A rapid prediction method for neutron radiation dose field under complex terrain conditions

Abstract

The present invention discloses a method for rapidly predicting neutron radiation dose fields under complex terrain conditions, solving the problem that existing radiation environment calculation methods are unable to achieve rapid prediction of neutron radiation dose fields under any actual complex terrain scenario. Specifically, the MC method is used to simulate the neutron radiation dose field distribution on an open surface under multiple source height conditions, and radiation data that changes in two dimensions with source height and source center projection distance are obtained as a control; then, multiple complex terrain MC calculation models based on DEM applications are established, and large-scale calculations are carried out to obtain the neutron radiation dose field distribution in complex terrain; then, based on the terrain elevation data and radiation field distribution, the characteristic parameters of the terrain samples are calculated to produce a sample data set; finally, the sample data set is used to perform model training based on an ensemble learning algorithm to obtain a rapid prediction model for the neutron radiation dose field, and the rapid prediction performance of the model is tested on a verification terrain sample.

Description

Technical Field

[0001] The present invention relates to the field of rapid prediction of neutron radiation dose fields, and in particular to a method for rapid prediction of neutron radiation dose fields under complex terrain conditions. Background Art

[0002] Neutron radiation emitted by a neutron source above the Earth's surface interacts with nitrogen and oxygen nuclei in the atmosphere before reaching the ground. Neutrons are then scattered and absorbed by the ground medium, forming a neutron radiation field. Compared to an ideal infinitely flat surface, the distribution of the radiation field is affected to varying degrees by the presence of various topographical features on the actual surface. This effect is qualitatively illustrated using typical valley and hill terrain as examples. In the case of a valley, especially when the neutron source is located within the valley, the line of sight between an observation point and the source is not blocked by other terrain features. Furthermore, the radiation level at the observation point is enhanced due to the continuous reflection of the surrounding or lateral mountains. In contrast, in the case of a hill, if the observation point is located behind a mountain, the line of sight between the observation point and the neutron source may be blocked by the hill. As a result, the hill provides a degree of shielding for the observation point. Although the radiation level at the observation point is reduced due to this shielding, the sky scattering contributes to this. Therefore, prediction methods and models for neutron radiation dose fields under actual terrain conditions must consider the influence of various complex topographical features.

[0003] Traditional research on neutron radiation environments has primarily employed three approaches: Monte Carlo (MC), deterministic methods, and deterministic-MC coupled methods. MC methods enable detailed modeling of complex terrain scenarios and provide high confidence in their results. MC simulations of long-distance particle transport in the near-surface atmosphere have been documented, such as the paper "Application of the Weighted Window Variance Reduction Method to Simulating Long-Distance Gamma-ray Transport," published in Modern Applied Physics in 2020. This method addresses the deep penetration problem encountered in long-distance particle transport, but the simulations are time-consuming and cannot achieve rapid predictions. Deterministic methods, such as the discrete ordinate (SN) method, offer rapid computational speeds but are limited in their ability to describe complex geometries, making them unsuitable for detailed modeling and calculations of realistic terrain scenarios. Deterministic-MC coupled methods complement these approaches to some extent, but their spatial description capabilities are still limited by deterministic methods, and due to the inherent MC simulation process, rapid prediction capabilities remain limited. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for quickly predicting neutron radiation dose fields under complex terrain conditions, so as to solve the technical problem that the existing radiation environment calculation method cannot achieve rapid prediction of neutron radiation dose fields in any actual complex terrain scenario.

[0005] To achieve the above-mentioned object, the present invention provides a method for rapidly predicting neutron radiation dose field under complex terrain conditions, which is special in that it comprises the following steps:

[0006] Step 1) establishing a first set of neutron radiation MC calculation models for open areas under multiple source height conditions, and simulating the first type of neutron radiation dose at different source center projection distances on each of the open areas according to the neutron source energy spectrum and intensity;

[0007] Step 2) using the first type neutron radiation dose obtained in step 1) to create a neutron radiation dose data table that varies two-dimensionally with source height and source center projection distance;

[0008] Step 3) using the terrain elevation data samples to establish a second set of neutron radiation MC calculation models for multiple actual complex terrains, simulating the second type of neutron radiation dose of each terrain unit under each actual complex terrain, and calculating the relative source height and source center projection distance of each terrain unit, combining the source height of the second set of neutron radiation MC calculation models, and using the neutron radiation dose data table obtained in step 2) to perform interpolation calculation to obtain the third type of neutron radiation dose of each terrain unit under the preset open surface;

[0009] Step 4) calculating the ratio of the second type neutron radiation dose obtained in step 3) to the third type neutron radiation dose to obtain the neutron radiation dose ratio of each terrain unit;

[0010] Step 5), for terrain samples with different terrain features, using their corresponding terrain elevation data samples, respectively perform steps 3) and 4) to obtain a first set of neutron radiation dose ratios for each terrain unit in each terrain sample;

[0011] Step 6) Calculate the relative source height, elevation angle deviation, slope aspect, visibility, and terrain openness of each terrain unit in each terrain sample respectively, and use them together with all source center projection distances calculated in step 3) as input parameters for machine learning model training, and use all neutron radiation dose ratios obtained in step 5) as output parameters for the machine learning model training, thereby completing the machine learning model training sample set;

[0012] Step 7) using an ensemble learning algorithm in a machine learning method to perform model training on the sample set obtained in step 6) to obtain a rapid prediction model for neutron radiation dose under complex terrain conditions;

[0013] Step 8) Using the rapid prediction model to rapidly predict the neutron radiation dose field under complex terrain conditions.

[0014] Furthermore, after step 7) and before step 8), a prediction model verification step is also included:

[0015] a) Based on the rapid prediction model obtained in step 7), a rapid prediction of the sample neutron radiation dose ratio is performed on the verification terrain sample;

[0016] b) performing the interpolation calculation of step 3) on the verification terrain sample to obtain the Category IV neutron radiation dose of each terrain unit in the verification terrain sample under a preset open surface, and multiplying the interpolation calculation by the sample radiation dose ratio obtained in step a) to obtain a rapid prediction result of the neutron radiation dose field of the verification terrain sample;

[0017] c) Comparing the neutron radiation dose field prediction results of the verification terrain sample with its MC simulation results to verify the actual prediction performance of the rapid prediction model:

[0018] If the prediction performance is good, save the quick prediction model and proceed to step 8);

[0019] If the prediction performance is poor, increase the terrain samples and return to step 5).

[0020] Furthermore, in step 3), the interpolation calculation method is as follows:

[0021] For any source height H and source center projection distance D, if X i <H≤X i+1 And Y j <D<Y j+1 , where the neutron radiation dose Dose(H,D) is given by Dose(X i ,Y j ), Dose(X i+1 ,Y j ), Dose(X i ,Y j+1 ) and Dose(X i+1 ,Y j+1 ) Four data points are interpolated six times and then the average is obtained; where X i and Y j They are the source height and source center projection distance data points in the two-dimensional data table, Dose(X i ,Y j ) is the data point X i and Y j The corresponding neutron radiation dose value.

[0022] Furthermore, in step 3), the specific geometric modeling method for establishing a second set of neutron radiation MC calculation models of multiple actual complex terrains using terrain elevation data samples is as follows:

[0023] For terrain data samples consisting of latitude and longitude coordinates and elevation values, two adjacent triangular terrain planes are constructed using the four elevation points that make up the grid. These planes are considered as two terrain units, and the triangular terrain planes are used as the upper surfaces of triangular prisms, so that the entire complex terrain area is composed of multiple triangular prisms.

[0024] In step 3), the source center projection distance D is the projection distance between the terrain unit and the corresponding source center.

[0025] Further, in step 6):

[0026] The relative source height of the terrain unit is the height difference between the radiation source height and the corresponding terrain unit;

[0027] The elevation angle deviation of the terrain unit is specifically the difference between the elevation angle β between the source center and the preset open ground surface and the elevation angle α between the source center and the actual complex terrain.

[0028] The specific calculation method of the slope aspect of the terrain unit is as follows:

[0029] The outer normal vector of the plane formed by the three points is calculated based on the coordinates of the three elevation points constituting the terrain unit, and the angle between the vector and the direction of the terrain unit pointing to the source center is calculated based on the coordinates of the source center of the terrain unit. The cosine value of the angle is used as the slope direction of the terrain unit.

[0030] The visibility of the terrain unit is specifically the difference between the altitude angle α formed by the source center relative to each terrain unit in the corresponding terrain sample and the maximum shielding angle γ formed by other terrains on the direction pointing to the source center to the terrain unit;

[0031] The terrain openness of the terrain unit includes the terrain openness in 16 directions. The specific calculation method of the terrain openness (Sky View Factor, SVF) is:

[0032] a. Starting from the direction of point P pointing to the source center, the corresponding azimuth angle φ0 = 0, and moving in a clockwise direction with a step length of Δφ = π / 8, the corresponding 16 azimuth angles are:

[0033] φ i =φ0+igΔφ,i=0,1,2,L,15

[0034] b. Calculate the terrain unit in all directions φ i Terrain openness on:

[0035] SVF i =1-sinθ i

[0036] Where θ i Point P is in the direction φ iThe maximum elevation angle within a certain distance L, that is, the azimuth φ i The terrain on the point P causes the maximum shielding angle; take point P as the starting point, along the azimuth φ i Draw a straight line L i , starting from point P along the straight line L i According to the step length ΔL, the elevation angles of other grid points to point P are judged in turn. ΔL can be taken as a value smaller than the grid spacing. i As the distance increases in step size ΔL, the elevation Z of any point j ij for:

[0037] Z ij =Z(x p +jΔL x ,y p +jΔL y ), j=1,2,L,N

[0038] Where, ΔL x , ΔL y is the coordinate increment in the x and y directions, ΔL x =ΔLsin(φ i ), ΔL y =ΔLcos(φ i ), the shielding angle θ of the point to the terrain unit P ij for:

[0039]

[0040] Azimuth φ i The terrain above causes the maximum shielding angle θ to point P i for:

[0041] θ i =max(θ ij ), j=1,2,L,N.

[0042] Furthermore, in step 7), the ensemble learning algorithm is a gradient boosting tree algorithm (GradientBoosting, GBDT) based on the Python open source machine learning library Scikit-Learn. The hyperparameters required by the GBDT algorithm (mainly the number of decision trees, learning rate and maximum depth) are determined by splitting the training sample set through "K-fold cross validation" and using the parameter grid search method.

[0043] Furthermore, in step 7), the neutron radiation dose prediction model has the following input parameters: relative source height, source center projection distance, altitude angle deviation, slope direction, visibility and terrain openness in 16 directions (21 in total), and the output parameter is the neutron radiation dose ratio.

[0044] Furthermore, after step 8), the method further includes:

[0045] Step 9) obtaining a verification terrain sample, performing the interpolation calculation of step 3) on the verification terrain sample, obtaining the neutron radiation dose of each terrain unit of the verification terrain sample under the preset open surface, and multiplying the result by the radiation dose ratio obtained in step 8) to obtain the neutron radiation dose field prediction result of the verification terrain sample;

[0046] Step 10) Compare the neutron radiation dose field prediction results of the verification terrain sample with its MC simulation results to verify the actual prediction performance of the neutron radiation dose rapid prediction model under the complex terrain conditions.

[0047] Furthermore, in step 10), the actual prediction performance is tested by comparing the model prediction results with the MC simulation results, and the performance evaluation index used is the determination coefficient R 2 and the mean relative percentage error MAPE, defined as follows:

[0048]

[0049]

[0050] Among them, y′i is the predicted value, yi is the actual value of the validation sample, is the average of the actual values of the validation sample.

[0051] Beneficial effects of the present invention:

[0052] 1. The present invention proposes a prediction method that uses MC simulation results to produce sample data, and then trains and generates a machine learning prediction model, and relies on relevant terrain feature parameters to quickly predict the radiation dose field. Leveraging the powerful geometric modeling capabilities and highly reliable simulation calculations of traditional MC methods, this method can achieve a realistic representation of complex terrain scenarios at the prediction model level. The training data is sufficiently reliable to strongly support the machine learning algorithm in exploring the physical laws hidden within the data.

[0053] 2. The characteristic parameters proposed or selected in the present invention include, on the one hand, intuitive expressions of terrain features such as height and slope, and on the other hand, introduce three-dimensional features such as visibility and terrain openness that reflect the impact of terrain undulations on radiation field propagation. They can highly characterize the influence of complex terrain on neutron transport in the atmosphere.

[0054] 3. The method for rapid prediction of neutron radiation dose field under complex terrain conditions proposed in this invention uses a neutron radiation dose prediction model obtained through machine learning training. For any actual terrain scenario, it can quickly provide neutron radiation dose prediction results within seconds on a conventionally configured computer. Compared with the traditional MC method that takes dozens of hours or even days to calculate on a mainframe, it significantly improves computing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Schematic diagram of the MC calculation model for neutron radiation on an open surface in an embodiment of the present invention;

[0056] Figure 2 Figure 2 is a schematic diagram of neutron radiation dose calculation using interpolation from a two-dimensional data table in an embodiment of the present invention; R3, R4, R5, and R6 are the logarithmic interpolation results of the four solid line segments, R1 and R2 are the logarithmic interpolation results of the two dashed line segments, and Dose(H, D) is the average of R1 and R2.

[0057] Figure 3 Schematic diagram of the neutron radiation MC calculation model under complex terrain conditions in an embodiment of the present invention (taking a 33×33 terrain grid as an example);

[0058] Figure 4 Schematic diagram of altitude angle deviation and visibility calculation of terrain units in an embodiment of the present invention;

[0059] Figure 5 Schematic diagram of a terrain unit being blocked by mountains on both sides in a two-dimensional plane in an embodiment of the present invention;

[0060] Figure 6 This is a flow chart of an embodiment of the method for rapid prediction of neutron radiation dose field under complex terrain conditions proposed by the present invention;

[0061] Figure 7 A schematic diagram of contour lines for verifying terrain sample 1 in an embodiment of the present invention;

[0062] Figure 8 Schematic diagram showing the comparison between the MC simulation results and the model prediction results for verifying the neutron radiation dose ratio of terrain sample 1 in an embodiment of the present invention;

[0063] Figure 9 Schematic diagram showing the comparison between the MC simulation results and the model prediction results for verifying the neutron radiation dose of terrain sample 1 in an embodiment of the present invention;

[0064] Figure 10 A schematic diagram of contour lines for verifying terrain sample 2 in an embodiment of the present invention;

[0065] Figure 11Schematic diagram showing the comparison between the MC simulation results and the model prediction results for verifying the neutron radiation dose ratio of terrain sample 2 in an embodiment of the present invention;

[0066] Figure 12 Schematic diagram comparing the MC simulation results and the model prediction results for verifying the neutron radiation dose of terrain sample 2 in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] Over the past decade, the ubiquity of big data and the continuous improvement of computing power have enabled machine learning methods to break through the limitations of traditional analytical models, rapidly identify inherent patterns in data, and become a powerful tool for solving data analysis and prediction problems influenced by the coupling of complex nonlinear characteristics. Given the various advantages of machine learning, applying machine learning methods to develop a model for predicting neutron radiation dose fields under complex terrain conditions has the potential to rapidly obtain interesting predictions while sacrificing some accuracy.

[0068] Existing published literature lacks relevant technologies for developing rapid algorithms for neutron radiation dose fields that consider complex terrain factors, particularly those that combine the use of digital elevation models (DEMs) with machine learning to establish rapid prediction models applicable to any complex terrain. This paper proposes a method that utilizes MC simulation results to generate sample data, trains a machine learning prediction model, and uses relevant terrain characteristic parameters to rapidly predict radiation dose fields.

[0069] The concept of the present invention is as follows: first, the MC method is used to simulate the neutron radiation dose field distribution of an open surface model under a series of source height conditions, and radiation data that changes in two dimensions with source height and source center projection distance are obtained as a control, providing a data source for subsequent interpolation calculation of the radiation field distribution under any source height; then, a series of complex terrain MC calculation models based on DEM application are established, and large-scale calculations are carried out to obtain neutron radiation dose field distribution data of complex terrain samples; secondly, according to the terrain elevation data and radiation field distribution, the characteristic parameters of the terrain samples are calculated, and a sample data set for machine learning model training is produced; finally, based on the ensemble learning algorithm, the sample data is used for model training to obtain a rapid prediction model of the neutron radiation dose field, and the rapid prediction performance of the model is tested on the verification terrain sample.

[0070] like Figure 6 As shown, the method for rapidly predicting neutron radiation dose field under complex terrain conditions provided by an embodiment of the present invention specifically includes the following steps:

[0071] Step 1) Establish a first set of neutron radiation MC calculation models for open areas under multiple source height conditions, and simulate the first type of neutron radiation dose at different source center projection distances on each open area based on the neutron source energy spectrum and intensity;

[0072] Specifically, the neutron radiation MC calculation models for open ground with source heights of 1m, 100m, 200m, ... 1000m were established (a total of 11 models, the model geometries are as shown in the attached Figure 1 The energy spectrum of the neutron radiation source is shown in Table 1. The first type of neutron radiation dose at different source center projection distances is obtained by simulation, and the source center projection distance range is 1 to 4000m.

[0073]

[0074] Table 1 Neutron radiation source energy spectrum

[0075] Step 2) using the first type neutron radiation dose obtained in step 1) to create a neutron radiation dose data table that varies two-dimensionally with source height and source center projection distance;

[0076] Specifically, the first type neutron radiation dose obtained in step 1) is used to prepare a neutron radiation dose data table that varies two-dimensionally with source height and source center projection distance, see Table 2 (containing only part of the data).

[0077]

[0078] Table 2 Two-dimensional data table of neutron radiation dose (rad)

[0079] Step 3) using the terrain elevation data samples to establish a second set of neutron radiation MC calculation models for multiple actual complex terrains, simulating the second type of neutron radiation dose of each terrain unit under each actual complex terrain, and calculating the relative source height and source center projection distance of each terrain unit, combining the source height of the second set of neutron radiation MC calculation models, and using the neutron radiation dose data table obtained in step 2) to perform interpolation calculation to obtain the third type of neutron radiation dose of each terrain unit under the preset open surface;

[0080] Step 4) calculating the ratio of the second type neutron radiation dose obtained in step 3) to the third type neutron radiation dose to obtain the neutron radiation dose ratio of each terrain unit;

[0081] Step 5), for terrain samples with different terrain features, use their corresponding terrain elevation data samples to perform steps 3) and 4) respectively to obtain the neutron radiation dose ratio of each terrain unit in each terrain sample;

[0082] Specifically (including steps 3 to 5), a second set of neutron radiation MC calculation models under actual complex terrain conditions was established using terrain elevation data samples with a resolution of 100m×100m (a total of 21 sets, with the same model geometry). Figure 3 The latitude and longitude range of each terrain sample is 12 km. The relevant parameters and the source height of the corresponding MC model are shown in Table 3.

[0083]

[0084] Table 3. Related parameters of terrain samples and corresponding source heights of MC models

[0085] The second type of neutron radiation dose of each terrain unit in the above 21 groups of actual terrain scenarios was simulated and calculated. The relative source height and source center projection distance of each terrain unit were calculated. Combined with the source height of the second group of neutron radiation MC calculation models, the third type of neutron radiation dose of each terrain unit in each group of terrain scenarios under the preset open surface was obtained by interpolation calculation using a two-dimensional dose data table.

[0086] like Figure 2 As shown, the above interpolation calculation method is as follows:

[0087] For any source height H and source center projection distance D, if X i <H≤X i+1 And Y j <D<Y j+1 , where the neutron radiation dose Dose(H,D) is given by Dose(X i ,Y j ), Dose(X i+1 ,Y j ), Dose(X i ,Y j+1 ) and Dose(X i+1 ,Y j+1 ) Four data points are interpolated six times and then the average is obtained; where X i and Y j They are the source height and source center projection distance data points in the two-dimensional data table, Dose(X i ,Y j ) is the data point X i and Y j The corresponding neutron radiation dose value; where the source center projection distance D is the projection distance between the terrain unit and the corresponding source center.

[0088] The neutron radiation dose ratios of each terrain unit were obtained by calculating the ratios of the second type neutron radiation doses of each terrain unit under 21 sets of actual terrain scenarios and the third type neutron radiation doses under assumed open conditions.

[0089] It should be noted that the above 21 sets of terrain elevation data samples all meet the requirement of “broadly representative terrain characteristics”, and the source height distribution range of the corresponding MC model is consistent with the open surface model in step 1).

[0090] Step 6) Calculate the relative source height, elevation angle deviation, slope aspect, visibility, and terrain openness of each terrain unit in each terrain sample respectively, and use them together with all source heights and source center projection distances calculated in step 3) as input parameters for machine learning model training, and use all neutron radiation dose ratios obtained in step 5) as output parameters for machine learning model training, thereby completing the machine learning model training sample set;

[0091] Specifically, such as Figure 4 、 Figure 5 As shown, for the 21 sets of actual terrain elevation data samples, the relative source height, elevation angle deviation, slope aspect, visibility, and terrain openness of each terrain unit in each sample are calculated respectively. Together with the source center projection distance calculated in steps 3) to 5), these are used as input parameters for machine learning model training. The calculated neutron radiation dose ratio of each terrain unit is used as the output parameter of model training, thereby completing the production of a machine learning training sample set based on the 21 sets of actual terrain elevation data. Among them, the elevation angle deviation of the terrain unit is specifically the difference between the elevation angle β formed by the source center and the preset open condition surface and the elevation angle α formed with the actual complex terrain; the relative source height of the terrain unit is specifically the difference between the radiation source height and the height of the corresponding terrain unit.

[0092] The specific calculation method of slope aspect is as follows:

[0093] The external normal vector of the plane formed by the three points is calculated based on the coordinates of the three elevation points that constitute the terrain unit, and the angle between the vector and the direction of the terrain unit pointing to the source center is calculated based on the coordinates of the source center. The cosine value of the angle is used as the slope direction of the terrain unit.

[0094] Visibility is specifically the difference between the elevation angle α of the source center relative to each terrain unit in the corresponding terrain sample and the maximum shielding angle γ of other terrains on the direction pointing to the source center to the terrain unit;

[0095] The openness of a terrain unit includes the terrain openness in 16 directions. The specific calculation method of terrain openness (Sky View Factor, SVF) is as follows:

[0096] a. Starting from the direction of point P pointing to the source center, the corresponding azimuth angle φ0 = 0, and moving in a clockwise direction with a step length of Δφ = π / 8, the corresponding 16 azimuth angles are:

[0097] φ i =φ0+igΔφ,i=0,1,2,L,15

[0098] b. Calculate the terrain unit in all directions φ i Terrain openness on:

[0099] SVF i =1-sinθ i

[0100] Where θ i Point P is in the direction φ i The maximum elevation angle within a certain distance L, that is, the azimuth φ i The terrain on the point P causes the maximum shielding angle; take point P as the starting point, along the azimuth φ i Draw a straight line L i , starting from point P along the straight line L i According to the step length ΔL, the elevation angles of other grid points to point P are judged in turn. ΔL can be taken as a value smaller than the grid spacing. i As the distance increases in step size ΔL, the elevation Z of any point j ij for:

[0101] Z ij =Z(x p +jΔL x ,y p +jΔL y ), j=1,2,L,N

[0102] Where, ΔL x , ΔL y is the coordinate increment in the x and y directions, ΔL x =ΔLsin(φ i ), ΔL y =ΔLcos(φ i ), the shielding angle θ of the point to the terrain unit P ij for:

[0103]

[0104] Azimuth φ i The terrain above causes the maximum shielding angle θ to point P i for:

[0105] θ i =max(θ ij ), j=1,2,L,N.

[0106] Step 7) using an ensemble learning algorithm in a machine learning method to perform model training on the sample set obtained in step 6) to obtain a rapid prediction model for neutron radiation dose under complex terrain conditions;

[0107] Specifically, the GBDT algorithm is used to train the neutron radiation dose prediction model. The input parameters are all the above 21 characteristic parameters, and the output parameter is the neutron radiation dose ratio.

[0108] During training, the "10-fold cross-validation" method was used to split the training sample set, and the "parameter grid search" method was used to determine the hyperparameter combination of the GBDT algorithm, and the optimal hyperparameters were obtained: the number of decision trees was 300, the learning rate was 0.05, and the maximum tree depth was 8; under this hyperparameter combination, a rapid prediction model for neutron radiation dose under complex terrain conditions was obtained.

[0109] Step 8) Using the rapid prediction model to quickly predict the neutron radiation dose field under complex terrain conditions;

[0110] Specifically, on two verification terrain samples (the contour lines are as follows Figure 7 and Figure 10 As shown in the figure, the source heights are 200m and 600m respectively), and the radiation dose ratio is quickly predicted using the trained neutron radiation dose rapid prediction model;

[0111] Step 9) performing the interpolation calculation of step 3) on the verification terrain sample to obtain the neutron radiation dose of each terrain unit of the verification terrain sample under the preset open surface, and multiplying it with the radiation dose ratio obtained in step 8) to obtain a rapid prediction result of the neutron radiation dose field of the verification terrain sample;

[0112] Specifically, the interpolation calculation of step 3) is performed on the two verification terrain samples using the two-dimensional neutron radiation dose data table to obtain the neutron radiation dose of each terrain unit in the verification terrain sample under the preset open surface, and the neutron radiation dose is multiplied by the radiation dose ratio obtained in step 8) to obtain the rapid prediction results of the neutron radiation dose field of the two verification terrain samples;

[0113] Step 10) Compare the rapid prediction results of the neutron radiation dose field of the verification terrain sample with its MC simulation results to test the actual prediction performance of the neutron radiation dose rapid prediction model under complex terrain conditions: if the prediction performance is good, save the rapid prediction model and execute step 8); if the prediction performance is poor, return to step 5) and execute steps 7)-8), add terrain samples according to the test results and retrain to obtain a rapid prediction model with better generalization performance.

[0114] Specifically, the actual prediction performance is tested by comparing the model prediction results with the MC simulation results. The performance evaluation index used is the coefficient of determination (R 2 ) and the mean relative percentage error (MAPE), defined as follows:

[0115]

[0116]

[0117] Among them, y′i is the predicted value, yi is the actual value of the validation sample, is the average of the actual values.

[0118] Specifically, as a comparison, MC simulations were performed on two validation terrain samples. Figure 8 and Figure 11 The comparison between the MC simulation results and the model prediction results of the neutron radiation dose ratio of two verification terrain samples is given. Figure 9 and Figure 12 The comparison of the neutron radiation dose MC simulation results and the model prediction results of the two verification terrain samples are given in the form of scatter plots. Table 4 gives the prediction performance evaluation of the rapid prediction model on the verification terrain samples.

[0119] In the verification terrain samples, the determination coefficients of the model prediction results are all close to 1, indicating that the model has a good degree of fit and strong explanatory ability; the neutron radiation dose prediction results are in good agreement with the MC simulation results, indicating that the above technical solution has significant reference value for engineering applications.

[0120]

[0121] Table 4 Prediction performance evaluation on validation terrain samples

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for rapidly predicting neutron radiation dose field under complex terrain conditions, characterized in that: The following steps are involved: Step 1) establishing a first set of neutron radiation MC calculation models for open areas under multiple source height conditions, and simulating the first type of neutron radiation dose at different source center projection distances on each of the open areas according to the neutron source energy spectrum and intensity; Step 2) using the first type neutron radiation dose obtained in step 1) to create a neutron radiation dose data table that varies two-dimensionally with source height and source center projection distance; Step 3) using the terrain elevation data samples to establish a second set of neutron radiation MC calculation models for multiple actual complex terrains, simulating the second type of neutron radiation dose of each terrain unit under each actual complex terrain, and calculating the relative source height and source center projection distance of each terrain unit, combining the source height of the second set of neutron radiation MC calculation models, and using the neutron radiation dose data table obtained in step 2) to perform interpolation calculation to obtain the third type of neutron radiation dose of each terrain unit under the preset open surface; Step 4) calculating the ratio of the second type neutron radiation dose obtained in step 3) to the third type neutron radiation dose to obtain the neutron radiation dose ratio of each terrain unit; Step 5), for terrain samples with different terrain features, using their corresponding terrain elevation data samples, respectively perform steps 3) and 4) to obtain a first set of neutron radiation dose ratios for each terrain unit in each terrain sample; Step 6) Calculate the relative source height, elevation angle deviation, slope aspect, visibility, and terrain openness of each terrain unit in each terrain sample respectively, and use them together with all source center projection distances calculated in step 3) as input parameters for machine learning model training, and use all neutron radiation dose ratios obtained in step 5) as output parameters for the machine learning model training, thereby completing the machine learning model training sample set; Step 7), using an ensemble learning algorithm in a machine learning method to perform model training on the sample set obtained in step 6), to obtain a rapid prediction model for neutron radiation dose under complex terrain conditions; the rapid prediction model for neutron radiation dose, the model input parameters are relative source height, source center projection distance, altitude angle deviation, slope aspect, visibility and terrain openness in 16 directions, and the output parameter is the neutron radiation dose ratio; Step 8) Using the rapid prediction model to rapidly predict the neutron radiation dose field under complex terrain conditions.

2. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 1 is characterized in that: After step 7) and before step 8), the method further includes: Steps 7-8) Testing the prediction model a) Based on the rapid prediction model obtained in step 7), a rapid prediction of the sample neutron radiation dose ratio is performed on the verification terrain sample; b) performing the interpolation calculation of step 3) on the verification terrain sample to obtain the Category IV neutron radiation dose of each terrain unit in the verification terrain sample under a preset open surface, and multiplying the interpolation calculation by the sample radiation dose ratio obtained in step a) to obtain a rapid prediction result of the neutron radiation dose field of the verification terrain sample; c) Comparing the neutron radiation dose field prediction results of the verification terrain sample with its MC simulation results to verify the actual prediction performance of the rapid prediction model: If the prediction performance is good, save the quick prediction model and proceed to step 8); If the prediction performance is poor, increase the terrain samples and return to step 5).

3. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 1 or 2, characterized in that: In step 3), the interpolation calculation method is as follows: For any source height H and source center projection distance D, if X i <H≤X i+1 And Y j <D<Y j+1 , where the neutron radiation dose Dose(H,D) is given by Dose(X i ,Y j ), Dose(X i+1 ,Y j ), Dose(X i ,Y j+1 ) and Dose(X i+1 ,Y j+1 ) Four data points are interpolated six times and then the average is obtained; where X i and Y j They are the source height and source center projection distance data points in the two-dimensional data table, Dose(X i ,Y j ) is the data point X i and Y j The corresponding neutron radiation dose value.

4. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 3 is characterized in that: In step 3), the specific geometric modeling method for establishing a second set of neutron radiation MC calculation models for multiple actual complex terrains using terrain elevation data samples is as follows: For terrain data samples consisting of latitude and longitude coordinates and elevation values, two adjacent triangular terrain planes are constructed using the four elevation points that make up the grid. These planes are considered as two terrain units, and the triangular terrain planes are used as the upper surfaces of triangular prisms, so that the entire complex terrain area is composed of multiple triangular prisms. In step 3), the source center projection distance D is the projection distance between the terrain unit and the corresponding source center.

5. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 4 is characterized in that: In step 6): The relative source height of the terrain unit is the height difference between the radiation source height and the corresponding terrain unit; The elevation angle deviation of the terrain unit is specifically the difference between the elevation angle between the source center and the preset open ground surface and the elevation angle between the source center and the actual complex terrain; The specific calculation method of the slope aspect of the terrain unit is as follows: The outer normal vector of the plane formed by the three points is calculated based on the coordinates of the three elevation points constituting the terrain unit, and the angle between the vector and the direction of the terrain unit pointing to the source center is calculated based on the coordinates of the source center of the terrain unit. The cosine value of the angle is used as the slope direction of the terrain unit. The visibility of the terrain unit is specifically the difference between the altitude angle α formed by the source center relative to each terrain unit in the corresponding terrain sample and the maximum shielding angle formed by other terrains on the direction pointing to the source center to the terrain unit; The terrain openness of the terrain unit includes the terrain openness in 16 directions. The specific calculation method of the terrain openness (Sky View Factor, SVF) is: a. Starting from the direction of point P pointing to the source center, the corresponding azimuth angle φ0 = 0, and moving in a clockwise direction with a step length of Δφ = π / 8, the corresponding 16 azimuth angles are: f i =φ0+iΔφ,i=0,1,2,…,15 b. Calculate the terrain unit in all directions φ i Terrain openness on: SVF i =1-sinθ i Where θ i Point P is in the direction φ i The maximum elevation angle within a certain distance L, that is, the azimuth φ i The terrain on the point P causes the maximum shielding angle; take point P as the starting point, along the azimuth φ i Draw a straight line L i , starting from point P along the straight line L i According to the step length ΔL, the elevation angles of other grid points to point P are judged in turn. ΔL can be taken as a value smaller than the grid spacing. i As the distance increases in step size ΔL, the elevation Z of any point j ij for: Z ij =Z(x p +jΔL x ,y p +jΔL y ),j=1,2,…,N Among them, Z(·,·) is in the direction φ i The terrain elevation of a point on the top, ΔL x , ΔL y is the coordinate increment in the x and y directions, ΔL x =ΔL sin(φ i ), ΔL y =ΔL cos(φ i ), N = L / ΔL is the maximum number of search steps, the shielding angle θ of the point to the terrain unit P ij for: Azimuth φ i The terrain above causes the maximum shielding angle θ to point P i for: i i =max(θ ij ), j=1,2,…,N.

6. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 5 is characterized by: In step 7), the ensemble learning algorithm is the gradient boosting tree algorithm (GBDT) based on the Python open source machine learning library Scikit-Learn. The hyperparameters required by the GBDT algorithm are determined by dividing the training sample set through K-fold cross-validation and adopting the parameter grid search method.

7. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 6 is characterized in that: After step 8), the method further includes: Step 9) obtaining a verification terrain sample, performing the interpolation calculation of step 3) on the verification terrain sample, obtaining the neutron radiation dose of each terrain unit of the verification terrain sample under the preset open surface, and multiplying the result by the radiation dose ratio obtained in step 8) to obtain the neutron radiation dose field prediction result of the verification terrain sample; Step 10) Compare the neutron radiation dose field prediction results of the verification terrain sample with its MC simulation results to verify the actual prediction performance of the neutron radiation dose rapid prediction model under the complex terrain conditions.

8. The method for rapid prediction of neutron radiation dose field under complex terrain conditions according to claim 7 is characterized by: In step 10), the actual prediction performance is tested by comparing the model prediction results with the MC simulation results. The performance evaluation index used is the determination coefficient R 2 and the mean relative percentage error MAPE, defined as follows: Among them, y i ′ is the predicted value, y i is the actual value of the validation sample, is the average value of the actual value of the validation sample, and n is the total number of samples.

Images