CN112784496A - Method and device for predicting motion parameters of hydrodynamics and storage medium - Google Patents
Method and device for predicting motion parameters of hydrodynamics and storage medium Download PDFInfo
- Publication number
- CN112784496A CN112784496A CN202110126239.3A CN202110126239A CN112784496A CN 112784496 A CN112784496 A CN 112784496A CN 202110126239 A CN202110126239 A CN 202110126239A CN 112784496 A CN112784496 A CN 112784496A
- Authority
- CN
- China
- Prior art keywords
- motion
- moving object
- motion parameter
- parameter
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Fluid Mechanics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Mathematical Optimization (AREA)
- Algebra (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a method, a device and a storage medium for predicting a motion parameter of hydrodynamics, wherein the method comprises the following steps: acquiring the motion time of each detection point of the target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time; inputting the obtained motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process. Therefore, the prior information is coupled in the neural network model, and the training of the neural network model can be completed through a small amount of samples, so that the robustness and the generalization capability of the model can be effectively improved, and a calculation result with higher precision can be obtained.
Description
Technical Field
The present application relates to the field of fluid mechanics technologies, and in particular, to a method and an apparatus for predicting a motion parameter of fluid mechanics, and a storage medium.
Background
Computational fluid dynamics is a typical interdisciplinary discipline that uses numerical methods and computer simulations to solve real-world physical, biological, and chemical problems. Such as determining the strength of the wing structural material by calculating the pressure at the wing surface.
In the prior art, the following method is adopted to calculate parameters in fluid mechanics: modeling a continuous calculation domain in the problem, dividing the calculation domain into a limited number of discrete grid nodes by using grid lines, selecting a proper path to convert a differential equation and a definite solution condition thereof into a corresponding algebraic equation set on the grid nodes, and solving the discrete equation set by using a computer to obtain approximate solutions of parameters such as fluid speed, pressure and the like on the grid nodes. However, the calculation result precision is poor due to sparse grid in the calculation domain, and the calculation efficiency is low due to too dense grid; due to the existence of the grid, the numerical method can only obtain the predicted values of parameters such as speed and pressure at grid nodes, and the positions among the grid nodes can only be obtained by other methods, such as interpolation, which may have adverse effects on the accuracy of the calculation result.
Disclosure of Invention
In view of this, an object of the present application is to provide a method, an apparatus, and a storage medium for predicting a motion parameter of hydrodynamics, in which prior information is coupled in a neural network model, and training of the neural network model can be completed through a small number of samples, so that not only can the robustness and the generalization capability of the model be effectively improved, but also a calculation result with higher accuracy can be obtained.
In a first aspect, the present application provides a method for predicting a motion parameter of fluid mechanics, the method comprising:
acquiring the motion time of each detection point of a target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time;
inputting the obtained motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
Preferably, the motion parameter prediction model is trained by:
the method comprises the steps of obtaining the motion time of each detection sample point of a moving object sample in the motion process, the position coordinate of each detection sample point corresponding to the motion time, the prior value of the motion parameter of the moving object sample in an prior information equation, and the actual value of the motion parameter of the moving object sample in the motion process;
based on an error back propagation algorithm, training the constructed neural network model through the motion time of each detection sample point of the moving object sample and the position coordinate corresponding to the motion time, the motion parameter prior value of the moving object sample and the motion parameter actual value of the moving object sample to obtain a trained motion parameter prediction model.
Preferably, the neural network model is trained by:
inputting the motion time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the motion time into the neural network model to obtain an initial motion parameter predicted value of the moving object sample in the motion process;
determining a mean square error loss function based on the initial motion parameter predicted value of the moving object sample and the motion parameter actual value of the moving object sample;
replacing the motion parameter prior value of the moving object sample with the initial motion parameter predicted value, and determining a prior loss function;
determining a model loss function of the motion parameter prediction model based on the mean square error loss function and the prior loss function;
and updating the parameters of the neural network model based on the model loss function to obtain the trained neural network model.
Preferably, the mean square error loss function is calculated by the following formula:
where loss1 represents the mean square error loss function, N represents the number of samples, yiRepresenting the actual value of said motion parameter, f (x)i) Representing the initial motion parameter predictor.
Preferably, the replacing the motion parameter prior value of the moving object sample with the initial motion parameter prediction value to determine a prior loss function includes:
constructing a prior information equation according to a parameter change rule of a moving object sample in a moving process, wherein the prior information equation takes the moving time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the moving time as independent variables, and takes a moving parameter prior value of the moving object sample as a dependent variable;
replacing the dependent variable in the prior information equation with the initial motion parameter predicted value to obtain an auxiliary function;
determining an a priori loss function from the auxiliary function.
Preferably, the a priori loss function is calculated by the formula:
where loss2 represents the prior loss function, N represents the number of samples, hiRepresenting the auxiliary function.
Preferably, the prior information equation is a navier-stokes equation.
In a second aspect, the present application further provides a kinematic parameter prediction device for fluid mechanics, the kinematic parameter prediction device comprising:
the parameter acquisition module is used for acquiring the motion time of each detection point of the target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time;
the model application module is used for inputting the acquired motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
In a third aspect, the present application further provides an electronic device, including: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is operating, the machine-readable instructions when executed by the processor performing the steps of the fluid-dynamic motion parameter prediction method as described above.
In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when being executed by a processor, performs the steps of the fluid-dynamic motion parameter prediction method as described above.
The application provides a method, a device and a storage medium for predicting motion parameters of hydrodynamics, wherein the method for predicting the motion parameters comprises the following steps: acquiring the motion time of each detection point of a target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time; inputting the obtained motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
Compared with the method for predicting the kinetic parameters of the hydrodynamics by carrying out grid division on a calculation domain and discretizing a equation in the prior art, the method has the advantages that the prior information is coupled in the neural network model, the training of the neural network model can be completed through a small amount of samples, the robustness and the generalization capability of the model can be effectively improved, and the calculation result with higher precision can be obtained.
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.
FIG. 1 is a flow chart of a method for predicting a fluid dynamic motion parameter provided in an embodiment of the present application;
FIG. 2 is a flow chart of a method for training a neural network model according to an embodiment of the present disclosure;
FIG. 3 is a block diagram of a process for training a neural network model according to an embodiment of the present disclosure;
FIG. 4 is a schematic flow chart illustrating a method for constructing an auxiliary function according to an embodiment of the present disclosure;
FIG. 5 is a schematic diagram of a fluid dynamic motion parameter prediction device according to an embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. Every other embodiment that can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present application falls within the protection scope of the present application.
First, an application scenario to which the present application is applicable will be described. The method can be applied to fluid mechanics, and mainly researches the static state and the moving state of the fluid and the interaction and the flow law when the fluid and the solid boundary wall have relative motion. The Navier-Stokes equations (N-S) describing fluid motion are a class of classical nonlinear partial differential equations. That is, in the production life and industrial fields, many problems can be solved as a nonlinear system process, and the solving of the problems is essentially to solve nonlinear control equations of the system, the nonlinear control equations are equations for describing nonlinear mapping relations between independent variables and dependent variables, and in most systems, the control equations exist in the form of nonlinear partial differential equations. The equation has wide application in meteorological science, fluid machinery, aerodynamics and other aspects, and describes the change rule of state quantities such as the speed, the pressure and the like of fluid in time space. Therefore, the research on the solution of the N-S equation can help us to understand the principle and the law of fluid motion and effectively predict the speed and the pressure of the fluid under certain conditions. Such as determining the material strength of the wing by calculating the magnitude of the wing surface forces.
In the prior art, because an N-S equation of an actual problem is quite complex, an analytic solution of the N-S equation cannot be obtained so far, and the N-S equation is mostly approximately solved by using a numerical method in the field of Computational Fluid Dynamics (CFD), so that the Fluid mechanics problem is analyzed and simulated. The numerical method mainly utilizes the idea of discretization, the discretization mainly comprises discretization of a calculation domain and discretization of an equation, and the method can be divided into a finite difference method, a finite volume method and the like according to different equation discretization modes.
The discretization of the calculation domain refers to modeling the continuous calculation domain in the problem, dividing the calculation domain into limited discrete grid nodes by using grid lines, and the discretization of the equation refers to selecting a proper path to convert a differential equation and a definite solution condition thereof into a corresponding algebraic equation set on the grid nodes. After the discretization is finished, the discretized equation set can be solved by using a computer to obtain approximate solutions of the fluid velocity, pressure and other state quantities on the grid nodes.
However, the numerical method currently used in CFD has the following disadvantages:
for the fluid flow problem with complex geometric shapes, a large amount of time is consumed for modeling the geometric bodies, and the labor cost is increased; the division of the calculation domain grids has great influence on the calculation efficiency and the result precision, the result precision is poor under the condition of sparse grids, and the problem of low calculation efficiency is caused by too dense grids; due to the existence of the grids, the numerical method can only obtain the predicted values of the state quantities such as the speed, the pressure and the like at the grid nodes, and the positions among the grid nodes can only be obtained by other methods, such as an interpolation method, which may bring adverse effects on the precision of the calculation result; different equation discretization methods are only effective for different problems, and in practical problems, the algorithm is poor in applicability.
Based on this, the embodiments of the present application provide a method and an apparatus for predicting a motion parameter of hydrodynamics, and a storage medium, which utilize the strong nonlinear modeling capability of deep learning to solve the problem of realizing approximate solution of an N-S equation without discretization, i.e., without performing mesh division on a computational domain and discretization on an equation, and further can directly calculate the predicted values of state quantities such as velocity pressure at all positions in the entire computational domain. Because the deep learning technology belongs to a black box model and needs to depend on labeled data, the problems of data dependence and physical interpretability exist, and aiming at the problem of deep learning, the method simultaneously constructs a general method for coupling prior information in the deep learning model, so that the model can realize approximate solution to an equation under the condition of a small amount of training data, and the problems of data dependence and physical interpretability are effectively reduced.
Referring to fig. 1, fig. 1 is a flowchart illustrating a method for predicting a fluid dynamic motion parameter according to an embodiment of the present disclosure. As shown in fig. 1, a method for predicting motion parameters provided in an embodiment of the present application includes:
s110, acquiring the motion time of each detection point of the target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time.
Here, the target moving object may be a structure used in many fields of aviation, aerospace, fluid machinery, and the like, such as an aircraft, an automobile, a missile, and the like. The target moving object may be divided into a plurality of detection points, and the position coordinates of the detection points corresponding to the movement time may be obtained in a coordinate system established in advance for each detection point during the movement of the target moving object. The target moving object is in continuous motion, and during the motion process, the position coordinates of each detection point of the target moving object change, so the coordinates of the detection points in the coordinate system need to be determined according to the motion time.
S120, inputting the acquired motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
Here, the motion time of each detection point and the position coordinates corresponding to the motion time are used as the input of the motion parameter prediction model, and the motion parameter prediction value of the target moving object in the motion process can be obtained through the prediction of the motion parameter prediction model.
The predicted value of the motion parameter may be a predicted value of a state quantity such as a velocity and a pressure, or may be a predicted value of a medium physical property parameter.
Specifically, the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process, namely, the parameter change rule is experience and historical data obtained before a test for obtaining a sample. The priori information is combined with the neural network model, then the neural network model is trained through a small amount of training sample data, and the trained model is used for obtaining the predicted values of state quantities or physical parameters such as speed, pressure and the like of other data in the calculated domain.
In specific implementation, the nonlinear partial differential equation is reconstructed through input and output of a deep learning model (a neural network model), prior information in the nonlinear partial differential equation is converted into an auxiliary function, a model loss function is modified by the auxiliary function, the equation and the deep learning model are combined, and the equation is approximately solved.
The application provides a method, a device and a storage medium for predicting motion parameters of hydrodynamics, wherein the method for predicting the motion parameters comprises the following steps: acquiring the motion time of each detection point of a target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time; inputting the obtained motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process. Compared with the method for predicting the kinetic parameters of the hydrodynamics by carrying out grid division on a calculation domain and discretizing a equation in the prior art, the method has the advantages that the prior information is coupled in the neural network model, the training of the neural network model can be completed through a small amount of samples, the robustness and the generalization capability of the model can be effectively improved, and the calculation result with higher precision can be obtained.
In the embodiment of the present application, as a preferred embodiment, the embodiment of the present application trains the motion parameter prediction model by:
the method comprises the steps of obtaining the motion time of each detection sample point of a moving object sample in the motion process, the position coordinate of each detection sample point corresponding to the motion time, the prior value of the motion parameter of the moving object sample in an prior information equation, and the actual value of the motion parameter of the moving object sample in the motion process.
The motion parameter prediction model is input into the motion time of each detection sample point, the position coordinates of each detection sample point corresponding to the motion time, and the motion parameter prior value and the motion parameter actual value which are needed by the motion parameter prediction model in the training process, and the training of the motion parameter prediction model can be completed through the parameter values.
Based on an error back propagation algorithm, training the constructed neural network model through the motion time of each detection sample point of the moving object sample and the position coordinate corresponding to the motion time, the motion parameter prior value of the moving object sample and the motion parameter actual value of the moving object sample to obtain a trained motion parameter prediction model.
Here, the learning process of the error Back Propagation (BP) algorithm is composed of two processes of forward propagation of a signal and back propagation of an error. In forward propagation, an input sample is transmitted from an input layer, is processed layer by layer through hidden layers and is transmitted to an output layer, and if the actual output of the output layer is not consistent with the expected output, the error is transmitted to a backward propagation stage; the error back propagation is to transmit the output error back to the input layer by layer through the hidden layer in a certain form, and distribute the error to all units of each layer, so as to obtain the error signal of each layer, and the error signal is used as the basis for correcting the weight of the unit. The weight value adjustment process of each layer of signal forward propagation and error backward propagation is carried out repeatedly, the process of continuously adjusting the weight value, namely the process of network learning training, is carried out until the error of network output is reduced to an acceptable degree or is carried out to preset learning times.
The trained motion parameter prediction model can be obtained through a BP algorithm.
Preferably, referring to fig. 2, fig. 2 is a flowchart of a method for training a neural network model according to an embodiment of the present disclosure, in which the neural network model is trained according to the following steps:
s210, inputting the motion time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the motion time into the neural network model to obtain the initial motion parameter predicted value of the moving object sample in the motion process.
Here, the neural network model may be a fully-connected neural network model, a convolutional neural network model, a self-coding neural network model, or the like. The type of the model is not particularly limited. I.e. using any kind of neural network model, parameter prediction can be achieved.
In step S210, a neural network model is first built, and the fully-connected neural network model is used as a training model in the embodiment of the present application. And the motion time of each detection sample point and the position coordinate of each detection sample point corresponding to the motion time are used as input data of the neural network model, and the initial motion parameter predicted value of the moving object sample in the motion process is used as output data of the neural network model. The output data is obtained by input data through a series of nonlinear mapping in a neural network model.
S220, determining a mean square error loss function based on the initial motion parameter predicted value of the moving object sample and the motion parameter actual value of the moving object sample.
Here, after the output of the neural network model (initial motion parameter prediction value) is determined, the mean square error loss function can be calculated by comparing the output with the actual output value (motion parameter actual value) of the neural network model.
Specifically, the mean square error loss function is calculated by the following formula:
where loss1 represents the mean square error loss function, N represents the number of samples, yiRepresenting the actual value of said motion parameter, f (x)i) Representing the initial motion parameter predictor.
And S230, replacing the motion parameter prior value of the moving object sample with the initial motion parameter predicted value, and determining a prior loss function.
In the prior art, a mean square error loss function is directly used as a loss function, and model parameters are updated for the loss function by using an error back propagation algorithm, but original information of a problem is lost, so that data dependence and physical interpretability problems are caused, so that the embodiment of the application considers that prior information is coupled with a neural network model.
In specific implementation, a prior information equation is constructed according to a parameter change rule of a moving object sample in a moving process, wherein the prior information equation takes the moving time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the moving time as independent variables, and takes a moving parameter prior value of the moving object sample as a dependent variable; replacing the dependent variable in the prior information equation with the initial motion parameter predicted value to obtain an auxiliary function; determining an a priori loss function from the auxiliary function.
Here, the prior information equation is Navier-Stokes equation (Navier-Stokes, N-S). According to the practical condition of the parameters of the fluid mechanics, a simplified form of the N-S equation in a one-dimensional space can be obtained: burgers equation (Burgers equation), i.e., one-dimensional Burgers equation.
For the one-dimensional Burgers equation, the independent variable is the motion time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the motion time, and the dependent variable is the motion parameter prior value of the moving object sample.
Thus, the a priori loss function can be determined by the auxiliary function, only by minimizing the auxiliary function so that it continuously approaches 0.
Specifically, the a priori loss function is calculated by the following formula:
where loss2 represents the prior loss function, N represents the number of samples, hiRepresenting the auxiliary function.
S240, determining a model loss function of the motion parameter prediction model based on the mean square error loss function and the prior loss function.
Here, the mean square error loss function and the prior loss function are summed to obtain a model loss function of the motion parameter prediction model.
And S250, updating the parameters of the neural network model based on the model loss function to obtain the trained neural network model.
Here, after the model loss function is determined, the parameters of the model are updated using an error back propagation algorithm. The trained model can also be used for predicting speeds of other independent variables, namely other positions in the calculation domain at different moments.
Referring to fig. 3 and 4, fig. 3 is a flowchart of a process for training a neural network model according to an embodiment of the present application, fig. 4 is a schematic flowchart of a process for constructing an auxiliary function according to an embodiment of the present application, and a motion parameter prediction method according to the embodiment of the present application is described in detail below by taking a speed solution in the field of hydrodynamics as an example, specifically, a one-dimensional Burgers equation is used to describe the speed parameter prediction method according to the embodiment of the present application:
(1) the one-dimensional Burgers equation has the following form:
firstly, defining independent variables and dependent variables of a nonlinear partial differential equation, wherein in a Burgers equation, the independent variables refer to position coordinates x and motion time t, and the dependent variables refer to speed u. Here, the independent variable is represented by vector X, the dependent variable is represented by vector Y, and the Burgers equation is simplified as: Ω (X, Y) ═ 0.
(2) A neural network model is built, a fully-connected neural network is used as a training model, in fig. 3, an independent variable X is used as input data, a predicted value of a dependent variable Y is used as output data, the output data is obtained by input data through a series of nonlinear mapping in the neural network model, and the predicted value of the dependent variable output by the model can be expressed as f (X) assuming the transformation performed by the model as a function f.
(3) Since the approximate solution problem of the equation belongs to the regression problem, after the output of the neural network model is determined, the output can be compared with the true value (actual value of the motion parameter) of the dependent variable to construct the mean square error loss function of the multidimensional data:
(4) since the model outputs the predicted value f (X) of the dependent variable, f (X) may be used instead of the dependent variable Y in the equation Ω, and an auxiliary function h ═ Ω (X, f (X)) may be established, as shown in fig. 4, so that it is equivalent to approximately solving the equation by minimizing h and continuously approaching 0.
(5) In order to couple the prior information into the model, the auxiliary function h needs to satisfy the equation to be solved, i.e. h needs to be close to 0 continuously in the model training process, so the prior loss function loss2 can be determined by h:
and combining it with the mean square error loss, as a model loss function loss:
loss＝loss1+loss2；
(6) and determining a model loss function, and updating the parameters of the model by using an error back propagation algorithm, so that the trained model can predict the speed of the target moving object at other positions at different moments.
It should be noted that, since the governing equation in the fluid mechanics mostly exists in the form of a nonlinear partial differential equation, the solution of the N-S equation in the fluid mechanics is taken as an example to describe the scheme in detail. It can be understood that the method provided by the embodiment of the present application is also applicable to nonlinear partial differential equations in other fields such as finance, electricity, aerodynamics, etc., and is not limited to the nonlinear partial differential equations, and for nonlinear ordinary differential equations, linear equations, equation sets, etc., even equation sets with boundary conditions and initial conditions, the method can be used to couple prior information to achieve the effect of approximate solution of equations, and details are not repeated here.
In the method for predicting the motion parameters of the fluid mechanics provided by the embodiment of the application, the prior information is considered to be equal to the control equation for describing the fluid mechanics problem, so that the essence of fusing the prior information in the embodiment of the application is to use a universal and normative mode to carry out deep coupling on the N-S equation and the deep learning model so as to complete the construction process of the model. Specifically, a nonlinear partial differential equation is reconstructed through input and output of a neural network model, an auxiliary function is introduced, the problem of solving the equation is converted into the problem of minimizing the auxiliary function, the problem is combined with the idea of minimizing a loss function in deep learning, prior information in the equation is successfully coupled into the deep learning model, then the deep learning model is trained through a small amount of training data, and the trained model is used for obtaining the predicted values of state quantities or physical property parameters such as speed, pressure and the like of other data in a calculation domain, so that the purpose of approximately solving the N-S equation is achieved.
Therefore, the method for predicting the motion parameter of the fluid mechanics provided by the embodiment of the application has the following advantages: the method has no grid dependency, does not need to divide a calculation domain into grids, does not need to discretize an equation, and can realize approximate solution of the N-S equation only by a small amount of training data; the method has no limitation on a deep learning model, and any neural network model can realize approximate solution of an N-S equation through the method; the method has strong generalization capability and high solving accuracy, and can effectively improve the robustness and the generalization capability of the model due to the coupling of prior information and the equivalence of the second term prior loss to the function of a regularization term in statistics; the method has low requirement on data quantity and physical interpretability, can still realize equation solution with high precision under the condition of few samples, and has certain physical interpretability due to the existence of prior information.
Based on the same inventive concept, the embodiment of the present application further provides a hydromechanical motion parameter prediction device corresponding to the hydromechanical motion parameter prediction method, and as the principle of the device in the embodiment of the present application for solving the problem is similar to the hydromechanical motion parameter prediction method in the embodiment of the present application, the implementation of the device may refer to the implementation of the method, and repeated details are not repeated.
Referring to fig. 5, fig. 5 is a schematic structural diagram of a fluid dynamic motion parameter prediction apparatus according to an embodiment of the present disclosure. As shown in fig. 5, the motion parameter prediction apparatus 500 includes:
a parameter obtaining module 510, configured to obtain a movement time of each detection point of the target moving object in a movement process, and a position coordinate of each detection point corresponding to the movement time;
the model application module 520 is configured to input the acquired motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model, so as to obtain a motion parameter prediction value of the target moving object in a motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
In the embodiment of the present application, as a preferred embodiment, the motion parameter prediction apparatus 500 further includes a model training module 530, and the model training module 530 is configured to train the motion parameter prediction model by:
the method comprises the steps of obtaining the motion time of each detection sample point of a moving object sample in the motion process, the position coordinate of each detection sample point corresponding to the motion time, the prior value of the motion parameter of the moving object sample in an prior information equation, and the actual value of the motion parameter of the moving object sample in the motion process;
based on an error back propagation algorithm, training the constructed neural network model through the motion time of each detection sample point of the moving object sample and the position coordinate corresponding to the motion time, the motion parameter prior value of the moving object sample and the motion parameter actual value of the moving object sample to obtain a trained motion parameter prediction model.
Preferably, the model training module 530 is configured to train the neural network model by:
inputting the motion time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the motion time into the neural network model to obtain an initial motion parameter predicted value of the moving object sample in the motion process;
determining a mean square error loss function based on the initial motion parameter predicted value of the moving object sample and the motion parameter actual value of the moving object sample;
replacing the motion parameter prior value of the moving object sample with the initial motion parameter predicted value, and determining a prior loss function;
determining a model loss function of the motion parameter prediction model based on the mean square error loss function and the prior loss function;
and updating the parameters of the neural network model based on the model loss function to obtain the trained neural network model.
Preferably, the model training module 530 is configured to calculate the mean square error loss function by the following formula:
where loss1 represents the mean square error loss function, N represents the number of samples, yiRepresenting the actual value of said motion parameter, f (x)i) Representing the initial motion parameter predictor.
Preferably, when the model training module 530 is configured to replace the motion parameter prior value of the moving object sample with the initial motion parameter prediction value to determine the prior loss function, the model training module 530 is configured to:
constructing a prior information equation according to a parameter change rule of a moving object sample in a moving process, wherein the prior information equation takes the moving time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the moving time as independent variables, and takes a moving parameter prior value of the moving object sample as a dependent variable;
replacing the dependent variable in the prior information equation with the initial motion parameter predicted value to obtain an auxiliary function;
determining an a priori loss function from the auxiliary function.
Preferably, the model training module 530 is configured to calculate the prior loss function by the following equation:
where loss2 represents the prior loss function, N represents the number of samples, hiRepresenting the auxiliary function.
Preferably, the prior information equation is a navier-stokes equation.
The hydrodynamics motion parameter prediction device provided by the embodiment of the application comprises a parameter acquisition module, a model training module and a model application module, wherein the parameter acquisition module is used for acquiring the motion time of each detection point of a target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time; the model training module trains the motion parameter prediction model by: the model application module inputs the acquired motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fusing prior information, and the prior information is used for describing a parameter change rule of a target moving object in a motion process. Therefore, the prior information is coupled in the neural network model, training of the neural network model can be completed through a small amount of samples, the robustness and the generalization capability of the model can be effectively improved, and a calculation result with higher precision can be obtained.
Referring to fig. 6, fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As shown in fig. 6, the electronic device 600 includes a processor 610, a memory 620, and a bus 630.
The memory 620 stores machine-readable instructions executable by the processor 610, when the electronic device 600 runs, the processor 610 communicates with the memory 620 through the bus 630, and when the machine-readable instructions are executed by the processor 610, the steps of the fluid dynamics motion parameter prediction method in the embodiment of the method shown in fig. 1 and the steps of training the neural network model in the embodiment of the method shown in fig. 2 may be performed.
An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method for predicting a kinetic parameter of fluid mechanics in the embodiment of the method shown in fig. 1 and the steps of training a neural network model in the embodiment of the method shown in fig. 2 may be executed.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the exemplary embodiments of the present application, and are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A method for predicting a kinetic parameter of fluid mechanics, the method comprising:
acquiring the motion time of each detection point of a target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time;
inputting the obtained motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
2. The method of claim 1, wherein the athletic parameter prediction model is trained by:
the method comprises the steps of obtaining the motion time of each detection sample point of a moving object sample in the motion process, the position coordinate of each detection sample point corresponding to the motion time, the prior value of the motion parameter of the moving object sample in an prior information equation, and the actual value of the motion parameter of the moving object sample in the motion process;
based on an error back propagation algorithm, training the constructed neural network model through the motion time of each detection sample point of the moving object sample and the position coordinate corresponding to the motion time, the motion parameter prior value of the moving object sample and the motion parameter actual value of the moving object sample to obtain a trained motion parameter prediction model.
3. The method of claim 2, wherein the neural network model is trained by:
inputting the motion time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the motion time into the neural network model to obtain an initial motion parameter predicted value of the moving object sample in the motion process;
determining a mean square error loss function based on the initial motion parameter predicted value of the moving object sample and the motion parameter actual value of the moving object sample;
replacing the motion parameter prior value of the moving object sample with the initial motion parameter predicted value, and determining a prior loss function;
determining a model loss function of the motion parameter prediction model based on the mean square error loss function and the prior loss function;
and updating the parameters of the neural network model based on the model loss function to obtain the trained neural network model.
4. A method for motion parameter prediction according to claim 3, characterized in that the mean square error loss function is calculated by the following formula:
where loss1 represents the mean square error loss function, N represents the number of samples, yiRepresenting the actual value of said motion parameter, f (x)i) Representing the initial motion parameter predictor.
5. The method according to claim 3, wherein the replacing the priori value of the motion parameter of the moving object sample with the predicted value of the initial motion parameter and determining an a priori loss function comprises:
constructing a prior information equation according to a parameter change rule of a moving object sample in a moving process, wherein the prior information equation takes the moving time of each detection sample point of the moving object sample and the position coordinate of each detection sample point corresponding to the moving time as independent variables, and takes a moving parameter prior value of the moving object sample as a dependent variable;
replacing the dependent variable in the prior information equation with the initial motion parameter predicted value to obtain an auxiliary function;
determining an a priori loss function from the auxiliary function.
6. The motion parameter prediction method of claim 5, wherein the prior loss function is calculated by the following formula:
where loss2 represents the prior loss function, N represents the number of samples, hiRepresenting the auxiliary function.
7. The method of claim 2, wherein the prior information equation is a navier-stokes equation.
8. A kinetic parameter prediction device for fluid mechanics, characterized in that the kinetic parameter prediction device comprises:
the parameter acquisition module is used for acquiring the motion time of each detection point of the target moving object in the motion process and the position coordinate of each detection point corresponding to the motion time;
the model application module is used for inputting the acquired motion time of each detection point and the position coordinate corresponding to the motion time into a pre-trained motion parameter prediction model to obtain a motion parameter prediction value of the target moving object in the motion process; the motion parameter prediction model is a neural network model fused with prior information, and the prior information is used for describing a parameter change rule of the target moving object in the motion process.
9. An electronic device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating via the bus when the electronic device is running, the processor executing the machine-readable instructions to perform the steps of the fluid dynamic motion parameter prediction method according to any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method for predicting a kinetic parameter of a fluid according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110126239.3A CN112784496A (en) | 2021-01-29 | 2021-01-29 | Method and device for predicting motion parameters of hydrodynamics and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110126239.3A CN112784496A (en) | 2021-01-29 | 2021-01-29 | Method and device for predicting motion parameters of hydrodynamics and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112784496A true CN112784496A (en) | 2021-05-11 |
Family
ID=75759760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110126239.3A Withdrawn CN112784496A (en) | 2021-01-29 | 2021-01-29 | Method and device for predicting motion parameters of hydrodynamics and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112784496A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282448A (en) * | 2021-11-16 | 2022-04-05 | 北京百度网讯科技有限公司 | Flow field information acquisition method, model training method and device and electronic equipment |
| CN114896886A (en) * | 2022-05-18 | 2022-08-12 | 北京百度网讯科技有限公司 | Flow field identification method, flow field identification device, electronic apparatus, flow field identification medium, and program product |
| CN115081677A (en) * | 2022-05-12 | 2022-09-20 | 暨南大学 | Wind speed prediction method and device, electronic equipment and readable storage medium |
| CN116341129A (en) * | 2023-02-02 | 2023-06-27 | 华北电力大学 | Simulation method and device for dynamic digital twin model of leading operation of wind turbine generator |
| WO2023151201A1 (en) * | 2022-02-10 | 2023-08-17 | 中山大学 | Fast and flexible holomorphic embedding type neural network wide-area optimization training method |
-
2021
- 2021-01-29 CN CN202110126239.3A patent/CN112784496A/en not_active Withdrawn
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282448A (en) * | 2021-11-16 | 2022-04-05 | 北京百度网讯科技有限公司 | Flow field information acquisition method, model training method and device and electronic equipment |
| WO2023151201A1 (en) * | 2022-02-10 | 2023-08-17 | 中山大学 | Fast and flexible holomorphic embedding type neural network wide-area optimization training method |
| CN115081677A (en) * | 2022-05-12 | 2022-09-20 | 暨南大学 | Wind speed prediction method and device, electronic equipment and readable storage medium |
| CN114896886A (en) * | 2022-05-18 | 2022-08-12 | 北京百度网讯科技有限公司 | Flow field identification method, flow field identification device, electronic apparatus, flow field identification medium, and program product |
| CN116341129A (en) * | 2023-02-02 | 2023-06-27 | 华北电力大学 | Simulation method and device for dynamic digital twin model of leading operation of wind turbine generator |
| CN116341129B (en) * | 2023-02-02 | 2023-09-26 | 华北电力大学 | Simulation method and device for dynamic digital twin model of leading operation of wind turbine generator |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112784496A (en) | Method and device for predicting motion parameters of hydrodynamics and storage medium | |
| Belbute-Peres et al. | Combining differentiable PDE solvers and graph neural networks for fluid flow prediction | |
| Baque et al. | Geodesic convolutional shape optimization | |
| Campana et al. | Shape optimization in ship hydrodynamics using computational fluid dynamics | |
| Toal et al. | Geometric filtration using proper orthogonal decomposition for aerodynamic design optimization | |
| Leifsson et al. | Simulation-driven aerodynamic design using variable-fidelity models | |
| Zhou et al. | An active learning radial basis function modeling method based on self-organization maps for simulation-based design problems | |
| Pelamatti et al. | Overview and comparison of gaussian process-based surrogate models for mixed continuous and discrete variables: Application on aerospace design problems | |
| Yamazaki et al. | Geometry parameterization and computational mesh deformation by physics-based direct manipulation approaches | |
| Da Ronch et al. | Adaptive design of experiments for efficient and accurate estimation of aerodynamic loads | |
| Tang et al. | Generation of aerodynamic data using a design of experiment and data fusion approach | |
| White et al. | Neural networks predict fluid dynamics solutions from tiny datasets | |
| Hines et al. | Graph neural networks for the prediction of aircraft surface pressure distributions | |
| Udoewa et al. | Computational fluid dynamics | |
| JP2020119189A (en) | Fluid analysis system, method for analyzing fluid, and fluid analysis program | |
| CN107526863B (en) | Optimal pressure projection method for incompressible transient and steady state NAVIER-STOKES equations | |
| Pena et al. | A surrogate method based on the enhancement of low fidelity computational fluid dynamics approximations by artificial neural networks | |
| Vanderwyst et al. | Reduced order models for generation of large, high speed aerodynamic databases with jet interactions | |
| Strönisch et al. | Flow field prediction on large variable sized 2D point clouds with graph convolution | |
| Andrés-Pérez et al. | Aerodynamic shape design by evolutionary optimization and support vector machines | |
| Nelson et al. | Experimental and numerical investigation of flight dynamics of a generic lambda wing configuration | |
| Pehlivanoglu | Direct and indirect design prediction in genetic algorithm for inverse design problems | |
| Neumann et al. | Coupling strategies for large industrial models | |
| Pan et al. | Iterative refinement of accelerations in real-time vehicle dynamics | |
| Görtz et al. | Variable-fidelity and reduced-order models for aero data for loads predictions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210511 |
|
| WW01 | Invention patent application withdrawn after publication |