Source item inversion optimization method and system of nuclide with long half-life period

Abstract

The invention discloses a source item inversion optimization method and system for nuclide with a long half-life period. The method comprises the following steps: determining a nuclear accident source item and meteorological parameters influencing nuclear accident source item inversion; based on the contribution degree of multiple nuclides to the environmental dose rate, adjusting the release rate change gradient of long-life nuclides with low contribution degree to the environmental dose rate so as to enhance the recognition capability of the long-life nuclides in dose rate inversion; simulating a nuclear accident scene, and generating a training data set and a test set required by the neural network model; performing data preprocessing on the meteorological parameters, the environmental dose rate and the nuclide release rate; determining a neural network model with time sequence analysis capability, and building a nuclear accident source item inversion model; determining a hyper-parameter optimization method of the neural network model, and optimizing hyper-parameters of the neural network; and testing the nuclear accident source item inversion model by using the test set to obtain an optimal nuclear accident source item inversion model. According to the method, the inversion error of the low-contribution nuclide is remarkably reduced, and higher robustness and practicability are achieved.

Description

Technical Field

[0001] This invention belongs to the field of nuclear application technology, and specifically relates to a source term inversion optimization method and system for long half-life nuclides. Background Technology

[0002] On March 11, 2011, a powerful earthquake and tsunami struck the Japanese coast, causing a major accident at the Fukushima Daiichi Nuclear Power Plant and releasing a large amount of radionuclides into the environment. These radionuclides decay during atmospheric diffusion and surface deposition, posing a serious threat to the health of local residents. Therefore, rapid and accurate determination of the source term (i.e., the release rate) of radionuclides is crucial to assist nuclear emergency decision-making bodies in developing emergency plans.

[0003] Generally, there are two methods for source term inversion: one is based on monitoring instrument data within the nuclear power plant, and the other is based on environmental monitoring data outside the plant. However, in severe nuclear accidents, monitoring instruments within the nuclear power plant are often destroyed, making it impossible to obtain operational data. Therefore, increasing research focuses on source term inversion, which predicts the types and release rates of radionuclides from environmental monitoring data outside the nuclear power plant. Currently, many studies have conducted source term inversion based on meteorological parameters and the concentration of radionuclides in the air. However, the concentration of radionuclides in the air is obtained by analyzing environmental samples, representing an average over a period of time, and cannot be obtained in real time. In contrast, the environmental dose rate, which can be measured in real time, is more suitable for source term inversion. Source term inversion methods based on environmental dose rate include source-acceptor equations, Kalman filtering, least squares methods, and data assimilation; however, the accuracy of these methods largely depends on the quality and reliability of the prior information of the source term.

[0004] Artificial neural networks (ANNs) are widely used in environmental radioactivity research due to their adaptive learning capabilities and the fact that they do not require prior information about source terms. Currently, research teams have successfully applied ANNs to the source term inversion of nuclear accidents involving multiple nuclides, demonstrating good results for some nuclides. However, when performing source term inversion based on environmental dose rates for multiple nuclides, the environmental dose rate is contributed by multiple nuclides, and the contribution of different nuclides varies. This may prevent the artificial neural network from recognizing nuclides with longer half-lives, thus making it difficult to obtain satisfactory source term inversion results. Summary of the Invention

[0005] The purpose of this invention is to provide a source term inversion optimization method and system for long half-life nuclides. By adjusting the release rate variation gradient of long half-life nuclides, the difference between them and other nuclides in the environmental dose rate is improved, thereby effectively improving their prediction accuracy in the nuclear accident source term inversion process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a source term inversion optimization method for long half-life nuclides.

[0008] A source term inversion optimization method for long half-life nuclides includes the following steps:

[0009] (1) Determine the nuclear accident source term and the meteorological parameters that affect the inversion of the nuclear accident source term;

[0010] (2) Based on the contribution of multiple nuclides to the environmental dose rate, adjust the release rate change gradient of long-lived nuclides with low contribution to the environmental dose rate to enhance their identification ability in dose rate inversion.

[0011] (3) Simulate nuclear accident scenarios to generate the training and test datasets required for the neural network model;

[0012] (4) Data preprocessing of meteorological parameters, environmental dose rate, and radionuclide release rate;

[0013] (5) Determine a neural network model with time series analysis capabilities and build a nuclear accident source term inversion model;

[0014] (6) Determine the hyperparameter optimization method for the neural network model and optimize the hyperparameters of the neural network;

[0015] (7) Test the nuclear accident source term inversion model using the test set to obtain the optimal nuclear accident source term inversion model.

[0016] Furthermore, in step (1), four fission products in the nuclear reactor, Kr-88, Sr-91, Te-132, and Cs-137, are identified as nuclear accident source terms, and wind speed, wind direction, atmospheric stability, mixing layer height, and precipitation are identified as meteorological parameters.

[0017] Furthermore, in step (2), because the environmental dose rate conversion factors of the four radionuclides are different, even if the release rate is the same, their contributions to the environmental dose rate will differ; the formula for calculating the environmental dose rate of radionuclides is:

[0018] H(t)=X(t)×DRF (1)

[0019] In the formula, H(t) represents the environmental dose rate at time t, X(t) represents the concentration of radionuclides at time t, and DRF represents the environmental dose rate conversion factor.

[0020] In air, the formula for calculating the air immersion dose rate conversion factor is:

[0021]

[0022] In the formula, K is a constant, and E γ For the energy of a monoenergetic ambient photon, ρ a air density, For monoenergetic photons E γ The ratio between the organ k and the organ k;

[0023] On the ground, the formula for calculating the ground deposition dose rate conversion factor is:

[0024]

[0025] In the formula, z represents the vertical distance between the monitoring point and the ground. represents the air specific absorption fraction, and r represents the horizontal distance between the monitoring point and the ground.

[0026] From formulas (1)-(3), it can be seen that under the same atmospheric environment and human tissue conditions, ignoring the effects of dry deposition, wet deposition and decay of radionuclides, the environmental dose rate is approximately proportional to the release rate and dose rate conversion factor of radionuclides. Due to the long half-life and small dose rate conversion factor of Cs-137, its contribution to the environmental dose rate is relatively low, resulting in insufficient source term inversion accuracy. Therefore, in order to improve the identifiability of Cs-137, a gradient enhancement strategy based on half-life characteristics is applied to its release rate in some samples.

[0027] For the data used to generate dose rate gradient changes, the release rates of the original three nuclides, Kr-88, Sr-91, and Te-132, are kept unchanged. Only Cs-137 is subjected to a release rate gradient change based on its half-life. The specific steps are as follows: First, based on the half-lives of the four nuclides, the "half-life ratio weight" between Cs-137 and the other nuclides is calculated. Using this weight, a proportional factor sequence of Cs-137 gradient changes is constructed (e.g., 5 levels of fluctuation, each level differing by a certain proportion, forming 10 gradient levels). Finally, the Cs-137 gradient release rate value is applied to each group of samples to form a new training enhancement subset, enhancing the model's inversion ability when facing low dose rate nuclides.

[0028] The formula for calculating the gradient release rate is as follows:

[0029]

[0030] In the formula, R0 represents the initial release rate, α represents the gradient growth factor, m represents the sequence number of the different gradient groups generated, and T represents the gradient value. 1 / 2 This represents the half-life of a nuclide, and n represents the number of nuclides.

[0031] Furthermore, in step (3), the International Radiation Consequences Assessment System is used to simulate a nuclear accident scenario; during the 10-hour simulation, environmental dose rate data is collected every 0.5 hours to form a time series data of 20 time steps; a total of 30,000 sets of data are generated, of which 20,000 sets of data are directly used for training the neural network, 1,000 sets of data are used to generate the release rate gradient change for training, and the remaining 9,000 sets of data are used for testing the neural network.

[0032] Furthermore, in step (4), the meteorological parameters, environmental dose rate, and nuclide release rate are normalized and mapped to the interval [0,1]. The meteorological parameters and environmental dose rate are determined as input variables of the neural network, and the release rates of the four radionuclides are used as output variables of the neural network.

[0033] Furthermore, in step (5), an inversion model of nuclear accident source terms is constructed using LSTM, Transformer, and fully connected neural network; the environmental dose rate is used as the input variable of LSTM, the output variable of LSTM is used as the input variable of Transformer, the output variable of Transformer is concatenated with meteorological parameters and used as the input variable of fully connected neural network, and finally the release rate of nuclides is output through fully connected neural network; since there are 20 time steps of environmental dose rate and 4 kinds of radionuclides, the input layer of LSTM is set to 20 nodes, and the output layer of fully connected neural network is set to 4 nodes.

[0034] Furthermore, in step (6), the Optuna optimization method is used to optimize the learning rate, the number of hidden layer nodes in the LSTM, the number and size of attention heads in the Transformer, and the number of hidden layer nodes in the fully connected neural network.

[0035] Furthermore, in step (7), the results of the nuclear accident source term inversion model are tested using the mean absolute percentage error as a metric to obtain the optimal nuclear accident source term inversion model.

[0036] Secondly, the present invention provides a source term inversion optimization system for long half-life nuclides, used to execute the source term inversion optimization method for long half-life nuclides.

[0037] A source term inversion optimization system for long half-life nuclides, comprising:

[0038] Unit for acquiring environmental dose rate and meteorological parameters;

[0039] The input data format processing unit is configured to normalize the input data;

[0040] The nuclear accident source term inversion unit is configured to estimate the release rate of radionuclides using input data.

[0041] Thirdly, the present invention provides an environmental monitoring device for a source term inversion optimization system for the long half-life nuclide.

[0042] An environmental monitoring device is used to perform the source term inversion optimization method for the long half-life nuclide, comprising:

[0043] An environmental dose rate monitor is used to collect real-time time-series data of the environmental dose rate in the area affected by a nuclear accident.

[0044] The meteorological parameter monitoring instrument is used to simultaneously collect time-series data of meteorological parameters such as wind speed, wind direction, atmospheric stability, mixing layer height, and precipitation in the area.

[0045] Fourthly, the present invention provides a processing apparatus for deploying a source term inversion optimization system for the long half-life nuclide.

[0046] A processing apparatus, comprising:

[0047] Processor and memory;

[0048] The memory stores an optimized nuclear accident source term inversion model obtained by the method.

[0049] The processor is configured to:

[0050] Receive environmental dose rate and meteorological parameters from environmental monitoring equipment;

[0051] The input data format processing unit is invoked to perform normalization processing on the environmental dose rate and meteorological parameters;

[0052] The nuclear accident source term inversion unit is invoked to perform nuclear accident source term inversion and output the nuclide release rate.

[0053] Beneficial effects:

[0054] 1) This invention uses real-time measured environmental dose rate as input data for the nuclear accident source term inversion model, which helps to achieve faster and more accurate nuclear accident source term inversion, thereby improving the efficiency of nuclear emergency response and prevention and control measures.

[0055] 2) By adjusting the release rate of long half-life nuclides, this invention effectively improves the problem of low accuracy in the inversion of individual nuclide source terms in multi-nucleon source terms.

[0056] 3) This invention employs an artificial neural network method, effectively avoiding the problem of low inversion accuracy caused by unreliable prior information in traditional methods. Furthermore, this invention uses the Optuna optimization algorithm to optimize the structure of the neural network model, reducing computational load and accelerating computation speed. This not only saves computational resources but also enables the model to better adapt to different nuclear accident scenarios, further improving the accuracy and robustness of source term inversion. Attached Figure Description

[0057] Figure 1 Flowchart for the design of a nuclear accident source term inversion model;

[0058] Figure 2 This is a structural diagram of the LSTM-Transformer neural network model;

[0059] Figure 3 Scatter plots of predicted values ​​for four nuclides; where: (a) is the original data, and (b) is the data after Cs-137 release rate gradient enhancement processing;

[0060] Figure 4 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0061] The present invention will now be further described with reference to the accompanying drawings.

[0062] The present invention provides a source term inversion optimization method for long half-life nuclides, comprising the following steps:

[0063] (1) Four fission products in the nuclear reactor, Kr-88, Sr-91, Te-132 and Cs-137, were identified as nuclear accident source terms, and wind speed, wind direction, atmospheric stability, mixing layer height and precipitation were identified as meteorological parameters.

[0064] (2) Because the environmental dose rate conversion factors of the four radionuclides are different, even if the release rate is the same, their contributions to the environmental dose rate will differ; the formula for calculating the environmental dose rate of radionuclides is:

[0065] H(t)=X(t)×DRF (1)

[0066] In the formula, H(t) represents the environmental dose rate at time t, X(t) represents the concentration of radionuclides at time t, and DRF represents the environmental dose rate conversion factor.

[0067] In air, the formula for calculating the air immersion dose rate conversion factor is:

[0068]

[0069] In the formula, K is a constant, and E γFor the energy of a monoenergetic ambient photon, ρ a air density, For monoenergetic photons E γ The ratio between the organ k and the organ k;

[0070] On the ground, the formula for calculating the ground deposition dose rate conversion factor is:

[0071]

[0072] In the formula, z represents the vertical distance between the monitoring point and the ground. represents the air specific absorption fraction, and r represents the horizontal distance between the monitoring point and the ground.

[0073] From formulas (1)-(3), it can be seen that under the same atmospheric environment and human tissue conditions, ignoring the effects of dry deposition, wet deposition and decay of radionuclides, the environmental dose rate is approximately proportional to the release rate and dose rate conversion factor of radionuclides. Due to the long half-life and small dose rate conversion factor of Cs-137, its contribution to the environmental dose rate is relatively low, resulting in insufficient source term inversion accuracy. Therefore, in order to improve the identifiability of Cs-137, a gradient enhancement strategy based on half-life characteristics is applied to its release rate in some samples.

[0074] For the data used to generate dose rate gradient changes, the release rates of the original three nuclides, Kr-88, Sr-91, and Te-132, are kept unchanged. Only Cs-137 is subjected to a release rate gradient change based on its half-life. The specific steps are as follows: First, based on the half-lives of the four nuclides, the "half-life ratio weight" between Cs-137 and the other nuclides is calculated. Using this weight, a proportional factor sequence of Cs-137 gradient changes is constructed (e.g., 5 levels of fluctuation, each level differing by a certain proportion, forming 10 gradient levels). Finally, the Cs-137 gradient release rate value is applied to each group of samples to form a new training enhancement subset, enhancing the model's inversion ability when facing low dose rate nuclides.

[0075] The formula for calculating the gradient release rate is as follows:

[0076]

[0077] In the formula, R0 represents the initial release rate, α represents the gradient growth factor, m represents the sequence number of the different gradient groups generated, and T represents the gradient value. 1 / 2 This represents the half-life of a nuclide, and n represents the number of nuclides.

[0078] (3) The international radiation consequences assessment system was used to simulate nuclear accident scenarios. During the 10-hour simulation, environmental dose rate data was collected every 0.5 hours to form a time series data of 20 time steps. A total of 30,000 sets of data were generated, of which 20,000 sets of data were directly used for training the neural network, 1,000 sets of data were used to generate the release rate gradient change for training, and the remaining 9,000 sets of data were used for testing the neural network.

[0079] (4) Normalize the meteorological parameters, environmental dose rate, and nuclide release rate to map the values ​​to the interval [0,1]; determine the meteorological parameters and environmental dose rate as input variables of the neural network, and the release rates of the four radionuclides as output variables of the neural network.

[0080] (5) A nuclear accident source term inversion model was constructed using a Long Short Term Memory (LSTM), Transformer, and fully connected neural network. The environmental dose rate was used as the input variable of the LSTM, the output variable of the LSTM was used as the input variable of the Transformer, the output variable of the Transformer was concatenated with meteorological parameters and used as the input variable of the fully connected neural network, and finally the release rate of the nuclide was output through the fully connected neural network. Since the environmental dose rate and the release rate of the four radionuclides are contained in 20 time steps, the input layer of the LSTM was set to 20 nodes and the output layer of the fully connected neural network was set to 4 nodes.

[0081] (6) The Optuna optimization method was used to optimize the learning rate, the number of hidden layer nodes of LSTM, the number and size of attention heads of Transformer, and the number of hidden layer nodes of fully connected neural network.

[0082] (7) The nuclear accident source term inversion model is tested using a test set. The mean absolute percentage error is used as a metric to test the results of the nuclear accident source term inversion model in order to obtain the best nuclear accident source term inversion model.

[0083] The present invention will be further described below with reference to the embodiments.

[0084] Example 1:

[0085] like Figure 1 As shown, the source term inversion optimization method for nuclides with low environmental dose rate contribution provided in this embodiment includes the following steps:

[0086] (1) During a nuclear accident, a large number of long-lived and short-lived radionuclides are usually released. Short-lived nuclides have a significant impact on the environment and public health during atmospheric diffusion, deposition, and decay; while long-lived nuclides, although released at higher intensities, have lower dose rate conversion factors, resulting in weaker performance in dose rate monitoring data and thus affecting the accuracy of inversion. To balance the source term inversion capabilities of different nuclides, four typical nuclides (Kr-88, Sr-91, Te-132, and Cs-137) were selected as inversion targets. Among them, long-lived nuclides, represented by Cs-137, have a lower dose rate contribution, which makes their inversion accuracy more susceptible to influence.

[0087] Meteorological parameters affecting the diffusion of radionuclides were determined based on the Gaussian plume atmospheric diffusion model. The principle of the Gaussian plume atmospheric diffusion model is shown in the following formula:

[0088]

[0089] In the formula, X a,i (x) represents the annual average air pollution concentration of near-ground air at a distance x within sector i; Q a Indicates the average annual release rate of a nuclide; σ z,j The vertical diffusion coefficient represents the stability of type j; x represents the distance from the center of the chimney; u j,k h represents the average wind speed at the effective release height for stability class j and wind speed level k; i,k P represents the effective release height at stability level j and wind speed level k; i,j,k This represents the joint frequency within sector i, at stability class j and wind speed level k.

[0090] Meteorological parameters affecting the diffusion of radionuclides in the air include: release altitude, wind speed, wind direction, and atmospheric stability. Release altitude directly affects the maximum diffusion range and maximum wind speed of the nuclide; wind speed directly affects the diffusion rate of the nuclide; wind direction directly affects the location of monitoring points; and atmospheric stability directly affects the vertical movement of air and the intensity of turbulence.

[0091] In many cases, atmospheric temperature stratification exhibits a decreasing gradient at the bottom and an inversion at the top, with a distinct inversion layer existing several hundred meters to 1-2 kilometers above the ground. At this point, the upward diffusion of radioactive nuclides transported over long distances is inhibited after reaching the bottom of the inversion layer at its vertical extension limit, confining it to a thickness of H between the ground and the bottom of the inversion layer. m Within an unstable layer of atmosphere, H mThis is called the mixing layer height, which directly affects the vertical diffusion range of radionuclides. Additionally, during the diffusion of radionuclides in the atmosphere, wet deposition occurs due to rainwater runoff; therefore, precipitation is also a major meteorological parameter affecting radionuclide diffusion. In summary, the main meteorological parameters affecting the inversion of nuclear accident source terms include: release height, wind speed, wind direction, atmospheric stability, mixing layer height, and precipitation.

[0092] (2) Ignoring the effects of dry deposition, wet deposition and decay of radionuclides, the environmental dose rate is approximately proportional to the release rate and dose rate conversion factor of radionuclides, and inversely proportional to the half-life.

[0093] Table 1 Physical properties of three radionuclides

[0094]

[0095] Therefore, fine-grained gradient enhancement processing of the release rate of Cs-137 is performed during the data generation stage. Formulas (4) and (5) are used to calculate the gradient value of the release rate of Cs-137, and new data is generated based on the gradient value. This can improve its recognition ability and prediction effect in the source term inversion task.

[0096] (3) After determining the nuclear accident source terms and meteorological parameters, the International Radiation Assessment System (InterRAS) was used to simulate the nuclear accident scenario. By inputting the source terms and meteorological parameters into InterRAS, the environmental dose rate at 5 monitoring points was continuously obtained over 10 hours, generating a total of 30,000 sets of data. Of these, 20,000 sets of data were directly used for training the neural network, 1,000 sets of data were used to generate the release rate gradient change and then used for training, and the remaining 9,000 sets of data were used for testing the neural network.

[0097] (4) The environmental dose rate, meteorological parameters, and nuclide release rates are normalized and mapped to the interval [0,1]. Furthermore, the environmental dose rate needs to be extended to 20 time steps to highlight the temporal sequence. A set of example environmental dose rate data is shown in Table 2. The first row represents the environmental dose rate at the first time step, containing one environmental dose rate; the second row represents the environmental dose rate at the second time step, containing two environmental dose rates; and the tenth row represents the environmental dose rate at ten time steps, containing ten environmental dose rates. The environmental dose rate and meteorological parameters are used as input variables to the neural network, while the release rates of the four nuclides are used as output variables.

[0098] Table 2. Example of normalized environmental dose rate data (data from 10 time steps)

[0099]

[0100] (5) Build the LSTM-Transformer model, such as Figure 2 As shown, the environmental dose rate at 20 time steps is used as the input variable of the LSTM. After feature extraction by the LSTM, the environmental dose rate is input into the Transformer. After further feature extraction by the Transformer, it is concatenated with release height, wind speed, wind direction, atmospheric stability, mixing layer height, and precipitation, and then input into a fully connected neural network. Finally, the fully connected neural network outputs the release rates of the four nuclides. The input nodes of the LSTM are set to 20, and the output nodes of the fully connected layer are set to 4.

[0101] (6) The Optuna optimization algorithm is used to optimize the learning rate, the number of hidden layer nodes in the LSTM, the number and size of attention heads in the Transformer, and the number of hidden layer nodes in the fully connected neural network. Optuna utilizes Bayesian optimization and sample parallelization to minimize the validation error of the neural network, which can improve the accuracy and convergence speed of the model. In this embodiment, during the optimization of hyperparameters using Optuna, the loss value of the test dataset is defined as the objective function of the model to be optimized.

[0102] (7) After adjusting the hyperparameters of the neural network, the model accuracy was verified using a test dataset. The mean absolute percentage error of the four nuclides at the last time step is shown in Table 2. Te-132, due to its short half-life and high dose conversion factor, contributes the most to the gamma dose rate, thus exhibiting the smallest relative error. In contrast, Cs-137, as a long-lived nuclide, has a relatively small dose rate conversion factor, resulting in a weaker contribution to the environmental dose rate even under the same release rate conditions, leading to lower inversion accuracy in the original model. In this invention, a release rate gradient enhancement strategy based on half-life is introduced to finely adjust the release rate of Cs-137 during the sample generation stage, improving the model's feature learning ability. The results show that the mean inversion error of Cs-137 decreases significantly, indicating that the enhancement strategy effectively improves the model's ability to identify and invert Cs-137.

[0103] Table 3 shows the mean absolute percentage error of the four nuclides at the last time step.

[0104]

[0105] like Figure 3As shown, the predicted values ​​for Kr-88 and Te-132 are very close to the observed values. Although the predicted values ​​for Cs-137 have a wider distribution, the scatter plot density shows that most of the predicted values ​​are concentrated near the observed values. This indicates that the LSTM-Transformer model has high accuracy in retrieving the source terms of the four nuclides. After introducing a gradient enhancement strategy based on half-life for the release rate of Cs-137, the scatter plot distribution of its predicted values ​​significantly approaches the true values, with most predicted values ​​clustered near the observed values.

[0106] Example 2:

[0107] Example 1 provides a source term inversion optimization method for nuclides with low environmental dose rate contribution. Correspondingly, this example provides a source term inversion optimization system for nuclides with low environmental dose rate contribution. Specifically, the source term inversion optimization system for nuclides with low environmental dose rate contribution provided in this example includes:

[0108] (1) Input data for the nuclear accident source term inversion model were obtained through environmental dose monitors and meteorological parameter monitors;

[0109] (2) Normalize the environmental dose rate data and meteorological parameters;

[0110] (3) Input environmental dose rate data and meteorological parameters into the nuclear accident source term inversion model to perform source term inversion for four types of radioactive nuclear accidents.

[0111] like Figure 4 As shown, the source term inversion optimization system for long-half-life nuclides includes an environmental dose rate monitor, a meteorological parameter monitor, a processor, a memory, a communication interface, and a bus. The environmental dose rate monitor and the meteorological parameter monitor are connected to the communication interface to acquire input data for the nuclear accident source term inversion model; the processor, memory, and communication interface are connected via the bus to complete communication between them.

[0112] This invention is applicable to rapid and accurate assessment in nuclear emergency response. Unlike traditional inversion methods that rely on prior information, this invention uses artificial neural networks to predict multi-nucleus source terms without prior knowledge, specifically addressing the problem of low inversion accuracy caused by the low dose contribution of long-half-life nuclides. During the data generation stage, while keeping the release rates of other nuclides constant, a gradient-based release rate variation based on half-life structure is applied to long-half-life nuclides, effectively enhancing their performance in monitoring data and improving the model's ability to identify them. The constructed LSTM-Transformer model possesses excellent time-series analysis and feature extraction capabilities, which can improve the accuracy and generalization ability of the inversion model. Results show that this invention significantly reduces the inversion error of low-contribution nuclides, exhibiting stronger robustness and practicality. The constructed system can acquire environmental dose rate and meteorological data in real time, automatically complete data processing and nuclide release rate estimation, providing efficient and intelligent decision support for nuclear accident emergency response.

[0113] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A source term inversion optimization method for long half-life nuclides, characterized in that: Includes the following steps: (1) Determine the nuclear accident source term and the meteorological parameters that affect the inversion of the nuclear accident source term; (2) Based on the contribution of multiple nuclides to the environmental dose rate, adjust the release rate change gradient of long-lived nuclides with low contribution to the environmental dose rate to enhance their identification ability in dose rate inversion. (3) Simulate nuclear accident scenarios to generate the training and test datasets required for the neural network model; (4) Data preprocessing of meteorological parameters, environmental dose rate, and radionuclide release rate; (5) Determine a neural network model with time series analysis capabilities and build a nuclear accident source term inversion model; (6) Determine the hyperparameter optimization method for the neural network model and optimize the hyperparameters of the neural network; (7) Test the nuclear accident source term inversion model using the test set to obtain the optimal nuclear accident source term inversion model.

2. The source term inversion optimization method for long half-life nuclides according to claim 1, characterized in that: In step (1), four fission products in the nuclear reactor, Kr-88, Sr-91, Te-132 and Cs-137, are identified as nuclear accident source terms, and wind speed, wind direction, atmospheric stability, mixing layer height and precipitation are identified as meteorological parameters.

3. The source term inversion optimization method for long half-life nuclides according to claim 2, characterized in that: In step (2), the formula for calculating the environmental dose rate of the radionuclide is: H(t)=X(t)×DRF (1) In the formula, H(t) represents the environmental dose rate at time t, X(t) represents the concentration of radionuclides at time t, and DRF represents the environmental dose rate conversion factor. In air, the formula for calculating the air immersion dose rate conversion factor is: In the formula, K is a constant, and E γ For the energy of a monoenergetic environmental photon, ρ a air density, For monoenergetic photons E γ The ratio between the organ k and the organ k; On the ground, the formula for calculating the ground deposition dose rate conversion factor is: In the formula, z represents the vertical distance between the monitoring point and the ground. represents the air specific absorption fraction, and r represents the horizontal distance between the monitoring point and the ground. From formulas (1)-(3), it can be seen that under the same atmospheric environment and human tissue conditions, ignoring the effects of dry deposition, wet deposition and decay of radionuclides, the environmental dose rate is approximately proportional to the release rate and dose rate conversion factor of radionuclides. Due to the long half-life and small dose rate conversion factor of Cs-137, its contribution to the environmental dose rate is relatively low, resulting in insufficient source term inversion accuracy. Therefore, in order to improve the identifiability of Cs-137, a gradient enhancement strategy based on half-life characteristics was applied to its release rate in some samples. For the data used to generate dose rate gradient changes, the release rates of the original three nuclides, Kr-88, Sr-91, and Te-132, are kept unchanged. Only the release rate gradient change based on the nuclide half-life is applied to Cs-137. The specific steps are as follows: First, based on the half-lives of the four nuclides, the "half-life proportional weight" between Cs-137 and the other nuclides is calculated. Using this weight, a proportional factor sequence of Cs-137 gradient changes is constructed. Finally, the Cs-137 gradient release rate value is applied to each group of samples to form a new training enhancement subset, which enhances the model's inversion ability when facing low dose rate nuclides. The formula for calculating the gradient release rate is as follows: In the formula, R0 represents the initial release rate, α represents the gradient growth factor, m represents the sequence number of the different gradient groups generated, and T represents the gradient value. 1 / 2 This represents the half-life of a nuclide, and n represents the number of nuclides.

4. The source term inversion optimization method for long half-life nuclides according to claim 3, characterized in that: In step (3), the International Radiation Consequences Assessment System is used to simulate a nuclear accident scenario. During the 10-hour simulation, environmental dose rate data is collected every 0.5 hours to form a time series data of 20 time steps. A total of 30,000 sets of data are generated, of which 20,000 sets of data are directly used for training the neural network, 1,000 sets of data are used to generate the release rate gradient change and then used for training, and the remaining 9,000 sets of data are used for testing the neural network.

5. The source term inversion optimization method for long half-life nuclides according to claim 4, characterized in that: In step (4), the meteorological parameters, environmental dose rate, and radionuclide release rate are normalized and mapped to the interval [0,1]. The meteorological parameters and environmental dose rate are determined as input variables of the neural network, and the release rates of the four radionuclides are used as output variables of the neural network.

6. The source term inversion optimization method for long half-life nuclides according to claim 5, characterized in that: In step (5), an inversion model of nuclear accident source terms is constructed using LSTM, Transformer, and fully connected neural network. The environmental dose rate is used as the input variable of LSTM, the output variable of LSTM is used as the input variable of Transformer, the output variable of Transformer is concatenated with meteorological parameters and used as the input variable of fully connected neural network, and finally the release rate of nuclides is output through fully connected neural network. Since there are 20 time steps of environmental dose rate and 4 release rates of radionuclides, the input layer of LSTM is set to 20 nodes and the output layer of fully connected neural network is set to 4 nodes. In step (6), the Optuna optimization method is used to optimize the learning rate, the number of hidden layer nodes of LSTM, the number and size of attention heads of Transformer, and the number of hidden layer nodes of fully connected neural network.

7. The source term inversion optimization method for long half-life nuclides according to claim 6, characterized in that: In step (7), the results of the nuclear accident source term inversion model are tested using the mean absolute percentage error as a metric to obtain the optimal nuclear accident source term inversion model.

8. A system for performing the long half-life nuclide source term inversion optimization method according to any one of claims 1-7, characterized in that: include: Unit for acquiring environmental dose rate and meteorological parameters; The input data format processing unit is configured to normalize the input data; The nuclear accident source term inversion unit is configured to estimate the release rate of radionuclides using input data.

9. An environmental monitoring device for the system described in claim 8, characterized in that: include: An environmental dose rate monitor is used to collect real-time time-series data of the environmental dose rate in the area affected by a nuclear accident. The meteorological parameter monitoring instrument is used to simultaneously collect time-series data of meteorological parameters such as wind speed, wind direction, atmospheric stability, mixing layer height, and precipitation in the area.

10. A processing device for deploying the system of claim 8, characterized in that: include: Processor and memory; The memory stores an optimized nuclear accident source term inversion model obtained by the method described in any one of claims 1-7; The processor is configured to: Receive environmental dose rate and meteorological parameters from environmental monitoring equipment; The input data format processing unit is invoked to perform normalization processing on the environmental dose rate and meteorological parameters; The nuclear accident source term inversion unit is invoked to perform nuclear accident source term inversion and output the nuclide release rate.

Images