A prediction method, system and storage medium for the angle coefficient between particles of a pebble bed
Through the prediction method based on neural network, the angular coefficient between the spherical bed particles is quickly and efficiently calculated, which solves the problem of long calculations in the prior art and improves the calculation accuracy and efficiency.
Patent Information
- Application Number
- CN202111395184.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-23
- Publication Date
- 2025-03-28
- Estimated Expiration
- 2041-11-23
AI Technical Summary
The prior art takes a long time to calculate the angle coefficient between the ball bed particles, which is difficult to meet the fast and efficient needs of engineering applications.
Using a neural network-based prediction method, by obtaining the three-dimensional coordinate data of particles, randomly generate a data set and using GPU to accelerate the calculation of angle coefficients, establish a regression model and build a neural network model for training, and obtain a prediction model for calculating the angle coefficients between particles.
The calculation time of angular coefficients is shortened, the accuracy of angular coefficients is improved, and the rapid and efficient calculation of radiated heat transfer between spherical bed particles is realized.
Smart Images
Figure CN114091318B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of angle coefficient calculation, and more particularly to a prediction method, system and storage medium for the angle coefficient between particles of a ball bed. Background Art
[0002] As one of the fourth generation reactors, the pebble bed high temperature gas-cooled reactor is a well-known advanced reactor in the global nuclear field. It has many advantages such as inherent safety and near zero carbon emissions. The analytical modeling of the heat transfer relationship between the pebble bed particles in the high temperature gas-cooled reactor affects the operation and maintenance of the entire reactor. Therefore, the establishment of a new model for the heat transfer relationship of the pebble bed reactor to more reasonably calculate the effective thermal conductivity of thermal radiation is of great significance to the reasonable design and safe operation of the pebble bed reactor.
[0003] In recent years, with the research and development of thermal radiation modeling of pebble beds, the number of models analyzed has increased. For particle flow, the discrete element method (DEM) can simulate the movement of particles very successfully, and the CFD-DEM simulation method has been widely used in the field of gas particle flow. However, most of the research on thermal radiation modeling of pebble beds focuses on developing continuous porous models using empirical formulas. Moreover, as a discrete particle model, the short-range model always needs to reduce the long-range part of thermal radiation to simplify the calculation of the angular coefficient between spherical surfaces. Mehrabian Ramin et al. studied a typical short-range model used in fixed beds and fluidized beds, which only considered the short-range radiation part of 1.5 times the particle diameter. In the study of Cheng and Yu, a numerical model only calculated the radiation flux of a single Voronoi layer in the pebble bed.
[0004] In traditional integration methods or Monte Carlo methods, it often takes a long time to calculate the angular coefficient from a free ball to all possible balls around it, which is unacceptable in engineering applications. Therefore, for those skilled in the art, how to quickly and efficiently calculate the angular coefficient and then calculate the radiation heat transfer between particles is an urgent problem to be solved. Summary of the invention
[0005] In view of this, the present invention provides a method, system and storage medium for predicting the angular coefficient between particles of a pebble bed, which can quickly and efficiently calculate the angular coefficient and then calculate the radiation heat transfer between particles.
[0006] In order to achieve the above object, the present invention adopts the following technical scheme: On the one hand, a method for predicting the angle coefficient between particles of a ball bed based on a neural network is provided, and the specific steps include the following:
[0007] Acquire a data set, randomly generate particle coordinate positions under limited conditions, and calculate corresponding angular coefficients using a GPU-accelerated numerical method, wherein the data set includes three-dimensional coordinates of particles and angular coefficients between particles, wherein the three-dimensional coordinates of particles are used as input data, and the angular coefficients between particles are used as output data;
[0008] The data set is divided into 80% training set, 10% verification set and 10% test set;
[0009] Establishing a regression model for calculating the angular coefficient, and building a neural network model based on the regression model;
[0010] Inputting the training set into the neural network model for training to obtain a prediction model;
[0011] The coordinates between the particles in the ball bed are input into the prediction model to obtain the inter-particle angular coefficients.
[0012] Optionally, the neural network model is expressed as:
[0013]
[0014] in, x is the input particle space coordinate position, y is the angular coefficient, is the activation function, is the weight matrix, and L is the number of layers of the neural network.
[0015] Optionally, the activation function is one of a logsig function, a tansig function, a poslin function, and a swish function.
[0016] Optionally, the specific steps of building the neural network model are:
[0017] Determine the number of neurons in the input layer of the neural network model according to the number of particles, wherein the neurons are the spatial dimensions of the input multiple particle coordinate vectors;
[0018] Determine the number of neurons in the output layer of the neural network model, i.e. the angular coefficient, to be 1;
[0019] Determine the number of neurons in the hidden layer of the neural network model and the number of neural network layers based on the actual data set.
[0020] Determine the appropriate activation function based on the actual data set training performance
[0021] On the other hand, a prediction system for the angle coefficient between particles of a ball bed based on a neural network is provided, comprising a data acquisition module, a data classification module, a model building module, and a training module; wherein:
[0022] The data acquisition module is used to acquire a data set, wherein the data set includes three-dimensional coordinates of particles and angular coefficients between particles, wherein the three-dimensional coordinates of particles are used as input data, and the angular coefficients between particles are used as output data;
[0023] The data classification module is used to divide the data set into 80% training set, 10% verification set and 10% test set;
[0024] The model building module is used to establish a regression model for calculating the angle coefficient, and to build a neural network model based on the regression model;
[0025] The training module is used to input the training set into the neural network model for training to obtain a prediction model.
[0026] Finally, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for predicting the angular coefficient between particles of a ball bed based on a neural network are implemented.
[0027] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method, system and storage medium for predicting the angular coefficient between particles of a ball bed. A ball bed radiation heat transfer model based on neural network theory is constructed using the neural network framework of MATLAB based on the neural network theory, which provides a new model for efficiently calculating the angular coefficient and thus calculating the radiation heat transfer between particles, shortens the calculation time of the angular coefficient, and improves the accuracy of the angular coefficient calculation. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0029] Figure 1 is a flow chart of the method of the present invention;
[0030] Figure 2 is a system structure diagram of the present invention;
[0031] Figure 3 A neural network structure diagram of the present invention;
[0032] Figure 4 (a) to Figure 4 (c) are schematic diagrams of the present invention showing a situation where three balls are not blocked;
[0033] FIG5 (a) is a diagram of two spherical coordinate data of the present invention;
[0034] FIG5( b ) is a diagram of three spherical coordinate data of the present invention;
[0035] FIG5( c ) is a diagram of four spherical coordinate data of the present invention;
[0036] FIG5( d ) is a diagram of 5 spherical coordinate data of the present invention;
[0037] FIG. 5( e ) is a diagram showing six spherical coordinate data of the present invention.
[0038] Figure 6 This is the overall idea diagram of the present invention. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] Embodiment 1 of the present invention discloses a method for predicting the angle coefficient between particles of a ball bed based on a neural network, such as Figure 1 As shown, the specific steps include the following:
[0041] S1. Obtain a data set, randomly generate particle coordinate positions under limited conditions and calculate the corresponding angular coefficients using a GPU-accelerated numerical method. The data set includes the three-dimensional coordinates of the particles and the angular coefficients between the particles. The three-dimensional coordinates of the particles are used as input data, and the angular coefficients between the particles are used as output data.
[0042] S2, divide the data set into 80% training set, 10% verification set, and 10% test set;
[0043] S3, establishing a regression model for calculating the angular coefficient, and building a neural network model based on the regression model;
[0044] S4, inputting the training set into the neural network model for training to obtain a prediction model;
[0045] S5. Input the coordinates between the ball bed particles into the prediction model to obtain the inter-particle angular coefficients.
[0046] Furthermore, the method for randomly generating particle coordinate positions under limited conditions is:
[0047] Put all the spheres on the ball bed into a three-dimensional space coordinate system, then the first sphere under study is at the origin , the second sphere is x On the positive half axis , the third sphere is on the xOy plane , the fourth sphere is in the entire space , the 5th and 6th spheres are possible throughout the space, just like the 4th sphere.
[0048] For the case of 3 to 6 balls, there are too many possibilities and they can be screened out. If one of the 3rd to 6th balls does not block the first two balls (target spheres), then it is the same as the case of the total number minus 1 sphere. To avoid repeated counting, all these cases are screened out.
[0049] Regarding the unobstructed case, take three balls as an example (assuming that the first ball is at the origin , the second ball is on the positive half axis of the x-axis :
[0050] If the third ball is to the left of the first ball ,As shown in Figure 4(a), there is no occlusion;
[0051] If the third ball is to the right of the second ball ,As shown in Figure 4(b), there is no occlusion;
[0052] If the third ball is between the first and second balls , but the distance between the center of the third ball and the line connecting the centers of the first two balls is greater than twice the radius , as shown in Figure 4 (c), there is still no occlusion.
[0053] The above three situations are consistent with the situation of 2 balls and do not need to be calculated repeatedly.
[0054] The following is a detailed analysis of the situations of 2 to 6 balls:
[0055] (1) For the case of two spheres, as shown in Figure 5 (a), it is relatively simple. First, the distance cannot be too close (so that they overlap each other), and second, the distance cannot be too far (in actual situations, if the distance between the two spheres is too far, then the straight channel between the two spheres is likely to be completely blocked by other spheres without any gap, and no thermal radiation can occur between the two spheres). Therefore, the distance between the centers of the two spheres can be set to be greater than twice the radius and less than eight times the radius, that is: .
[0056] (2) For the case of three balls, the first two balls are consistent with (1), as shown in Figure 5(b). The third ball is on the xOy plane. x On the axis, the third ball situation and The situation is axisymmetric, so we can set the third ball If the third ball does not block the heat radiation of the first two balls, this situation should be excluded: ,and , and it is also necessary to exclude the situation where the third ball overlaps with the first two balls.
[0057] (3) For the case of four balls, as shown in Figure 5(c), the requirements for the first three balls are the same as those in (2). The three points form a plane, so the fourth ball should be within the three-dimensional space. The symmetric situation mentioned in (2) is as follows: and The situation is symmetric about the plane where the centers of the first three balls are located, so we can assume that , and considering the unobstructed situation, the coordinates of the center of the fourth ball are required to be: , , , and considering the spatial relationship, we should further filter by analogy with the situation where the three balls are not blocked: In addition, it is necessary to exclude the situation where the third ball overlaps with the first two balls (the distance between the sphere centers is less than the sum of the two radii).
[0058] (4) For the case of 5 balls, as shown in Figure 5(d), the coordinate requirements for the first 4 balls are the same as those in (3), and the requirement for the 5th ball is: , , ,and , and then exclude the overlapping parts between particles on this basis.
[0059] (5) For the case of 6 balls, as shown in Figure 5(e), the coordinate requirements for the first 5 balls are the same as those in (4), and the requirement for the 6th particle is: , , ,and , and then exclude the overlapping parts between particles on this basis.
[0060] This embodiment generates 100 sets, 1000 sets and 10,000 sets of coordinate data for 2 balls respectively, and generates 100 sets, 1000 sets, 10,000 sets and 100,000 sets of coordinate data for 3 to 6 balls respectively.
[0061] Furthermore, after obtaining the three-dimensional coordinate data of the particles, the corresponding angular coefficient is calculated using GPU acceleration. According to the concept of angular coefficient, it can be known that the value range of the angular coefficient is [0,1]. As the coordinates of the ball change, the value of the angular coefficient will also change accordingly. This is not a discrete value, but a continuous value. Therefore, this embodiment uses a regression model to predict the calculation rule of the angular coefficient. The regression model is: the activation function of the output layer of the neural network model is set to the y=x identity function.
[0062] Further, for the neural network model of the present invention, the dimension of its input layer is determined by the number of coordinates of the corresponding particles, because a set of coordinate data calculates an angle coefficient, so the dimension of the output layer is 1, and the number of hidden layers and the number of neurons in each layer also need to be determined according to the specific situation. In the design, the total parameters of the network generally do not exceed 1 / 10 of the number of data sets. From 2 balls to 6 balls, the number of hidden layers can be gradually increased, 2 hidden layers of 2 balls can be used, 2 to 3 hidden layers of 3 balls, and the last 6 balls can reach dozens of layers. The activation function of the output layer uses the identity function y=x. The activation function of the hidden layer can use logsig function, tansig function, poslin function and swish function. In this embodiment, the logsig function is used.
[0063] Use fitnet (function fitting neural network) in matlab to build the framework of the neural network. After building the neural network model and adjusting the activation function, import the data set into the neural network model. However, not all the data sets imported into the neural network are used as training sets, and verification and testing are required. In this embodiment, 80% of the data set is set as a training set, 10% is used for verification, and 10% is used for testing. The training set is used to train the model and update parameters, the verification set is used to change the hyperparameters of the model, and the test set is used to verify the generalization of the model.
[0064] The schematic diagram of the deep neural network with activation function used in this embodiment is as follows Figure 3 As shown in the figure, w and b are weight matrix and bias respectively. The weight matrix and bias coefficient are trained to obtain the trained prediction model.
[0065] The basic element of a multi-layer neural network is a neuron. Each neuron has real-valued inputs and outputs, which is different from formal neurons that only allow discrete values. In addition, a differentiable increasing function, such as the sigmoid function, can also be used as an activation function.
[0066] Assume that a neuron has multiple different output values , due to the different binding strengths between neurons, the actual weighted sum is:
[0067]
[0068] This is the total input of the neuron. u is called the activity value, and the bias b is added to the activity value. ， Then, through the action of the activation function f, the result obtained is the output h of the neuron.
[0069]
[0070] The layered structure connected by such neurons is called a multi-layer neural network, which consists of an input layer, hidden layer and an output layer.
[0071] (1) Input layer
[0072] Input Vector The number of input neurons is related to the input vector. x The number of elements of is equal, and the output of the first layer is:
[0073]
[0074] (2) Hidden Layer
[0075] Take the jth neuron in the lth layer as an example. This neuron receives The input values of each layer ， Its activity is
[0076]
[0077] in, For the Layer The weight value of the connection between the jth neuron in the lth layer.
[0078] Select The activity of the first neuron in the layer is:
[0079]
[0080] The output of a neuron is its activity value. Adding the bias of the neuron Substitute the result into formula (2).
[0081]
[0082] Different activation functions can also be used for each neuron , but the same activation function is generally used on the same layer.
[0083] These formulas can also be expressed in matrix form; the activity values of all neurons in layer l are represented by the vector To represent, the output value can be represented by a vector To express; and Layer The weights of the connections between the jth neuron in the lth layer and the jth neuron in the lth layer can be expressed as the matrix Substitute the elements into the weight matrix .
[0084]
[0085] No. The first layer has m neurons and the lth layer has n neurons.
[0086] Then the bias is represented as a column vector , so that the output of the hidden layer can be expressed as follows:
[0087]
[0088] The action of a function f on a vector is defined as the individual action of the function f on each element of the vector.
[0089] To make it easier to understand, a neuron with an output of 1 is added to all input layers and hidden layers and is recorded as
[0090]
[0091] In the Layer Neurons and layer l Introducing weights between neurons , this weight is the bias mentioned above. The following representation of the activity value already includes the bias.
[0092]
[0093] Weight Matrix The 0th column of is the bias.
[0094] (3) Output layer
[0095] For a feedforward neural network consisting of L layers, the output layer is the last layer, i.e., the Lth layer. The output value of the layer as input and transform it into the final output of the neural network
[0096]
[0097] In the output layer, in order to make the range of the activation function consistent with the range of the target variable, it is necessary to select an appropriate activation function .
[0098] (4) Function model
[0099] The multi-layer neural network starts from the input coordinate position x, and each layer of the network processes the corresponding input vector in turn, and finally gives a function model from input x to output angle coefficient y. In this embodiment, the expression of the neural network model is:
[0100]
[0101] Where x is the input coordinate position, y is the angular coefficient, is the activation function, is the weight matrix.
[0102] The present invention studies the thermal radiation in the core of a pebble bed high-temperature gas-cooled reactor. Aiming at the problem that the calculation of the core parameter angular coefficient in thermal radiation takes too long, a deep neural network regression model is constructed using MATLAB software to fit the relationship between the pebble bed coordinates and the angular coefficient. The calculation of the angular coefficient in the radiation heat transfer of particles in the pebble bed is simplified and accelerated by training the constructed neural network.
[0103] Embodiment 2 of the present invention discloses a prediction system for the angle coefficient between particles of a ball bed based on a neural network, such as Figure 2 As shown, it includes a data acquisition module, a data classification module, a model building module, and a training module; wherein,
[0104] A data acquisition module is used to acquire a data set, wherein the data set includes three-dimensional coordinates of particles and angular coefficients between particles, wherein the three-dimensional coordinates of particles are used as input data, and the angular coefficients between particles are used as output data;
[0105] Data classification module, used to divide the data set into 80% training set, 10% verification set, and 10% test set;
[0106] A model building module is used to build a regression model for calculating the angle coefficient and to build a neural network model based on the regression model;
[0107] The training module is used to input the training set into the neural network model for training to obtain a prediction model.
[0108] The present invention uses the neural network framework of MATLAB to construct a ball bed radiation heat transfer model based on the neural network theory, and provides a new model for efficiently calculating the angle coefficient and thus calculating the radiation heat transfer between particles. The overall idea of calculating the angle coefficient between particles of the ball bed through the multi-layer neural network model is shown in the figure. Figure 6 The method of the present invention is used to calculate the angle coefficient, which shortens the angle coefficient calculation time and improves the accuracy of the angle coefficient calculation.
[0109] Finally, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a method for predicting the angular coefficient between particles of a ball bed based on a neural network are implemented.
[0110] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0111] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for predicting the angle coefficient between particles of a pebble bed based on a neural network, characterized in that: The specific steps include the following: Obtain a data set, randomly generate particle coordinate positions under limited conditions, and use a GPU-accelerated numerical method to calculate the corresponding angular coefficients; The data set includes three-dimensional coordinates of particles and angular coefficients between particles, the three-dimensional coordinates of particles are used as input data, and the angular coefficients between particles are used as output data; Divide the data set into 80% training set, 10% verification set, and 10% test set; Establishing a regression model of particle angle coefficient, and building a neural network model based on the regression model; Inputting the training set into the neural network model for training to obtain a prediction model; Inputting the coordinates between the particles of the ball bed into the prediction model to obtain the inter-particle angular coefficient; The expression of the neural network model is: y=f (l) (W (L) f( L-1) (W (L-1) f (L-2) (…W (2) f (1) (x)))) Among them, x is the input particle space coordinate position, y is the angular coefficient, and f (l) is the activation function, W (L) is the weight matrix, L is the number of layers of the neural network; The method for randomly generating particle coordinate positions under limited conditions is: Put all the spheres on the ball bed into a three-dimensional space coordinate system. The first sphere under study is at the origin (0,0,0), the second sphere is on the positive half axis of the x-axis (x2,0,0), the third sphere is on the xoy plane (x3,y3,0), and the fourth sphere is in the entire space (x4,y4,z4). The fifth and sixth spheres are possible in the entire space just like the fourth sphere. For the cases of 3 to 6 balls, we screen out the cases where one of the 3rd to 6 balls does not block the first two balls, which is the same as the case where the total number is reduced by 1 ball. To avoid repeated counting, we screen out all these cases. Regarding the unobstructed case, take three balls as an example, the first ball is at the origin (0,0,0), and the second ball is at the positive half axis of the x-axis (x2,0,0): The third ball is on the left of the first ball, and its x3 < x1 = 0, so there is no occlusion. x1 represents the x-axis coordinate of the first ball. The third ball is to the right of the second ball, its x3 > x2, and there is no occlusion; The third ball is between the first and second balls, its 0 = x1 < x3 < x2, but the distance between the center of the third ball and the line connecting the centers of the first two balls is greater than twice the radius, its y3 > 2, and there is still no occlusion; The above three situations are consistent with the situation of two balls, and there is no need to repeat the calculation; The following is a detailed analysis of the situations of 2 to 6 balls: (1) For the case of two spheres, the distance cannot be too close so that they overlap, and the distance cannot be too far. In actual situations, if the distance between the two spheres is too far, then the straight channel between the two spheres is likely to be completely blocked by other spheres without any gaps, and no thermal radiation can occur between the two spheres. The distance between the centers of the two spheres is greater than two times the radius and less than eight times the radius, that is, 2 < x2 < 8; (2) For the case of 3 balls, the first two balls are the same as in (1). The third ball is in the xoy plane. Since the first two balls are both on the x-axis, the cases where y3 > 0 and y3 < 0 for the third ball are axisymmetric. So, we can assume y3 > 0 for the third ball. Additionally, if the third ball does not block the thermal radiation of the first two balls, this case needs to be excluded, that is: 0 < x3 < x2 and 0 < y3 < 2. Also, the case where the third ball overlaps with the first two balls needs to be excluded; (3) For the case of 4 balls, the requirements for the first 3 balls are the same as (2). The three points form a plane, so the fourth ball should be in the three-dimensional space. The symmetry mentioned in (2) is as follows: the fourth ball is symmetric with respect to the plane where the centers of the first three balls are located when z4 > 0 and z4 < 0, so we can assume that z4 > 0. Considering the unobstructed situation, the coordinates of the center of the fourth ball are required to be: 0 < x4 < x2, -2 < y4 < 2, 0 < z4 < 2. Considering the spatial relationship, we should further screen the three balls in the unobstructed situation by analogy: In addition, it is necessary to exclude the case where the third ball overlaps with the first two balls, that is, to delete the case where the distance between the ball centers is less than the sum of the radii of the two balls; (4) For the case of 5 balls, the coordinate requirements for the first 4 balls are the same as those in (3). The requirements for the 5th ball are: 0 < x5 < x2, -2 < y5 < 2, -2 < z5 < 2, and , and then, based on this, the cases where there are overlapping parts between the particles are excluded. (x5, y5, z5) are the coordinates of the 5th ball; (5) For the case of 6 balls, the coordinate requirements for the first 5 balls are the same as those in (4), and the requirements for the 6th particle are: 0 < x6 < x2, -2 < y6 < 2, -2 < z6 < 2, and , and then on this basis, we exclude the overlapping parts between particles, and (x6, y6, z6) is the coordinate of the sixth ball.
2. The method for predicting the angle coefficient between particles of a pebble bed based on a neural network according to claim 1, characterized in that: The activation function is one of the logsig function, tansig function, poslin function, and swish function.
3. The method for predicting the angle coefficient between particles of a pebble bed based on a neural network according to claim 1, characterized in that: The specific steps for building the neural network model are as follows: Determine the number of neurons in the input layer of the neural network model according to the number of particles. The neuron is the spatial dimension for inputting multiple particle coordinate vectors; Determine the number of neurons in the output layer of the neural network model, that is, the view factor, which is 1; Determine the number of neurons in the hidden layer of the neural network model and the number of neural network layers according to the actual dataset; Determine the activation function according to the actual dataset for training.
4. The method for predicting the angle coefficient between particles of a pebble bed based on a neural network according to claim 1, characterized in that: The training set is used to train the model and update the parameters. The validation set is used to change the hyperparameters of the model. The test set is used to test the generalization of the model.
5. A prediction system for the angle coefficient between particles of a ball bed based on a neural network, characterized in that: It includes a data acquisition module, a data classification module, a model establishment module, and a training module; among them, The data acquisition module is used to acquire the dataset. The dataset includes the three-dimensional coordinates of the particles and the view factor between the particles. The three-dimensional coordinates of the particles are used as input data, and the view factor between the particles is used as output data; The data classification module is used to divide the dataset into an 80% training set, a 10% validation set, and a 10% test set; The model establishment module is used to establish a regression model for calculating the view factor and build a neural network model according to the regression model; The training module is used to input the training set into the neural network model for training to obtain a prediction model; The expression of the neural network model is: y=f (l) (W (L) f( L-1) (W (L-1) f (L-2) (…W (2) f (1) (x)))) Among them, x is the input particle space coordinate position, y is the angular coefficient, and f (l) is the activation function, W (L) is the weight matrix, L is the number of layers of the neural network; The method for randomly generating the particle coordinate positions under the limited conditions is: Put all the spheres on the ball bed into a three-dimensional space coordinate system. The first sphere under study is at the origin (0, 0, 0), the second sphere is on the positive x-axis (x2, 0, 0), the third sphere is in the xoy plane (x3, y3, 0), the fourth sphere is in the whole space (x4, y4, z4), and the fifth to sixth spheres may be in the whole space like the fourth sphere; For the cases of 3 to 6 balls, filter them. If a certain ball among the third to sixth balls does not block the first two balls, it is the same as the case of the total number minus 1 sphere, avoiding duplicate calculations, and all these cases are filtered out; Regarding the unobstructed case, taking 3 balls as an example, the first ball is at the origin (0, 0, 0), and the second ball is on the positive x-axis (x2, 0, 0): The third ball is on the left of the first ball, with x3 < x1 = 0 and no blockage. x1 represents the x-axis coordinate of the first ball; The third ball is on the right of the second ball, with x3 > x2 and no blockage; The third ball is between the first and second balls 0 = x1 < x3 < x2, but the distance between the center of the third ball and the line connecting the centers of the first two balls is greater than twice the radius, y3 > 2, and there is still no occlusion; The above three situations are consistent with the situation of two balls, and there is no need to repeat the calculation; The following is a detailed analysis of the situations of 2 to 6 balls: (1) For the case of two spheres, the distance cannot be too close so that they overlap, and the distance cannot be too far. In actual situations, if the distance between the two spheres is too far, then the straight channel between the two spheres is likely to be completely blocked by other spheres without any gaps, and no thermal radiation can occur between the two spheres. The distance between the centers of the two spheres is greater than two times the radius and less than eight times the radius, that is, 2 < x2 < 8; (2) For the case of three balls, the first two balls are consistent with (1), and the third ball is on the xoy plane. Since the first two balls are on the x-axis, the case of y > 0 for the third ball is symmetrical with the case of y < 0, so we can assume that y3 > 0 for the third ball. In addition, if the third ball does not block the thermal radiation of the first two balls, this situation should be excluded, that is, the requirements are: 0 < x3 < x2, and 0 < y3 < 2. In addition, the situation where the third ball overlaps with the first two balls should be excluded. (3) For the case of 4 balls, the requirements for the first 3 balls are the same as in (2). The three points form a plane. Then the fourth ball should be within three-dimensional space. The symmetric situation mentioned in (2) here is that the cases of z4 > 0 and z4 < 0 for the fourth ball are symmetric about the plane where the centers of the first three balls lie. So, it can be assumed that z4 > 0. At the same time, considering the non-occlusion situation, the requirements for the coordinates of the center of the fourth ball are: 0 < x4 < x2, -2 < y4 < 2, 0 < y4 < 2. And considering the spatial relationship, it should be further screened by analogy with the non-occlusion situation of 3 balls: In addition, the situation where the third ball overlaps with the first 2 balls needs to be excluded, where the distance between the centers is less than the sum of the two radii; (4) For the case of 5 balls, the coordinate requirements for the first 4 balls are the same as those in (3). The requirements for the 5th ball are: 0 < x5 < x2, -2 < y5 < 2, -2 < z5 < 2, and , and then on this basis, exclude the cases where there are overlapping parts between the particles. (x5, y5, z5) are the coordinates of the 5th ball; (5) For the case of 6 balls, the coordinate requirements for the first 5 balls are the same as those in (4), and the requirements for the 6th particle are: 0 < x6 < x2, -2 < y6 < 2, -2 < z6 < 2, and , and then on this basis, we exclude the overlapping parts between particles, (x6, y6, z6) is the coordinates of the 6th ball.
6. A computer storage medium, characterized in that: The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for predicting the angular coefficient between particles of a ball bed based on a neural network as described in any one of claims 1 to 4.