A method, system and readable medium for training and predicting a nuclear diffusion model

Abstract

The present application relates to the technical field of nuclide diffusion prediction, and particularly relates to a nuclide diffusion model training and prediction method, system and readable medium.A nuclide diffusion prediction method is provided, which first converts the collected time series data into a Markov transition field image, then reconstructs the Markov transition field image through a preprocessing module to obtain reconstructed data that can highlight more feature information; and then inputs the reconstructed data into a DDPM network for processing to generate a radioactive nuclide diffusion image in a prediction time period; the radioactive nuclide diffusion image output by the DDPM network is actually a Markov transition field image corresponding to the predicted value of the time series data in the prediction time period, which can be obtained through Markov inverse coding of the radioactive nuclide diffusion image.In the present application, the nuclide diffusion model is based on the Markov transition field image for prediction, the characteristics of the input data are considered more comprehensively, and the prediction result is more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclide diffusion prediction, in particular to a nuclide diffusion model training and prediction method, system and readable medium. BACKGROUND

[0002] Radioactive pollution causes irreversible harm to humans and the natural environment. Therefore, the prediction of radionuclide diffusion is of great importance and helps to guide nuclear emergency decision-making scientifically.

[0003] The prediction of radionuclide diffusion involves multiple disciplines such as nuclear physics, atmospheric environmental science and computer science, and has developed mature theoretical basis and application mode. Traditional radionuclide diffusion prediction methods include atmospheric tracer experiments, wind tunnel experiments and numerical simulation prediction. Atmospheric tracer experiments refer to releasing tracers in a specific location to obtain the atmospheric flow characteristics and tracer diffusion law in the regional environment through sampling monitoring. Common gas tracers are SF6 gas, which is stable in nature and can be detected by gas chromatography. Wind tunnel experiments are similar in principle to atmospheric tracer experiments, and also use tracers to reflect the atmospheric flow characteristics and diffusion law of the region. Unlike atmospheric tracer experiments, which are carried out in real scenes, wind tunnel experiments are carried out in models scaled according to a certain scale of real scenes. The advantage of wind tunnel experiments is that they can reduce costs while controlling meteorological conditions and underlying surface environment, and can be used for quantitative research to predict diffusion.

[0004] Traditional atmospheric diffusion numerical models mainly include Gaussian plume model, Lagrangian puff model, Euler model and computational fluid dynamics. However, these traditional radionuclide diffusion prediction methods are highly dependent on the source term information of radionuclides, which makes it difficult for traditional radionuclide simulation methods to quickly adapt to unexpected radionuclide diffusion scenarios with uncertain locations, complex terrain and variable weather. SUMMARY

[0005] In order to overcome the defects of the above-mentioned prior art radionuclide diffusion prediction method which is too dependent on the source term information of radionuclides and is difficult to adapt to unexpected radionuclide diffusion scenarios, the present application proposes a radionuclide diffusion model training method which is suitable for any scenario and can accurately predict radionuclide diffusion at high speed.

[0006] The radionuclide diffusion model training method proposed by the present application comprises the following steps:

[0007] SA1, constructing a basic model, a first data set and a second data set;

[0008] The first data set is used for storing the time series data with missing time points; the second data set is used for storing the time series data with complete and labeled radionuclide diffusion images; the time series data contains monitoring objects of a monitoring area collected at multiple continuous monitoring time points; the monitoring objects include radionuclide concentrations;

[0009] The base model comprises a preprocessing module and a DDPM network, an input of the preprocessing module is an input of the radionuclide diffusion model, and an output of the DDPM network is an output of the radionuclide diffusion model; the preprocessing module is used for reconstructing the input Markov transition field image, and the DDPM network generates a Markov transition field image in a predicted time period based on the reconstructed Markov transition field image;

[0010] SA2, the time series data in the first data set and the second data set are all processed into Markov transition field images;

[0011] SA3, the preprocessing module is trained on the first data set until convergence;

[0012] SA4, a second training sample is extracted from the second data set, the time series data of the second training sample is processed for missing, and then converted into a Markov transition field image as a missing sample;

[0013] SA5, the missing sample is processed by the preprocessing module to obtain a blurred image, and the DDPM network processes the blurred image to obtain a predicted radionuclide diffusion image as a predicted value;

[0014] SA6, the radionuclide diffusion image associated with the second training sample is taken as a true value, and the loss is calculated by combining the predicted value and the true value;

[0015] SA7, whether the loss converges is judged; if not, the DDPM network is updated by loss back propagation, and then the step SA4 is returned; if yes, the base model is fixed as the radionuclide diffusion model.

[0016] Preferably, in SA2, the method for processing the time series data into a Markov transition field image is: processing the time series data into time series of each monitoring object, processing the time series of each monitoring object into a Markov encoding image by using a Markov transition matrix, and then weighting and superimposing the Markov encoding images of each monitoring object to obtain a Markov transition field image.

[0017] Preferably, in SA3, the method for training the preprocessing module on the first data set is: the preprocessing module processes the Markov transition field image in the first data set and outputs a reconstructed image as a blurred image, the DDPM generates a radionuclide diffusion image according to the blurred image; the parameters of the preprocessing module are updated according to an optimization objective until the preprocessing module converges; the optimization objective is:

[0018]

[0019] a Markov transition field image representing the input of the pre-processing module, output data of the pre-processing module; x represents the radionuclide diffusion image output by the DDPM; into the DDPM network for guided generation; representing the probability of the base model under the control of the parameter η being and the probability of x being generated, i.e. the pre-processing module input being the output being the probability of the DDPM network outputting x when

[0020] Preferably, the monitoring objects include one or more of atmospheric pressure, wind speed, temperature, and humidity.

[0021] Preferably, the wind speed is the mean value of the X-axis wind speed, the Y-axis wind speed, and the Z-axis wind speed.

[0022] Preferably, the pre-processing module is a beta-VAE module.

[0023] The present application proposes a radionuclide diffusion prediction method using the radionuclide diffusion model training method.

[0024] S1, using the radionuclide diffusion model training method to obtain a radionuclide diffusion model; collecting time series data in a set monitoring area;

[0025] S2, randomly sampling the time series data to form new time series data with missing time points, and converting the new time series data into a Markov transition field image;

[0026] S3, inputting the Markov transition field image into the radionuclide diffusion model; the pre-processing module reconstructs the Markov transition field image; the DDPM network processes the reconstructed data and outputs a radionuclide diffusion image;

[0027] S4, Markov inverse encoding the radionuclide diffusion image to obtain a prediction value of the time series data of the monitoring objects in the monitoring area in the prediction time period, and extracting a radionuclide concentration prediction sequence from the prediction value as the radionuclide diffusion prediction result.

[0028] Preferably, in S2, the method for converting the new time series data into a Markov transition field image is: processing the new time series data into time series of each monitoring object, processing the time series of each monitoring object into a Markov encoding image using a Markov transition matrix, and then weighting and superimposing the Markov encoding images of each monitoring object to obtain the Markov transition field image.

[0029] The application provides a radionuclide diffusion prediction system, which comprises a memory and a processor, the memory stores a computer program, and the processor is connected with the memory and used for executing the computer program to realize the radionuclide diffusion prediction method.

[0030] The application provides a readable medium, which stores a computer program, and the computer program is used for realizing the radionuclide diffusion prediction method when executed.

[0031] The application has the advantages that:

[0032] (1) The radionuclide diffusion prediction method provided by the application firstly converts the collected time series data into a Markov transition field image, then reconstructs the Markov transition field image through a preprocessing module to obtain reconstructed data which can highlight more feature information, and then inputs the reconstructed data into a DDPM network for processing to generate a radionuclide diffusion image in a prediction time period; the radionuclide diffusion image output by the DDPM network is actually a Markov transition field image corresponding to a predicted value of the time series data in the prediction time period, which can be obtained through Markov inverse coding of the radionuclide diffusion image. In the application, the radionuclide diffusion model is based on the Markov transition field image for prediction, the characteristics of the input data are considered more comprehensively, and the prediction result is more accurate.

[0033] (2) In the application, the preprocessing module adopts a beta-VAE module, and an image generated by the beta-VAE module is taken as a prior condition and input into the DDPM network, so that the DDPM can conditionally generate a corresponding radionuclide diffusion Markov transition field image, and the reliability of prediction is further improved.

[0034] (3) In the application, Markov coding images of time series of each monitoring object are firstly obtained, and then a Markov transition field image of multi-feature coupling is obtained through weighted superposition; the input of the radionuclide diffusion model is in the form of a simple and intuitive image, and the monitoring features are deeply described, so that the feature mining in the model processing process is facilitated.

[0035] (4) The training method of the radionuclide diffusion model provided by the application, in the training process, the supervised training of the preprocessing module and the unsupervised training of the DDPM are independent of each other; the overall convergence speed in the model training process is improved. The generation result of the preprocessing module is taken as one of the inputs of the DDPM, and guides the DDPM to generate a final image, so that the training accuracy and efficiency of the DDPM network are further improved. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is a flow chart of the training method of the radionuclide diffusion model;

[0037] Figure 2Fig. 1 is a schematic diagram of a training process of a nuclide diffusion model;

[0038] Figure 3 Fig. 2 is a flowchart of a nuclide diffusion prediction method;

[0039] Figure 4 Fig. 3 is a complex underlying surface geometric model composed of three different vegetations;

[0040] Figure 5 Fig. 4 is a schematic diagram of a monitoring area;

[0041] Figure 6 Fig. 5 is a nuclide concentration change of each node;

[0042] Figure 7 Fig. 6 is a nuclide concentration prediction result of node A1. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0044] DDPM (conditional diffusion model) is a generative model that learns the target data distribution from samples, which consists of two Markov processes: one is a fixed forward process that simulates from real pictures to random Gaussian noise, and the other is a learning-based backward process that generates images with similar features to the original image. The form of the forward process in DDPM can be summarized as follows:

[0045]

[0046] Wherein: c represents the control condition added in DDPM; q(x 1：T |x0, c) represents the joint conditional probability distribution from the initial state x0 to the time step T under the guidance of the condition c, which describes the whole process of gradually adding noise from the original data x0 to x T ; T is the set maximum time step; t is the time step number;

[0047] q(x t |x t-1 , c) represents the conditional probability distribution from the state x t-1 at time step t-1 to the state x t at time step t under the guidance of the condition c; this process usually involves adding random noise to x t-1 to obtain x t ; δ1, δ2,..., δ t..., delta T represents the noise-adding strategy in the forward process; represents a Gaussian distribution with mean and standard deviation delta t I; I represents an identity matrix;

[0048] q(x t | x0, c) represents the conditional probability distribution from the initial state x0 to the state x t at time step t under the guidance of the condition c; represents a Gaussian distribution with mean and standard deviation , and has gamma t = 1-delta t .

[0049] The reverse process of the DDPM can also be parameterized by a Gaussian transition distribution learned through learning, and the specific process is as follows:

[0050]

[0051] Where: p(x 0：T | c) represents the joint probability distribution from the final noise state x T back to the original state x0 under the influence of the condition c, which describes the process of gradually recovering data from noise; p(x T | c) represents the prior probability distribution of the final noise state x T under the condition c;

[0052] p λ (x t-1 | x t , c) represents the conditional probability distribution of the state x t at time step t back to the state x t-1 at time step t-1 under the influence of the condition c, and this process is parameterized by a learnable model, and lambda is a parameter of the DDPM.

[0053] Let q(x T | x0, c) represent the conditional probability distribution from the initial state x0 to x T under the guidance of the condition c; when T is large enough, the distribution q(x T | x0, c) can be approximated as an isotropic Gaussian distribution, and lambda can be trained based on variational inference. However, it can be seen from the generation process of the DDPM in formula 1 and formula 1 that adding an auxiliary condition c in the DDPM can improve the accuracy of sample generation and reduce the instability of the DDPM. Therefore, the present application proposes a DDPM framework that can be coupled with any preprocessing module. The DDPM framework is adopted The optimization objective function of the general pre-processing module can be represented as:

[0054]

[0055] η is the parameter of the overall model formed by connecting the pre-processing module and the DDPM framework before and after; represents the original time series data, that is, the input data of the pre-processing module; x represents the original time series data the image reconstructed by the DDPM after the pre-processing module; represents the prior information, that is, the output data of the pre-processing module, which is also the condition c in formula 1; into the DDPM network for guiding the generation of x; represents the probability of generating x under the control of the model parameter η and . and .

[0056] To obtain the original state x0, replace x with x0 in formula (3), so that the DDPM outputs x0 according to and , and obtains the following inequality according to the evidence lower bound:

[0057]

[0058] where φ is the parameter of the inverse process of the DDPM, represents the probability that the DDPM outputs x0 when the input is and . represents the expectation of , represents the original time series data the data x0 obtained by reconstructing the data sequence x through the pre-processing module and the DDPM 1:T . 0:T represents the joint probability distribution from the final noise state x T back to the original state x0; represents the probability that the DPPM combines the output data of the pre-processing module and the original time series data to obtain the data sequence x 0:T . represents the probability that the pre-processing module generates from , represents the distribution of ; ψ represents the parameter of the encoder in the pre-processing model.

[0059] The inequality (4) shows that in the controllable DDPM training guided by the preprocessing module, the supervised training of the preprocessing module and the unsupervised training of the DDPM are independent of each other; the generation result of the preprocessing module As one of the inputs of the DDPM, the DDPM is guided to generate the final image.

[0060] Based on the theoretical derivation above, the present application adopts a β-VAE module (β-Variational Autoencoder, β-variational autoencoder) as the preprocessing module, and the data input into the preprocessing module The Markov transition field image is used, and the image generated by the preprocessing module is used as a prior condition The Markov transition field image is used, and the image generated by the preprocessing module is used as a prior condition

[0061] In the present application, the whole model composed of the β-VAE module and the DDPM network is denoted as a β-VDRDPF model, the input of the β-VAE module is used as the input of the β-VDRDPF model, the input data of the β-VDRDPF model adopts the Markov transition field image, and the DDPM network generates the radionuclide diffusion image in combination with the Markov transition field image input into the β-VDRDPF model and the output data of the β-VAE module.

[0062] The data processing processes of the β-VAE module, the DDPM model and the Markov transition field will be introduced below.

[0063] The Markov transition field image input into the β-VAE module as the initial state x0 is converted from the collected time series data, which includes: collection time (s); collection point radionuclide I-131 concentration (kBq / m 3 ); collection point atmospheric pressure (Pa); X-axis wind speed (m / s), Y-axis wind speed (m / s), Z-axis wind speed (m / s) and three-dimensional coordinates of the collection point.

[0064] The β-VAE module includes an encoder and a decoder, and the input data x0 is processed by the encoder, and the encoded feature data output by the encoder is weighted to obtain the latent code z. The latent code z does not directly participate in the training of the DDPM and is not output; the latent code z is only used in the training of the β-VAE module, which is specifically shown in the following formulas (5)-(8).

[0065] During the training process of the β-VAE module, an auxiliary signal y is required as input, the β-VAE module processes the auxiliary signal y to generate output data, and the β-VAE module generates a latent code z; the DDPM generates a time series x according to the output data of the β-VAE module 1:T , T is the total number of time steps; in this way, the joint distribution p(x 0：T , y, z) of the β-VDRDPF model can be expressed as:

[0066] p(x 0：T , y, z) = p(z)p θ (y|z)pφ(x 0:T |y, z) (5)

[0067] where θ represents the parameters of the decoder in the β-VAE module, and φ represents the parameters of the inverse process of the DPPM network; p(z) represents the probability that the β-VAE module generates a latent code z; p θ (y|z) represents the probability that the β-VAE module generates a latent code z according to the input data y when the decoder parameters of the β-VAE module are θ; p φ (x 0：T |y, z) represents the probability that the β-VDRDPF model generates time series data x 0：T according to the input data y when the inverse process parameters of the DPPM network are φ and the latent code generated by the β-VAE module is z.

[0068] Since the real joint posterior distribution p(x 1：T , z|y, x0) is difficult to calculate, an approximate proxy distribution q(x 1：T , z|y, x0) is used to approximate it, and q(x 1：T , z|y, x0) can also be described as the following conditional distribution:

[0069] q(x 1：T , z|y, x0) = q ψ (z|y, x0)q(x 1：T |y, z, z0) (6)

[0070] where p(x 1：T , z|y, x0) represents the real joint posterior distribution of the β-VDRDPF model, that is, the probability that the β-VDRDPF model generates a latent code z and time series data x1: T when the input is y and the output of the β-VAE module is x0; ψ represents the parameters of the β-VAE module; q ψ (z|y, x0) represents the probability that the β-VAE module outputs a latent code z when the encoder parameters of the β-VAE module are ψ, the input is y, and the output is x0; q(x 1：T|y,z,x0) means that when the β-VAE module input is y, the output is x0 and the potential code is z, the β-VDRDPF model generates the time series x 1:T probability.

[0071] During the training process, the forward process of DDPM does not perform parameter updates or training, but keeps its parameters fixed. Therefore, the likelihood logarithm of the Markov transition field data used for training can be expressed as:

[0072] lgp(x0,y)=lg∫p(x 0：T ,y,z)dx 1：T dz (7)

[0073] Among them, p(x0, y) represents the probability that the output of the β-VAE module is x0 when the input is y; p(x 0:T , y, z) means that when the β-VAE module takes input y and generates a potential code z, the β-VDRDPF model generates a time series x 0:T probability;

[0074] Since formula (7) is difficult to estimate analytically, the present invention optimizes the lower bound of evidence corresponding to the log-likelihood

[0075]

[0076] Where ψ is the parameter of the encoder in the β-VAE module, θ is the parameter of the decoder in the β-VAE module, and φ is the parameter of the DDPM inverse process; p θ (y|z) represents the probability that the decoder of the β-VAE module generates the input data y according to the latent code z; q ψ (z|y, x0) represents the probability that the encoder of the β-VAE module generates the latent code z when the input is y and the output is x0; Indicates that in the probability distribution q ψ p calculated on (z|y, x0) θ expectations of (y|z);

[0077] q ψ The KL divergence between (z|y, x0) and p(z); p(z) represents the probability of the β-VAE module generating the latent code z; β represents the hyperparameter in the β-VAE module that controls the KL divergence weight and determines the degree of regularization of the β-VAE module;

[0078] p φ (x 0:T |y,z) represents the inverse process of DDPM to generate x based on the input data y under the guidance of the latent code z 0:T The probability of q(x 1:Trepresents the probability of the DDPM generating x in the β-VDRDPF model when the input data of the β-VAE module is y and the output is x0 and the generated latent code is z 1:T . represents the expectation of the ratio of p 1：T (x φ |y, z) to q 0：T (x 1：T |y, z, x0) on q φ (x 0：T |y, z). 1：T represents the expectation calculated on the z distribution of the specified β-VAE module with the input being y and the output being x0.

[0079] In the above formula (8), is the loss term of the β-VAE module, representing the expectation that the latent code z generated by the encoder thereof can reconstruct the conditional signal y; is the loss term of the DDPM, representing the expected ratio between the reverse diffusion process p φ (x 0：T |y, z) and the forward diffusion process q 1：T (x φ |y, z, x0). represents the expectation calculation on the latent variable z output by the VAE encoder network; represents the loss function and contribution of the β-VAE module, represents the loss function and contribution of the DDPM;

[0080] In order to enable the β-VDRDPF model proposed in the present application to converge quickly in the training process, the following design method is used when designing parameters to simplify part of the network parameters, so that formula (8) is rewritten as the following formula (9) and formula (10).

[0081] 1) In order to ensure the mapping relationship between y and x0, it is assumed that y is the Markov transition field image x0 itself, and the reverse diffusion process of the diffusion model is not performed conditioned on y in the training process, but is regarded as p φ (x 0：T |z) in formula (5). At this time, formula (8) can be rewritten as:

[0082]

[0083] where q ψ (z|x0) represents the posterior distribution of the latent variable z generated by the encoder of the β-VAE module according to the input image x0, p θ (x0|z) represents the probability of the input image x0 generated by the decoder of the β-VAE module according to the latent code z; represents the calculation of p ψ (z|x0) on the qθ The expectation of (x0|z) is used to characterize the posterior distribution q of z given x0. ψ The expected reconstruction loss of (z|x0);

[0084] p(x0) represents the probability that the β-VAE module generates x0, and p(z) represents the probability that the β-VAE module generates z; represents the posterior distribution q ψ The KL divergence between (z|x0) and the prior distribution p(z); β represents the hyperparameter that controls the weight of the KL divergence and determines the degree of regularization of the model;

[0085] p φ (x 0:T |z) represents the inverse process of DDPM, which generates x based on the input data y under the guidance of the latent code z. 0:T The probability of gradually recovering the image x from the noise given z 0：T The inverse process distribution of q(x 1:T |z,x0) represents the x generated by DDPM in the β-VDRDPF model when the input data of the β-VAE module is x0 and the generated latent code is z. 1:T The probability that the noise diffusion process x is given z and x0. 1：T The forward process distribution of is q(x 1:T |z,x0) on p φ (x 0:T |z) and q(x 1:T |z,x0) ratio expectation, that is, the reverse diffusion process p φ (x 0：T |z) and the forward diffusion process q(x 1：T |z, x0); Represents the expectation calculated on the z distribution of the input image x0, that is, the expectation calculation of the latent variable z output by the encoder in the β-VAE module given x0.

[0086] 2) When performing the reverse diffusion of the model, the invention does not directly use the potential code z of x0 generated by the β-VAE module as the input of the diffusion model, but reconstructs the potential code z of x0 according to the β-VAE module. Guide the training of the reverse transfer process of the diffusion model, at this time β-VAE reconstructs is a deterministic function of z. In this case, formula (8) can be rewritten as:

[0087]

[0088] represents the output of the β-VAE module when it takes x0 as input; Indicates a given and x0, the noise diffusion process x 1:T The forward process distribution of Indicates a given When , the time series x is gradually recovered from the noise 0：T The inverse process distribution of It represents the expectation of the ratio of the inverse process distribution of DDPM to the forward process distribution; For a given input image x0, Sampling from the distribution to reconstruct the image expectations; Reconstructed image representing the β-VAE module Expected calculation.

[0089] The evidence lower bound expression of formula (10) consists of two parts: the first part is the ELBO (variational lower bound) of the β-VAE module, which measures the reconstruction loss and the KL divergence between the distribution of the latent variable z and the prior distribution; the second part is the loss of DDPM, which measures the loss from the reconstructed image The degree of match between the inverse process and the forward process of restoring the original image x0.

[0090] 3) According to formula (8), it can be seen that the present invention uses a two-step training strategy in the training process. The first step is to optimize Second step optimization The second training phase fixes the parameters θ and ψ at the same time, and freezes the parameters of the β-VAE module in the first step; therefore, its final joint distribution can be defined as:

[0091]

[0092] Where, It means that the β-VDRDPF model is given Generate x under the condition 0：T The conditional probability of is generated by β-VAE Probability, Indicates that DDPM is given The conditional generation of x 0：T The conditional probability of .

[0093] A sensors, data from the A sensors are used to obtain A Markov transfer field images, and the A Markov transfer field images are weightedly superimposed to obtain a final Markov transfer field image.

[0094] Based on the above derivation, a nuclide diffusion model composed of a beta-VAE module and a DDPM network is proposed in the embodiment; the input of the beta-VAE module is the input of the nuclide diffusion model, the output of the beta-VAE module is connected to the input of the DDPM network, and the output of the DDPM network is the output of the nuclide diffusion model.

[0095] The input of the nuclide diffusion model is the Markov transition field image y converted from the time series data collected in the monitoring area, the beta-VAE module encodes and decodes the Markov transition field image, and the decoding data output by the beta-VAE module is connected to the input of the DDPM network. The nuclide diffusion image x0 generated by the DDPM network is the nuclide diffusion prediction result, that is, the Markov transition field image corresponding to the time series data in the prediction time period.

[0096] In the embodiment, the time series data contains monitoring objects collected in the monitoring area at multiple consecutive monitoring time points, and the monitoring objects include the concentration of radioactive nuclides (kBq / m 3 ), atmospheric pressure (Pa), wind speed (m / s), and temperature and humidity. The wind speed adopts the average value of the X-axis wind speed (m / s), Y-axis wind speed (m / s), and Z-axis wind speed (m / s) in the monitoring area.

[0097] The time series data is essentially composed of the time series of each monitoring object, and the method for processing the time series data into a Markov transition field image is as follows: using a Markov transition matrix to process the time series of each monitoring object into a Markov encoding image, and then weighting and superimposing the Markov encoding images of each monitoring object to obtain a Markov transition field image with multiple feature couplings as the input of the nuclide diffusion model.

[0098] When weighting and superimposing the Markov encoding images of each monitoring object, the weight of the concentration of radioactive nuclides is not less than 0.95; and the sum of the weights of temperature and humidity is not greater than 0.01.

[0099] It is worth noting that the number of monitoring time points contained in the input of the nuclide diffusion model is equal to the amount of time point data contained in the output of the nuclide diffusion model; assuming that the time series data composed of monitoring objects collected at 1-M monitoring time points in the monitoring area is converted into a Markov transition field image and input into the nuclide diffusion model, the nuclide diffusion image output by the nuclide diffusion model can be obtained by Markov inverse encoding, and the predicted value of the time series data composed of monitoring objects at M+1 to 2M monitoring time points in the monitoring area can be obtained, and then the predicted time series of the concentration of radioactive nuclides can be extracted from the predicted value.

[0100] Referring to Figure 1 、 Figure 2 , the training method of the nuclide diffusion model proposed in the embodiment includes the following steps:

[0101] SA1, construct a base model, a first data set and a second data set; the base model comprises a preprocessing module and a DDPM network; the preprocessing module specifically adopts a beta-VAE module.

[0102] The preprocessing module comprises an encoder and a decoder; the encoder encodes the input Markov transition field image and generates a latent feature z; the decoder decodes the latent feature z and outputs a reconstructed image; the reconstructed image is processed into a radionuclide diffusion image by the DDPM network and output.

[0103] The first data set is used to store time series data missing part of time points; the second data set is used to store time series data complete and labeled with radionuclide diffusion images;

[0104] SA2, process the time series data in the first data set and the second data set into Markov transition field images;

[0105] SA3, train the preprocessing module on the first data set to convergence; during the training process, the preprocessing module processes the Markov transition field images in the first data set and outputs the reconstructed images as blurred images, and the DDPM generates radionuclide diffusion images according to the blurred images; during the training process, the parameters of the preprocessing module are updated according to the following optimization objective, and the parameters of the DDPM are fixed; the optimization objective is:

[0106]

[0107] The Markov transition field image represents the input of the preprocessing module, The output data of the preprocessing module; x represents the radionuclide diffusion image output by the DDPM; Into the DDPM network for guided generation; The base model under the control of parameters η is generated by And The probability that x is generated, that is, the input of the preprocessing module is The output is The probability that the DDPM network outputs x when the input is

[0108] Specifically, step SA3 comprises the following sub-steps:

[0109] SA31, extract a first training sample from the first data set, and substitute the Markov transition field image of the first training sample into the base model to optimize the parameters of the preprocessing module in combination with the above optimization objective;

[0110] SA32, judge whether the iteration number of the preprocessing module reaches a set value; if yes, complete the pre-training of the preprocessing module and execute step S4; if no, return to step SA31.

[0111] SA4, extracting a second training sample from the second data set, and performing missing processing on time series data of the second training sample to convert the time series data into a Markov transition field image as a missing sample;

[0112] Specifically, in this step, first, the time series data in the second training sample is randomly sampled to form a data sequence with missing time points, and then the data sequence with missing time points is converted into a Markov transition field image as a missing sample;

[0113] SA5, processing the missing sample through the preprocessing module to obtain a blurred image, and processing the blurred image through the DDPM network to obtain a predicted radionuclide diffusion image as a prediction value;

[0114] SA6, taking the radionuclide diffusion image associated with the second training sample as a true value, and calculating a loss by combining the prediction value and the true value;

[0115] SA7, determining whether the loss converges; if not, updating the DDPM network through loss backpropagation, and then returning to step SA4; if yes, fixing the parameters of the preprocessing module and the DDPM network, and substituting them into the base model to obtain a radionuclide diffusion model with input as a Markov transition field image and output as a radionuclide diffusion image.

[0116] Reference Figure 3 When the trained radionuclide diffusion image is used for radionuclide diffusion prediction, the steps are as follows:

[0117] S1, obtaining the radionuclide diffusion model by using the above steps SA1-SA7; and collecting time series data in a set monitoring area;

[0118] S2, randomly sampling the time series data to form new time series data with missing time points, and converting the new time series data into a Markov transition field image;

[0119] S3, inputting the Markov transition field image into the radionuclide diffusion model; the preprocessing module encodes and decodes the Markov transition field image and outputs decoded data; the DDPM network processes the decoded data and outputs a radionuclide diffusion image;

[0120] S4, performing Markov inverse encoding on the radionuclide diffusion image to obtain a prediction value of time series data of the monitoring object in the monitoring area in the prediction time period, and extracting a radionuclide concentration prediction sequence from the prediction value as a radionuclide diffusion prediction result.

[0121] The radionuclide diffusion prediction method described above will be described in combination with specific embodiments.

[0122] The embodiment simulates the radionuclide diffusion original time series data by CFD software OpenFOAM. The data simulation scenario is a leakage accident of spent fuel during transportation, and a complex underlying surface geometric model composed of three different vegetation is established, as shown in Figure 4 The model calculation area is a three-dimensional space of 1000m x 1000m x 100m, the left plane of the cube is the air inlet, the right plane is the air outlet, the airflow direction is set to be along the positive x axis, and the wind speed is set to be 4m / s. Figure 4 The medium red solid point represents the release source of the radionuclide I-131, and the coordinates are (400m, 500m, 1m). The present application assumes that the release source continuously releases for 1 hour, and the release rate is 3.3 (kBq / m

[0123] The underlying surface model contains three different vegetation groups, and different vegetation has different adsorption and flow field changing capabilities, which is called vegetation effect. The essence of the vegetation effect is that the vegetation as a porous medium will cause pressure loss to the diffusion fluid passing through its surface, and the pressure loss caused by different vegetation types and different densities is not the same. Therefore, different pressure loss coefficients are set to represent different vegetation underlying surfaces in the present application, wherein the pressure loss coefficient of vegetation group A is set to 0.5 / m, and the spatial size is 120m x 60m x 15m; the pressure loss coefficient of vegetation group B is set to 2 / m, and the spatial size is 60m x 140m x 15m; the pressure loss coefficient of vegetation group C is set to 8 / m, and the spatial size is 60m x 140m x 15m.

[0124] According to the above settings, the diffusion data of I-131 is simulated by CFD method, wherein the time step of the diffusion data collection is 10 seconds, the total simulation diffusion time is 120 minutes, and finally 720 groups of original data are obtained. The original data include time (s), radionuclide I-131 concentration (kBq / m 3 ), atmospheric pressure (Pa), X-axis wind speed (m / s), Y-axis wind speed (m / s), Z-axis wind speed (m / s), and three-dimensional coordinates. Table 1 shows part of the data samples generated by the CFD software OpenFOAM.

[0125] Table 1 shows part of the original data of the diffusion at the 60th minute

[0126]

[0127] Table 2 shows three selected monitoring areas in the three different vegetation, and the monitoring points are given by Figure 5 , wherein the red solid point represents the radionuclide release source, and the blue solid point represents the monitoring point.

[0128] Table 2 monitoring point coordinates

[0129]

[0130]

[0131] In the application, the monitoring area is abstracted as a node, the radionuclide concentration, wind speed, atmospheric pressure and the characteristics of the underlying surface in the monitoring area are abstracted as node characteristics, and the Markov transition field is used according to different node characteristics to realize image coding, and then weighted superposition is performed to obtain a Markov transition field image coupled with multiple characteristics. Figure 6 The radionuclide concentration changes of each node in the application are shown. Figure 5

[0132] Specifically, in the embodiment, first, the time sequence of the radionuclide I-131 concentration is Markov coded to obtain a Markov coded image of the concentration; the time sequence of the atmospheric pressure is Markov coded to obtain a Markov coded image of the pressure; the mean values of the X-axis wind speed (m / s), the Y-axis wind speed (m / s) and the Z-axis wind speed at each time point are calculated as the wind speed value, and the time sequence of the wind speed value is Markov coded to obtain a Markov coded image of the wind speed;

[0133] Then, the Markov coded image of the concentration, the Markov coded image of the pressure and the Markov coded image of the wind speed are weighted and superimposed to form a Markov transition field image.

[0134] Finally, the Markov transition field image is input into a radionuclide diffusion model to obtain a predicted radionuclide diffusion image, and the radionuclide diffusion image is Markov inverse coded to obtain the radionuclide I-131 concentration at the continuous M time points.

[0135] In the application, the radionuclide diffusion model in the method and the LSTM, RNN and GAT-LSTM as the comparative model all take the Markov transition field image of the node as the input to predict the radionuclide diffusion image of the node, so as to obtain the radionuclide concentration change of the node.

[0136] In the experiment, the data from the 600th second to the 3310th second are selected as the model training set, and the data from the 3320th second to the 3620th second are selected as the model test set. Table 3 shows the quantitative analysis of the prediction accuracy of each point in the three detection areas. Figure 5 Figure 7 The results of the radionuclide concentration prediction of the node A1 by the method of the application, the LSTM, the RNN and the GAT-LSTM are shown. From the results, it can be seen that the method of the application has the highest prediction accuracy. Figure 7 ​​It can be seen from the table 1 that the predicted results of the beta-VDRDPF are the closest to the true values. Meanwhile, according to the quantitative result analysis in the table 3, it can be seen that the method proposed in the present application maintains high accuracy in predicting the radionuclide concentration changes of different nodes in different regions, and is superior to the other three methods, proving that the algorithm proposed in the present application has high accuracy in predicting the radionuclide concentration changes.

[0137] Table 1 quantitative analysis of the radionuclide concentration prediction accuracy of different methods at different nodes

[0138]

[0139] Of course, for those skilled in the art, the present application is not limited to the details of the above exemplary embodiments, but also includes the same or similar structures that can be realized in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0140] In addition, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that those skilled in the art can understand.

[0141] The technologies, shapes, and structural parts not described in detail in the present application are well-known technologies.

Claims

1. A method for training a nuclide diffusion model, characterized in that: The following steps are involved: SA1, build the basic model, the first data set and the second data set; The first data set is used to store time series data with some missing time points; The second data set is used to store complete time series data annotated with radionuclide diffusion images; the time series data includes monitoring objects collected at multiple consecutive monitoring time points in the monitoring area; the monitoring objects include radionuclide concentrations; The basic model includes a preprocessing module and a DDPM network. The input of the preprocessing module is the input of the nuclide diffusion model, and the output of the DDPM network is the output of the nuclide diffusion model. The preprocessing module is used to reconstruct the input Markov transfer field image, and the DDPM network generates a Markov transfer field image for the prediction time period based on the reconstructed Markov transfer field image. SA2, processing the time series data in the first data set and the second data set into Markov transition field images; SA3, training the preprocessing module on the first data set until convergence; SA4. Extract a second training sample from the second data set, perform missing processing on the time series data of the second training sample, and then convert it into a Markov transition field image as a missing sample; SA5: The missing samples are processed by the preprocessing module to obtain a blurred image. The DDPM network processes the blurred image to obtain a predicted radionuclide diffusion image as the predicted value. SA6. Taking the radionuclide diffusion image associated with the second training sample as the true value, and calculating the loss by combining the predicted value and the true value; SA7. Determine whether the loss has converged. If not, update the DDPM network through loss backpropagation and then return to step SA4. If yes, fix the basic model as the nuclide diffusion model.

2. The method for training a nuclide diffusion model according to claim 1, wherein: In SA2, the method for processing time series data into Markov transition field images is as follows: processing the time series data into the time series of each monitored object, using the Markov transfer matrix to process the time series of each monitored object into a Markov coded image, and then performing weighted superposition on the Markov coded images of each monitored object to obtain a Markov transition field image.

3. The method for training a nuclide diffusion model according to claim 1, wherein: In SA3, the method for training the preprocessing module on the first dataset is as follows: the preprocessing module processes the Markov transition field image in the first dataset and outputs a reconstructed image as a blurred image, and the DDPM generates a radionuclide diffusion image based on the blurred image; the parameters of the preprocessing module are updated according to the optimization objective until the preprocessing module converges; the optimization objective is: The Markov transition field image representing the input of the preprocessing module, Represents the output data of the preprocessing module; Represents the radionuclide diffusion image output by DDPM; Send to DDPM network for booting generate; It means that the basic model is controlled by the parameter η. and The probability of generating x, that is, the input of the preprocessing module is The output is The probability that the DDPM network outputs x when .

4. The method for training a nuclide diffusion model according to claim 1, wherein: The monitoring objects also include one or more of atmospheric pressure, wind speed, temperature and humidity.

5. The method for training a nuclide diffusion model according to claim 4, wherein: The wind speed value is the average of the X-axis wind speed, Y-axis wind speed, and Z-axis wind speed.

6. The method for training a nuclide diffusion model according to claim 4, wherein: The preprocessing module uses the β-VAE module.

7. A method for predicting nuclide diffusion using the training method of the nuclide diffusion model according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1. Using the training method of a nuclide diffusion model according to any one of claims 1 to 5 to obtain a nuclide diffusion model; collecting time series data in a set monitoring area; S2. Randomly sample the time series data to generate new time series data with missing time points, and convert the new time series data into a Markov transition field image; S3, inputting the Markov transfer field image into the nuclide diffusion model; The preprocessing module reconstructs the Markov transfer field image; the DDPM network processes the reconstructed data and outputs the radionuclide diffusion image; S4. Perform Markov inverse coding on the radionuclide diffusion image to obtain predicted values ​​of the time series data consisting of the monitoring objects in the monitoring area in the predicted time period, and extract the radionuclide concentration prediction sequence from the predicted values ​​as the nuclide diffusion prediction result.

8. The method for predicting nuclide diffusion according to claim 7, wherein: In S2, the method for converting the new time series data into the Markov transition field image is as follows: processing the new time series data into the time series of each monitored object, using the Markov transfer matrix to process the time series of each monitored object into a Markov coded image, and then performing weighted superposition on the Markov coded images of each monitored object to obtain the Markov transition field image.

9. A nuclide diffusion prediction system, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, the processor is connected to the memory, and the processor is used to execute the computer program to implement the nuclide diffusion prediction method according to claim 7.

10. A readable medium, characterized in that A computer program is stored, and when the computer program is executed, it is used to implement the nuclide diffusion prediction method according to claim 7.

Images