CN-122021479-A - Gate water flow field prediction method, device and server based on brain-like heuristics

CN122021479ACN 122021479 ACN122021479 ACN 122021479ACN-122021479-A

Abstract

The invention provides a sluice water flow field prediction method, a sluice water flow field prediction device and a sluice water flow field prediction server based on brain-like heuristics, and relates to the technical field of flow field prediction, wherein the sluice water flow field prediction method comprises the steps of obtaining a calculation area of a sluice, boundary conditions, initial conditions and control equations of the calculation area, and collecting training data; the method comprises the steps of obtaining a feature vector set by processing each dimension in coordinate values of training data through mutually independent modularized sub-network sets, obtaining a predicted value of a flow field variable by tensor decomposition type feature fusion processing of the feature vector set, constructing a total loss function by utilizing the predicted value, boundary condition, initial condition and control equation of the flow field variable, and performing brain-like heuristic training processing on the modularized sub-network set based on the total loss function to obtain a brain-like heuristic separated physical information neural network so as to predict the flow field of a drain gate under different working conditions to obtain a target predicted result. The invention can remarkably improve the prediction efficiency and the prediction accuracy.

Inventors

CAI XINWEI
Lang Chaohao
WU JIANMING

Assignees

浙江远算科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. A floodgate water flow field prediction method based on brain-like heuristics, the method comprising: Acquiring a calculation region of a drain gate, and boundary conditions, initial conditions and control equations of the calculation region, and acquiring training data, wherein the training data comprises configuration points, boundary constraint points and initial condition constraint points in the calculation region; processing each dimension in the coordinate values of the training data through mutually independent modularized subnetwork sets to obtain a feature vector set, and performing tensor decomposition type feature fusion processing on the feature vector set to obtain a predicted value of a flow field variable; and constructing a total loss function by using the predicted value of the flow field variable, the boundary condition, the initial condition and the control equation, and performing brain-like heuristic training treatment on the modularized subnetwork set based on the total loss function to obtain a brain-like heuristic separated physical information neural network, so as to predict the flow field of the drain gate under different working conditions by using the brain-like heuristic separated physical information neural network to obtain a target predicted result.
2. The method for predicting a sluice water flow field based on brain-like heuristics according to claim 1, wherein the step of collecting training data comprises: acquiring sampling ranges of a longitudinal dimension, a transverse dimension, a depth direction dimension and a time dimension of a river channel, and respectively carrying out Latin hypercube sampling processing on each dimension based on the sampling ranges to obtain sampling coordinate values corresponding to each dimension; Combining the sampling coordinate values through Cartesian products to obtain a four-dimensional structured configuration point set, and determining the four-dimensional structured configuration point set as the configuration point in the computing domain; And carrying out sampling processing on a boundary surface of the calculation region to generate the boundary constraint point, and carrying out sampling processing on the initial moment of the calculation region to generate the initial condition constraint point so as to determine the training data according to the configuration point in the calculation region, the boundary constraint point and the initial condition constraint point.
3. The method for predicting the water gate flow field based on brain-like heuristics according to claim 1, wherein the step of processing each dimension in the coordinate values of the training data through mutually independent modularized subnetwork sets to obtain a feature vector set comprises the following steps: and respectively transmitting the river longitudinal coordinate, the river transverse coordinate, the water depth direction coordinate and the time coordinate in the coordinate values of the training data to the corresponding modularized subnetworks in the modularized subnetwork set so as to process the river longitudinal dimension information, the river transverse dimension information, the water depth direction dimension information and the time dimension information to obtain the feature vector set.
4. The method for predicting a water gate flow field based on brain-like heuristics according to claim 1, wherein the step of performing tensor decomposition type feature fusion processing on the feature vector set to obtain a predicted value of a flow field variable comprises: And carrying out element-by-element multiplication on the values of the same rank positions in the eigenvector set corresponding to each flow field variable, and carrying out summation processing on the products along the rank dimension to obtain predicted values corresponding to the flow field variables, wherein the flow field variables comprise a river channel longitudinal flow rate, a river channel transverse flow rate, a water depth direction flow rate and a pressure field.
5. The method for predicting a water gate flow field based on brain-like heuristics according to claim 1, wherein said constructing a total loss function using the predicted values of the flow field variables, the boundary conditions, the initial conditions and the control equation comprises: Determining the residual mean square value of the configuration point of the control equation in the calculation domain as partial differential equation residual loss, comparing the predicted value of the flow field variable with the boundary condition and the initial condition respectively, and calculating boundary condition loss and initial condition loss; applying regularization constraint to the connection weights in the modularized subnetwork set, generating regularization loss, and determining brain-like optimization loss based on the regularization loss; and carrying out weighted summation processing on the partial differential equation residual loss, the boundary condition loss, the initial condition loss and the brain-like optimization loss to obtain the total loss function.
6. The method of water gate flow field prediction based on brain-like heuristics of claim 5, wherein said step of determining brain-like optimization loss based on said regularization loss comprises: giving two-dimensional geometric coordinates to neurons in each modularized subnetwork, and carrying out loss value calculation processing based on the distance between the neurons and the absolute value of the connection weight to obtain locality punishment loss; Determining a sum of the locality penalty and the regularization penalty as the brain-like optimization penalty.
7. The method for predicting a water gate water flow field based on brain-like heuristics according to claim 1, wherein the step of performing brain-like heuristics training on the modularized subnetwork set based on the total loss function to obtain a brain-like heuristics separated physical information neural network comprises the following steps: And adopting a staged training strategy, and respectively carrying out brain-like heuristic training treatment on the modularized subnetwork set based on the total loss function in a preliminary fitting stage, a sparse evolution stage and a fine tuning stage, so that the modularized subnetwork set fits the features of a flow field, and the brain-like heuristic separated physical information neural network is obtained, wherein the preliminary fitting stage, the sparse evolution stage and the fine tuning stage are divided by a preset iteration frequency threshold.
8. A floodgate rivers flow field prediction device based on class brain heuristic, characterized in that, the device includes: the data acquisition module acquires a calculation region of the sluice, boundary conditions, initial conditions and control equations of the calculation region, and acquires training data, wherein the training data comprises configuration points, boundary constraint points and initial condition constraint points in the calculation region; The grid construction module is used for respectively processing each dimension in the coordinate values of the training data through mutually independent modularized subnetwork sets to obtain a feature vector set, and carrying out tensor decomposition type feature fusion processing on the feature vector set to obtain a predicted value of a flow field variable; And the model training module is used for constructing a total loss function by utilizing the predicted value of the flow field variable, the boundary condition, the initial condition and the control equation, performing brain-like heuristic training on the modularized sub-network set based on the total loss function to obtain a brain-like heuristic separated physical information neural network, and predicting the flow field of the drain gate under different working conditions by utilizing the brain-like heuristic separated physical information neural network to obtain a target predicted result.
9. A server comprising a processor and a memory, the memory storing computer executable instructions executable by the processor, the processor executing the computer executable instructions to implement the method of any one of claims 1 to 7.
10. A computer readable storage medium storing computer executable instructions which, when invoked and executed by a processor, cause the processor to implement the method of any one of claims 1 to 7.

Description

Gate water flow field prediction method, device and server based on brain-like heuristics Technical Field The invention relates to the technical field of flow field prediction, in particular to a sluice water flow field prediction method, a sluice water flow field prediction device and a sluice water flow field prediction server based on brain-like heuristics. Background At present, when complex front-edge problems such as multi-scale coupling, parameterized identification, inverse problem solving and flow field reconstruction are solved based on a traditional grid-based hydrodynamic calculation method and a grid-particle-free method, challenges of overhigh calculation cost or insufficient accuracy are often faced. The related technology proposes that the known physical priori knowledge and the available flow field data can be fused through a physical information neural network by a machine learning method, so that the complex fluid problem is solved, but the scheme still has the problems of more memory consumption, lower calculation efficiency, lower calculation precision caused by spectrum deviation and the like. Disclosure of Invention Accordingly, the present invention aims to provide a sluice water flow field prediction method, a sluice water flow field prediction device and a sluice water flow field prediction server based on brain-like heuristics, which can remarkably improve prediction efficiency and prediction accuracy. The embodiment of the invention provides a sluice water flow field prediction method based on brain-like heuristics, which comprises the steps of obtaining a calculation area of a sluice, a boundary condition, an initial condition and a control equation of the calculation area, and collecting training data, wherein the training data comprise calculation area internal configuration points, boundary constraint points and initial condition constraint points, respectively processing each dimension in coordinate values of the training data through mutually independent modularized sub-network sets to obtain a feature vector set, carrying out tensor decomposition type feature fusion processing on the feature vector set to obtain a predicted value of a flow field variable, constructing a total loss function by utilizing the predicted value of the flow field variable, the boundary condition, the initial condition and the control equation, carrying out brain-like heuristics training processing on the modularized sub-network set based on the total loss function to obtain a brain-like heuristics separation type physical information neural network, and carrying out prediction processing on the flow field of the sluice under different working conditions by utilizing the brain-like heuristics separation type physical information neural network to obtain a target prediction result. In one embodiment, the training data acquisition step comprises the steps of acquiring sampling ranges of a longitudinal dimension of a river channel, a transverse dimension of the river channel, a depth direction dimension and a time dimension, carrying out Latin hypercube sampling processing on each dimension based on the sampling ranges to obtain sampling coordinate values corresponding to each dimension, combining the sampling coordinate values through Cartesian products to obtain a four-dimensional structured configuration point set, determining the four-dimensional structured configuration point set as a configuration point in a calculation region, carrying out sampling processing on a boundary surface of the calculation region to generate boundary constraint points, carrying out sampling processing on the boundary constraint points at the initial moment of the calculation region to generate initial condition constraint points, and determining the training data according to the configuration point in the calculation region, the boundary constraint points and the initial condition constraint points. In one embodiment, each dimension in the coordinate values of the training data is processed through a mutually independent modularized sub-network set to obtain a feature vector set, and the method comprises the steps of sending a river channel longitudinal coordinate, a river channel transverse coordinate, a water depth direction coordinate and a time coordinate in the coordinate values of the training data to a corresponding modularized sub-network in the modularized sub-network set to process river channel longitudinal dimension information, river channel transverse dimension information, water depth direction dimension information and time dimension information to obtain the feature vector set. In one embodiment, the step of performing tensor decomposition type feature fusion processing on the feature vector set to obtain a predicted value of a flow field variable comprises the steps of performing element-by-element multiplication processing on values of the same rank position in the feature vector set corresponding