand that Figure 6 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0112] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for accelerating numerical simulation of sedimentation, characterized in that, include: Determine the governing equations for river deposition; Constructing a physical information neural network; The physical information neural network includes a hidden layer; Add a low-rank adapter to the hidden layer; Obtain a training dataset; the training dataset includes water and sediment movement elements at different monitoring points at different times; The control equations are embedded in the loss function of the physical information neural network in the form of residuals; Based on the training dataset, with the goal of minimizing the loss calculated by the loss function, a physical information neural network with a low-rank adapter is trained, and the hidden layer parameters are adjusted to obtain a well-trained physical information neural network. The spatial coordinates and prediction time of the river to be simulated are input into a trained physical information neural network to obtain water and sediment movement elements of the river at different spatial coordinates and prediction times during the deposition process.

2. The method for accelerating numerical simulation of sedimentation according to claim 1, characterized in that, The governing equations for river deposition include: the continuity equation, the momentum equation, and the sediment transport equation. The continuity equation is: In the formula, ρ is the fluid density, and t is time. For gradient operators, For fluid velocity vector; The momentum equation is: In the formula, p is the pressure, and μ is the dynamic viscosity. It is the acceleration due to gravity; The sediment transport equation is as follows: In the formula, C is the sediment concentration, D is the sediment diffusion coefficient, and S is the sediment source-sink term.

3. The accelerated numerical simulation method for sedimentation according to claim 1, characterized in that, The physical information neural network includes: an input layer, multiple hidden layers, and an output layer connected in sequence; The hidden layers employ a multi-layer fully connected layer structure; activation functions are added between the hidden layers.

4. The accelerated numerical simulation method for sedimentation according to claim 1, characterized in that, Adding a low-rank adapter to the hidden layer specifically includes: Add a low-rank adapter to some or all of the hidden layers; the low-rank adapter is used to decompose the original weight matrix of the hidden layer into a product of low-rank matrices.

5. The method for accelerating numerical simulation of sedimentation according to claim 1, characterized in that, Obtaining the training dataset specifically includes: The water and sediment movement elements at different monitoring points at different times during the sedimentation process are obtained through numerical simulation or experimental measurement; the water and sediment movement elements include river velocity, pressure and sediment concentration.

6. The method for accelerating numerical simulation of sedimentation according to claim 1, characterized in that, The loss function of the physical information neural network is: L=αL data +βL physics ; In the formula, L is the total loss of the physical information neural network, L data For data-driven loss, L physics For physical constraint loss, α is the weighting coefficient of data-driven loss, and β is the weighting coefficient of physical constraint loss. The expression for the data-driven loss is: In the formula, N is the number of data samples in the training dataset. Let i be the network prediction value for the i-th data sample. This represents the actual measured value of the i-th data sample. The expression for the physical constraint loss is: In the formula, M is the sample size for calculating the residuals of the governing equations, and R... j Let be the residual of the j-th governing equation.

7. The method for accelerating numerical simulation of sedimentation according to claim 6, characterized in that, The residual R of the j-th governing equation j The expression is: In the formula, F is the differential operator. The solution is predicted by a physical information neural network, where t is time. Here, x is the gradient operator, and x is the x-axis coordinate in spatial coordinates.

8. The method for accelerating numerical simulation of sedimentation according to claim 1, characterized in that, When the spatial coordinates and prediction time of the river to be simulated are input into the trained physical information neural network, the initial conditions and boundary conditions of the river to be simulated are also input into the trained physical information neural network; the initial conditions are the initial water and sediment movement elements of the river to be simulated; the boundary conditions are the physical laws at the boundary of the river to be simulated.

9. A deposition numerical simulation acceleration device, characterized in that, include: The governing equation determination module is used to determine the governing equations for river deposition. The neural network building block is used to construct physical information neural networks; The physical information neural network includes a hidden layer; A low-rank adapter addition module is used to add a low-rank adapter in the hidden layer; The training data acquisition module is used to acquire the training dataset; the training dataset includes water and sediment movement elements at different monitoring points at different times; The control equation embedding module is used to embed the control equations into the loss function of the physical information neural network in the form of residuals; The training module is used to train a physical information neural network with a low-rank adapter based on the training dataset, with the goal of minimizing the loss calculated by the loss function, and to adjust the hidden layer parameters to obtain a trained physical information neural network. The simulation module is used to input the spatial coordinates and prediction time of the river to be simulated into a trained physical information neural network to obtain the water and sediment movement elements of the river at different spatial coordinates and different prediction times during the deposition process.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the deposition numerical simulation acceleration method according to any one of claims 1-8.

Images