Simulation and prediction method for indoor radioactivity distribution

Abstract

The invention discloses a simulation and prediction method for indoor radioactive distribution, and the method comprises the steps: constructing an indoor building model, carrying out the radioactive absorption simulation of different radionuclide sources, and obtaining the photon flux of a corresponding position in a room of the indoor building model; calculating an ideal absorbed dose rate of room irradiation under different radionuclide source specific activity conditions; constructing a training data set by taking the photon flux as input data and the target absorbed dose rate as output data, training the random forest regression model, and outputting the trained random forest regression model; and inputting photon fluxes of radionuclides at different positions in a room collected by a photon detector into the trained random forest regression model to obtain the distribution condition of the irradiation absorbed dose rate of the room in the room. The method can be used for radioactivity level analysis in a multi-target scene, and solves the problem that the prior radioactivity level analysis of the building material can only be calculated through an empirical formula and is not accurate enough.

Description

Technical Field

[0001] The invention relates to the field of indoor radioactivity research, and in particular to a simulation prediction method for indoor radioactivity distribution. Background Art

[0002] The radioactive hazards of building materials mainly include external exposure and internal exposure. External exposure mainly comes from the radioactive nuclides uranium, thorium and their decay products contained in them. These nuclides attack human cells by emitting gamma rays, causing tissue damage and thus producing radioactive hazards. The main nuclides are 226Ra, 232Th and 40K. Internal exposure is produced by the decay daughter 222Rn of the nuclide 226Ra in the air of the building. Radon is an inert gas, colorless and odorless. Since radon gas itself decays and produces radioactive decay products, when the human body breathes, the decay products of radon can be inhaled by the lungs. These decay products further decay into alpha particles that damage human tissues, eventually leading to lung cancer.

[0003] In the existing technology, when studying the radioactivity in a room, the air absorption dose rate at the center of the standard room model can be obtained through an empirical formula to preliminarily evaluate the radioactivity level of building materials, and the air absorption dose rate at a specific location can only be obtained through a complex calculation program; currently, empirical formulas such as the internal exposure index and the external exposure index are mainly used to evaluate the radioactivity level of building materials, and no simulation and quantification of the actual irradiation scene is performed; in fact, the distribution of indoor radioactivity is also affected by the thickness and density of the wall, the size of the doors and windows, and the structure of the house. Therefore, the existing evaluation methods are mainly to quickly judge the radioactivity of building materials, and the accuracy and radioactivity distribution in the actual application process still need to be further studied. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a simulation prediction method for indoor radioactivity distribution, which can accurately predict the absorbed dose rate by the photon flux at different indoor locations of a building through simulation of radioactivity at different indoor locations based on machine learning.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A method for simulating and predicting indoor radioactivity distribution is provided, which comprises the following steps:

[0007] S1: Construct an indoor building model, simulate the radioactive absorption of different radionuclide sources on the indoor building model, and obtain the photon flux at the corresponding position in the room of the indoor building model;

[0008] S2: according to the specific activity of the radionuclide source used in the radioactive absorption simulation process in step S1, the specific activity of each radionuclide source is substituted into the standard room irradiation model to calculate the ideal absorbed dose rate of the room irradiation under different radionuclide source specific activities;

[0009] S3: Using the photon flux obtained in step S1 as input data and the target absorbed dose rate obtained in step S2 as output data, constructing a training data set, inputting it into the random forest regression model, training the random forest regression model, and outputting the trained random forest regression model;

[0010] S4: The photon flux of radionuclides at different positions in the room collected by the photon detector is input into the trained random forest regression model, and the radiation absorption dose rate at different positions in the room is output to obtain the distribution of the radiation absorption dose rate in the indoor room.

[0011] Further, step S1 includes:

[0012] S11: Construct an indoor building model in the OpenMC software according to the set indoor room size, wall thickness and density, and determine the radionuclide source term;

[0013] S12: Calculate the photon source intensity S of each radionuclide source based on the specific activity of the radionuclide source term u ;

[0014]

[0015] Where u is the number of the radionuclide source, C u is the specific activity of the radionuclide source, ρ is the wall density, E i is the energy of gamma photon emission, Y i is the branching ratio corresponding to the gamma photon emission energy, i is the number of gamma rays in the radiation process of the radionuclide source, and I is the number of gamma rays;

[0016] S13: Based on the calculated photon source intensity S of each radionuclide source u , the photon source intensity S of each radionuclide source u Assign to different areas of the wall of the indoor building model and calculate the gamma intensity S of the wall v ; S v =S u ×V v , where v is the wall number of the indoor building model, V v is the wall volume;

[0017] S14: Divide the interior of the room of the indoor architectural model into n 1 ×n 2 ×n3 The regular grid, n 1 ,n 2 ,n 3 are the number of regular grids distributed in length, width and height respectively. The length Δx, width Δy and height Δz of each grid are;

[0018]

[0019] Among them, L x ,L y ,L z are the length, width and height of the room.

[0020] S15: Use the Tally module in OpenMC software to record the gamma intensity S of the wall v The flux distribution of gamma rays inside the room under the condition, and the number of photons N of each regular grid inside the room are obtained p , p is the number of the regular grid;

[0021] S16: According to the number of photons N p Calculate the photon flux φ at the center of each regular grid p ;

[0022]

[0023] Where A is the volume of the regular grid, Δt is the time interval for recording the gamma-ray flux;

[0024] S17: Return to step S12 and reset the specific activity of the radionuclide source to C u ′, and satisfies C u ′≠C u , and execute steps S12-S16 to calculate the specific activity of the radionuclide source as C u The photon flux at the point where the center of each regular grid is located at ′;

[0025] S18: Until the specific activity of the radionuclide source is reset m times After that, the photon flux at the center of each regular grid is obtained after each resetting of the specific activity of the radionuclide source, and the photon flux data at the corresponding position in the room simulated by the radionuclide source is obtained. is the photon flux at the center of the regular grid p after the specific activity of the radionuclide source is reset for the mth time;

[0026] S19: Repeat steps S12-S18 to obtain the photon flux data of the corresponding position in the room simulated by each radionuclide source Reset the specific activity of the radionuclide source for the Uth radionuclide source at the center of the regular grid p for the mth time The photon flux obtained is U, where U is the number of types of radionuclide sources.

[0027] Furthermore, the method for calculating the radiation absorbed dose rate of the room in step S2 is:

[0028]

[0029] Among them, D m is the ideal absorbed dose rate for all radionuclide sources in the room, λ u is the ideal absorption coefficient of the u-th radionuclide source.

[0030] Further, step S3 includes:

[0031] S31: Retrieve the photon flux corresponding to each reset of the specific activity of each radionuclide source as input data And calculate the ideal absorbed dose rate D in step S2 m As output data y m =D m , forming a training data set (x m ,y m );

[0032] S32: Create a training data set (X, Y), X = (x 1 ,x 2 ,…,x m ), Y=(y 1 ,y 2 ,…,y m ), X is the input training data set, y is the output training data set;

[0033] S33: Build a random forest regression model;

[0034]

[0035] Among them, y is the output of the random forest regression model, g is the number of the decision tree, and f is the g (x) is the prediction function of the g-th decision tree, x is the input variable, n trees is the number of decision trees, j is the number of training samples of the nodes in the g-th decision tree, y j is the training target value of the jth node in the gth decision tree, n L is the number of training samples of leaf nodes in the g-th decision tree;

[0036] S34: Randomly extract M training data groups with replacement from the training data set (X, Y) as training samples to generate the training set S of the prediction function of the g-th decision tree g ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x M ,y M )},(x M ,y M ) is the Mth training data group extracted;

[0037] S35: The training set S g Input the prediction function f of the g-th decision tree g (x), for the prediction function f g (x) is trained. During the training process, when each node of the g-th decision tree is split, the input data Randomly select several sub-data as input subset x′ M ,

[0038] S36: Using input data Selected input subset x′ M Train the split nodes and calculate the node impurity of each node after training;

[0039]

[0040] Among them, t is the node number, d is the training sample number of the input node, I(g t ) is the impurity of the t-th node, |D| is the number of training sample nodes under node t, y d is the target absorbed dose rate of the node’s training sample, It is the average absorbed dose rate output after node training;

[0041] S37: Calculate the total impurity I(g) after the g-th decision tree is split;

[0042]

[0043] Among them, |D 1 | is the number of left nodes of the decision tree split, |D 2 | is the number of right nodes of the decision tree split, is the left node impurity, is the impurity of the right node;

[0044] S38: Set the minimum threshold I of the total impurity after the decision tree splits 0 , if I(g)≤I0 , then the decision tree stops node splitting, and the currently split node is used as the leaf node of the g-th decision tree; if I(g)>I 0 , then return to step S35 and re-split the nodes of the g-th decision tree until I(g)≤I 0 When , output the leaf node of the g-th decision tree;

[0045] S39: Repeat steps S34-S38, train each decision tree in the random forest regression model in turn, obtain the leaf nodes of each decision tree, and output the trained random forest regression model.

[0046] The beneficial effects of the present invention are as follows: the present invention can be used for radioactivity level analysis in multi-target scenarios, solving the problem that the radioactivity level analysis of building materials can only be calculated by empirical formulas and is not accurate enough. By simulating the design of radionuclide source terms, geometric parameters, and detectors of indoor building models and training photon flux and absorbed dose rate by machine learning, the indoor radioactivity distribution under different wall thickness, density, material and other structural parameters can be predicted and simulated, providing more sophisticated data support for radioactivity level monitoring in buildings.

[0047] The present invention can reduce the complex influence of wall self-shielding and energy attenuation effect of gamma rays propagating in the air on the simulation results in actual research by means of a black box, and can realize the simulation of indoor radioactivity level distribution in multiple application scenarios, especially for laboratory design with strict requirements on background radiation dose, and can provide certain data support. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Flow chart of the simulation prediction method for indoor radioactivity distribution. DETAILED DESCRIPTION

[0049] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0050] like Figure 1 As shown, a simulation prediction method for indoor radioactivity distribution includes the following steps:

[0051] S1: Construct an indoor building model, perform radioactive absorption simulation of different radioactive nuclide sources on the indoor building model, and obtain the photon flux at the corresponding position in the room of the indoor building model. Step S1 specifically includes:

[0052] S11: According to the set indoor room size, wall thickness and density, build an indoor building model in the OpenMC software and determine the radionuclide source term. The indoor room structure of this embodiment is constructed with standard size, 4m long, 5m wide, 2.8m high, 0.2m thick wall, and 2350kg / m density. 3 The determination of radioactive nuclide source items is based on the naturally existing uranium, thorium and potassium-40 radionuclides in building materials. The main parameters include the activity concentration unit Bq / kg of the three nuclides 226Ra, 232Th and 40K; the main gamma ray energy of the nuclides and their branching ratios; potassium-40 (K-40): main energy 1460.8keV, emission probability 0.1066; uranium series (Ra-226): including multiple gamma rays released by the decay of 214Pb and 214Bi; thorium series (Th-232): including multiple gamma rays released by the decay of 228Ac and 212Pb.

[0053] S12: Calculate the photon source intensity S of each radionuclide source based on the specific activity of the radionuclide source term u ;

[0054]

[0055] Where u is the number of the radionuclide source, C u is the specific activity of the radionuclide source, ρ is the wall density, E i is the energy of gamma photon emission, Y i is the branching ratio corresponding to the gamma photon emission energy, i is the number of gamma rays in the radiation process of the radionuclide source, and I is the number of gamma rays;

[0056] S13: Based on the calculated photon source intensity S of each radionuclide source u , the photon source intensity S of each radionuclide source u Assign to different areas of the wall of the indoor building model and calculate the gamma intensity S of the wall v ; S v =S u ×V v , where v is the wall number of the indoor building model, V v is the wall volume;

[0057] S14: Divide the interior of the room of the indoor architectural model into n 1 ×n 2 ×n 3 The regular grid, n 1 ,n 2 ,n 3are the number of regular grids distributed in length, width and height respectively. The length Δx, width Δy and height Δz of each grid are;

[0058]

[0059] Among them, L x ,L y ,L z are the length, width and height of the room.

[0060] S15: Use the Tally module in OpenMC software to record the gamma intensity S of the wall v The flux distribution of gamma rays inside the room under the condition, and the number of photons N of each regular grid inside the room are obtained p , p is the number of the regular grid;

[0061] S16: According to the number of photons N p Calculate the photon flux φ at the center of each regular grid p ;

[0062]

[0063] Where A is the volume of the regular grid, Δt is the time interval for recording the gamma-ray flux;

[0064] S17: Return to step S12 and reset the specific activity of the radionuclide source to C u ′, and satisfies C u ′≠C u , and execute steps S12-S16 to calculate the specific activity of the radionuclide source as C u The photon flux at the point where the center of each regular grid is located at ′;

[0065] S18: Until the specific activity of the radionuclide source is reset m times After that, the photon flux at the center of each regular grid is obtained after each resetting of the specific activity of the radionuclide source, and the photon flux data at the corresponding position in the room simulated by the radionuclide source is obtained. is the photon flux at the center of the regular grid p after the specific activity of the radionuclide source is reset for the mth time;

[0066] S19: Repeat steps S12-S18 to obtain the photon flux data of the corresponding position in the room simulated by each radionuclide source Reset the specific activity of the radionuclide source for the Uth radionuclide source at the center of the regular grid p for the mth time The photon flux obtained is U, where U is the number of types of radionuclide sources.

[0067] In this embodiment, the matplotlib module in the OpenMC software can be used to draw a heat map of the gamma photon flux and the distribution of the photon flux at different slices along the X, Y, and Z axis directions.

[0068] S2: According to the specific activity of the radionuclide source used in the radioactive absorption simulation process in step S1, the specific activity of each radionuclide source is substituted into the standard room irradiation model to calculate the ideal absorbed dose rate of the room irradiation under different radionuclide source specific activities. The method for calculating the room irradiation absorbed dose rate in step S2 is:

[0069]

[0070] Among them, D m is the ideal absorbed dose rate for all radionuclide sources in the room, λ u is the ideal absorption coefficient of the u-th radionuclide source.

[0071] In this embodiment, the dimensions of the concrete room are 4m long, 5m wide, and 2.8m high, and the thickness and density of the wall are 20cm and 2350kg / m3 respectively. The calculation formula for the absorbed dose rate of the entire room is:

[0072] D=0.92C Ra +1.1C Th +0.08C K ;

[0073] Where D is the absorbed dose rate of the whole room, in nGy / h; C Ra , C Th , C K They are the specific activities of 226Ra, 232Th and 40K, respectively, in Bq / kg.

[0074] S3: Use the photon flux obtained in step S1 as input data and the target absorbed dose rate obtained in step S2 as output data to construct a training data set, input it into the random forest regression model, train the random forest regression model, and output the trained random forest regression model. Step S3 specifically includes:

[0075] S31: Retrieve the photon flux corresponding to each reset of the specific activity of each radionuclide source as input data And calculate the ideal absorbed dose rate D in step S2 m As output data y m =D m , forming a training data set (x m ,y m );

[0076] S32: Create a training data set (X, Y), X = (x 1 ,x 2 ,…,x m ), Y=(y 1 ,y 2 ,…,y m ), X is the input training data set, y is the output training data set;

[0077] S33: Build a random forest regression model;

[0078]

[0079] Among them, y is the output of the random forest regression model, g is the number of the decision tree, and f is the g (x) is the prediction function of the g-th decision tree, x is the input variable, n trees is the number of decision trees, j is the number of training samples of the nodes in the g-th decision tree, y j is the training target value of the jth node in the gth decision tree, n L is the number of training samples of leaf nodes in the g-th decision tree;

[0080] S34: Randomly extract M training data groups with replacement from the training data set (X, Y) as training samples to generate the training set S of the prediction function of the g-th decision tree g ={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x M ,y M )},(x M ,y M ) is the Mth training data group extracted;

[0081] S35: The training set S g Input the prediction function f of the g-th decision tree g (x), for the prediction function f g (x) is trained. During the training process, when each node of the g-th decision tree is split, the input data Randomly select several sub-data as input subset x′ M ,

[0082] S36: Using input data Selected input subset x′ M Train the split nodes and calculate the node impurity of each node after training;

[0083]

[0084] Among them, t is the node number, d is the training sample number of the input node, I(g t ) is the impurity of the t-th node, |D| is the number of training sample nodes under node t, y d is the target absorbed dose rate of the node’s training sample, It is the average absorbed dose rate output after node training;

[0085] S37: Calculate the total impurity I(g) after the g-th decision tree is split;

[0086]

[0087] Among them, |D 1 | is the number of left nodes of the decision tree split, |D 2 | is the number of right nodes of the decision tree split, is the left node impurity, is the impurity of the right node;

[0088] Each decision tree selects a random sample to obtain a training subset, and the feature selection during node splitting is a subset randomly selected from the training subset. This randomness ensures the diversity between different decision trees and reduces the risk of overfitting.

[0089] S38: Set the minimum threshold I of the total impurity after the decision tree splits 0 , if I(g)≤I 0 , then the decision tree stops node splitting, and the currently split node is used as the leaf node of the g-th decision tree; if I(g)>I 0 , then return to step S35 and re-split the nodes of the g-th decision tree until I(g)≤I 0 When , output the leaf node of the g-th decision tree;

[0090] S39: Repeat steps S34-S38, train each decision tree in the random forest regression model in turn, obtain the leaf nodes of each decision tree, and output the trained random forest regression model.

[0091] S4: The photon flux of radionuclides at different positions in the room collected by the photon detector is input into the trained random forest regression model, and the radiation absorption dose rate at different positions in the room is output to obtain the distribution of the radiation absorption dose rate in the indoor room.

[0092] The present invention can be used for radioactivity level analysis in multi-target scenarios, solving the problem that the radioactivity level analysis of building materials can only be calculated by empirical formulas and is not accurate enough. Through the simulation design of radionuclide source terms, geometric parameters, and detectors of indoor building models, and the training of photon flux and absorbed dose rate by machine learning, the indoor radioactivity distribution under different wall thickness, density, material and other structural parameters can be predicted and simulated, providing more refined data support for radioactivity level monitoring in buildings.

[0093] The present invention can reduce the complex influence of wall self-shielding and energy attenuation effect of gamma rays propagating in the air on the simulation results in actual research by means of a black box, and can realize the simulation of indoor radioactivity level distribution in multiple application scenarios, especially for laboratory design with strict requirements on background radiation dose, and can provide certain data support.

Claims

1. A simulation prediction method for indoor radioactivity distribution, characterized in that: The following steps are involved: S1: Construct an indoor building model, simulate the radioactive absorption of different radionuclide sources on the indoor building model, and obtain the photon flux at the corresponding position in the room of the indoor building model; S2: according to the specific activity of the radionuclide source used in the radioactive absorption simulation process in step S1, the specific activity of each radionuclide source is substituted into the standard room irradiation model to calculate the ideal absorbed dose rate of the room irradiation under different radionuclide source specific activities; S3: Using the photon flux obtained in step S1 as input data and the target absorbed dose rate obtained in step S2 as output data, constructing a training data set, inputting it into the random forest regression model, training the random forest regression model, and outputting the trained random forest regression model; S4: The photon flux of radionuclides at different positions in the room collected by the photon detector is input into the trained random forest regression model, and the radiation absorption dose rate at different positions in the room is output to obtain the distribution of the radiation absorption dose rate in the indoor room.

2. The method for simulating and predicting indoor radioactivity distribution according to claim 1, characterized in that: The step S1 comprises: S11: Construct an indoor building model in the OpenMC software according to the set indoor room size, wall thickness and density, and determine the radionuclide source term; S12: Calculate the photon source intensity S of each radionuclide source based on the specific activity of the radionuclide source term u ; Where u is the number of the radionuclide source, C u is the specific activity of the radionuclide source, ρ is the wall density, E i is the energy of gamma photon emission, Y i is the branching ratio corresponding to the gamma photon emission energy, i is the number of gamma rays in the radiation process of the radionuclide source, and I is the number of gamma rays; S13: Based on the calculated photon source intensity S of each radionuclide source u , the photon source intensity S of each radionuclide source u Assign to different areas of the wall of the indoor building model and calculate the gamma intensity S of the wall v ; S v =S u ×V v , where v is the wall number of the indoor building model, V v is the wall volume; S14: Divide the interior of the room of the indoor building model into regular grids of n1×n2×n3, where n1, n2, and n3 are the number of regular grids distributed in the length, width, and height directions, respectively, and the length Δx, width Δy, and height Δz of each grid are; Among them, L x ,L y ,L z are the length, width and height of the room. S15: Use the Tally module in OpenMC software to record the gamma intensity S of the wall v The flux distribution of gamma rays inside the room under the condition, and the number of photons N of each regular grid inside the room are obtained p , p is the number of the regular grid; S16: According to the number of photons N p Calculate the photon flux φ at the center of each regular grid p ; Where A is the volume of the regular grid, Δt is the time interval for recording the gamma-ray flux; S17: Return to step S12 and reset the specific activity of the radionuclide source to C u ′, and satisfies C u ′≠C u , and execute steps S12-S16 to calculate the specific activity of the radionuclide source as C u The photon flux at the point where the center of each regular grid is located at ′; S18: Until the specific activity of the radionuclide source is reset m times After that, the photon flux at the center of each regular grid is obtained after each resetting of the specific activity of the radionuclide source, and the photon flux data at the corresponding position in the room simulated by the radionuclide source is obtained. is the photon flux at the center of the regular grid p after the specific activity of the radionuclide source is reset for the mth time; S19: Repeat steps S12-S18 to obtain the photon flux data of the corresponding position in the room simulated by each radionuclide source Reset the specific activity of the radionuclide source for the Uth radionuclide source at the center of the regular grid p for the mth time The photon flux obtained is U, where U is the number of types of radionuclide sources.

3. The method for simulating and predicting indoor radioactivity distribution according to claim 2, characterized in that: The method for calculating the radiation absorption dose rate of the room in step S2 is: Among them, D m is the ideal absorbed dose rate for all radionuclide sources in the room, λ u is the ideal absorption coefficient of the u-th radionuclide source.

4. The method for simulating and predicting indoor radioactivity distribution according to claim 3, characterized in that: The step S3 comprises: S31: Retrieve the photon flux corresponding to each reset of the specific activity of each radionuclide source as input data And calculate the ideal absorbed dose rate D in step S2 m As output data y m =D m , forming a training data set (x m ,y m ); S32: Create a training data set (X, Y), X = (x1, x2, ..., x m ), Y=(y1,y2,…,y m ), X is the input training data set, y is the output training data set; S33: Build a random forest regression model; Among them, y is the output of the random forest regression model, g is the number of the decision tree, and f is the g (x) is the prediction function of the g-th decision tree, x is the input variable, n trees is the number of decision trees, j is the number of training samples of the nodes in the g-th decision tree, y j is the training target value of the jth node in the gth decision tree, n L is the number of training samples of leaf nodes in the g-th decision tree; S34: Randomly extract M training data groups with replacement from the training data set (X, Y) as training samples to generate the training set S of the prediction function of the g-th decision tree g ={(x1,y1),(x2,y2),…,(x M ,y M )},(x M ,y M ) is the Mth training data group extracted; S35: The training set S g Input the prediction function f of the g-th decision tree g (x), for the prediction function f g (x) is trained. During the training process, when each node of the g-th decision tree is split, the input data Randomly select several sub-data as input subset x′ M , S36: Using input data Selected input subset x′ M Train the split nodes and calculate the node impurity of each node after training; Among them, t is the node number, d is the training sample number of the input node, I(g t ) is the impurity of the t-th node, |D| is the number of training sample nodes under node t, y d is the target absorbed dose rate of the node’s training sample, It is the average absorbed dose rate output after node training; S37: Calculate the total impurity I(g) after the g-th decision tree is split; Among them, |D1| is the number of left nodes of the decision tree split, |D2| is the number of right nodes of the decision tree split, is the left node impurity, is the right node impurity; S38: Set the minimum threshold value I0 of the total impurity after the decision tree splits. If I(g)≤I0, stop the decision tree node splitting, and the currently split node is used as the leaf node of the g-th decision tree; if I(g)>I0, return to step S35, re-split the node of the g-th decision tree until I(g)≤I0 is satisfied, and output the leaf node of the g-th decision tree; S39: Repeat steps S34-S38, train each decision tree in the random forest regression model in turn, obtain the leaf nodes of each decision tree, and output the trained random forest regression model.

Images