CNN-LSTM-based radionuclide atmospheric diffusion prediction method and device

Abstract

The invention relates to a CNN-LSTM-based radionuclide atmospheric diffusion prediction method and device. The method comprises the following steps: acquiring a test set and a training set of radionuclide atmospheric diffusion prediction; a CNN-LSTM model is trained according to the training set, and a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction is obtained; the trained CNN-LSTM model is used to predict the test set, and a radionuclide concentration prediction value of radionuclide atmospheric diffusion prediction is obtained; and carrying out reverse normalization operation on the radionuclide concentration prediction value, and determining a final prediction value of radionuclide atmospheric diffusion prediction. The trained CNN-LSTM model is high in prediction accuracy and high in prediction speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclear applications, and more particularly to a radionuclide atmospheric diffusion prediction method and device based on CNN-LSTM. BACKGROUND

[0002] Radioactive contamination can cause irreversible harm to human health and the natural environment. Nuclear safety is of great concern. In the evaluation of nuclear accident consequences, radionuclide diffusion simulation research is particularly important and plays a key role in guiding nuclear emergency decision-making. Radionuclide diffusion simulation has gradually developed into a cross-disciplinary research field, involving nuclear physics, atmospheric environmental science and other disciplines, and has established a relatively mature theoretical basis and application model.

[0003] Most of the current research on radionuclide atmospheric diffusion uses complex and time-consuming physical models for simulation, which includes source term estimation, calculation of wind field patterns, and simulation of plume diffusion. The accuracy of these simulations is closely related to the ability of the model to reflect the physical laws of the real world. Considering the complex mechanism of radionuclide diffusion from diffusion to radiation damage caused by human body absorption, such as soil migration and plant metabolism, the complexity of the simulation may escalate to the extent that traditional computers cannot perform real-time calculations. Deep learning neural networks can extract complex relationships from large amounts of high-dimensional data. These hidden information and relationships in large data sets are multi-dimensional, non-linear, and multi-scale, often beyond the scope of traditional modeling based on physical principles. Moreover, trained neural networks can usually make predictions within a fraction of a second. Therefore, deep learning has unique advantages in dealing with complex radionuclide diffusion problems. SUMMARY

[0004] The embodiments of the present application provide a radionuclide atmospheric diffusion prediction method and device based on CNN-LSTM. To have a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not a general review, nor does it determine the key / important components or delineate the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0005] In a first aspect, the present application provides a radionuclide atmospheric diffusion prediction method based on CNN-LSTM, which comprises:

[0006] obtaining a test set and a training set for radionuclide atmospheric diffusion prediction;

[0007] training a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction;

[0008] using the trained CNN-LSTM model to predict the test set, to obtain the radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction;

[0009] The radionuclide concentration prediction value is subjected to a reverse normalization operation to determine the final prediction value of the radionuclide atmospheric diffusion prediction.

[0010] According to a preferred embodiment, the test set and the training set for the radionuclide atmospheric diffusion prediction are obtained, comprising:

[0011] Obtain radionuclide diffusion related data for radionuclide atmospheric diffusion prediction;

[0012] The radionuclide diffusion related data is preprocessed to determine the test set and the training set for the radionuclide atmospheric diffusion prediction.

[0013] According to a preferred embodiment, the CNN-LSTM model comprises a CNN network module, an LSTM network module and an FC network module.

[0014] According to a preferred embodiment, the CNN-LSTM model is trained according to the training set to obtain the trained CNN-LSTM model for the radionuclide atmospheric diffusion prediction, comprising:

[0015] The training set is input into the CNN network module for training to extract the potential data features of the radionuclide atmospheric diffusion prediction;

[0016] The potential data features are input into the LSTM network module for training to extract the time features of the radionuclide atmospheric diffusion prediction;

[0017] The time features are input into the FC network module for training to obtain the radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction;

[0018] After training, the trained CNN-LSTM model for the radionuclide atmospheric diffusion prediction is obtained.

[0019] According to a preferred embodiment, further comprising:

[0020] The trained CNN-LSTM model and the LSTM model are compared to comprehensively evaluate the performance of the trained CNN-LSTM model for the radionuclide atmospheric diffusion prediction.

[0021] In a second aspect, the application provides a radionuclide atmospheric diffusion prediction device based on CNN-LSTM, comprising:

[0022] The acquisition module is configured to acquire a test set and a training set of the radionuclide atmospheric diffusion prediction.

[0023] The training module is configured to train a CNN-LSTM model according to the training set, to obtain a trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction.

[0024] The prediction module is configured to use the trained CNN-LSTM model to predict the test set, to obtain a radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction.

[0025] The determination module is configured to perform an inverse normalization operation on the radionuclide concentration prediction value, to determine a final prediction value of the radionuclide atmospheric diffusion prediction.

[0026] According to a preferred embodiment, the acquisition module is specifically configured to:

[0027] Acquire radionuclide diffusion related data of the radionuclide atmospheric diffusion prediction.

[0028] Preprocess the radionuclide diffusion related data, to determine the test set and the training set of the radionuclide atmospheric diffusion prediction.

[0029] According to a preferred embodiment, the CNN-LSTM model comprises a CNN network module, an LSTM network module, and an FC network module.

[0030] According to a preferred embodiment, the training module is specifically configured to:

[0031] Input the training set into the CNN network module for training, to extract potential data features of the radionuclide atmospheric diffusion prediction.

[0032] Input the potential data features into the LSTM network module for training, to extract time features of the radionuclide atmospheric diffusion prediction.

[0033] Input the time features into the FC network module for training, to obtain a radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction.

[0034] After the training is completed, a trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction is obtained.

[0035] According to a preferred embodiment, the method further comprises:

[0036] The evaluation module is configured to compare the trained CNN-LSTM model and an LSTM model, and comprehensively evaluate the performance of the trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction.

[0037] In a third aspect, the present application provides a computer storage medium, which stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and performing the method steps described above.

[0038] In a fourth aspect, the present application provides a terminal, which can include a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and performing the method steps described above.

[0039] The technical solutions provided by the present application can include the following beneficial effects:

[0040] In the present application, the radionuclide atmospheric diffusion prediction method based on the CNN-LSTM model obtains a test set and a training set for radionuclide atmospheric diffusion prediction; trains a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction; uses the trained CNN-LSTM model to predict the test set to obtain a radionuclide concentration prediction value for radionuclide atmospheric diffusion prediction; and performs an inverse normalization operation on the radionuclide concentration prediction value to determine a final prediction value for radionuclide atmospheric diffusion prediction. The present application fully considers the influence of radionuclide diffusion related data on radionuclide diffusion, improves the accuracy and prediction speed of the trained CNN-LSTM model; compared with traditional physical models, the trained CNN-LSTM model of the present application has significantly enhanced prediction speed and high accuracy, which is of great significance for promoting the research of deep learning methods in the field of nuclear accident atmospheric diffusion, and can also provide more reference information for nuclear emergency decision-making. The trained CNN-LSTM model of the present application combines multiple deep learning models including a CNN network module, an LSTM network module and an FC network module, and is improved on the basis of traditional artificial neural networks, effectively solving the problems of gradient explosion or information loss that may occur when traditional methods process radionuclide diffusion related data; while keeping the algorithm complexity low, the model significantly enhances the information processing capability. The present application integrates and optimizes the information processed by the CNN network module and the LSTM network module, aiming to maximize the comprehensive utilization of radionuclide diffusion related data. By improving the utilization rate of data, the trained CNN-LSTM model reduces the risk caused by the inability to predict unknown situations in the prediction process, thereby improving the reliability of the prediction performance of the trained CNN-LSTM model.

[0041] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0043] Figure 1 is a flowchart of a radionuclide atmospheric diffusion prediction method based on a CNN-LSTM provided by an embodiment of the present application;

[0044] Figure 2 is a one-dimensional CNN convolution operation schematic diagram of a radionuclide atmospheric diffusion prediction method based on a CNN-LSTM provided by an embodiment of the present application;

[0045] Figure 3 is a LSTM network layer principle schematic diagram of a radionuclide atmospheric diffusion prediction method based on a CNN-LSTM provided by an embodiment of the present application;

[0046] Figure 4 is a final prediction value, LSTM model prediction value and radionuclide concentration true value comparison schematic diagram of a radionuclide atmospheric diffusion prediction method based on a CNN-LSTM provided by an embodiment of the present application;

[0047] Figure 5 is a structural schematic diagram of a radionuclide atmospheric diffusion prediction device based on a CNN-LSTM provided by an embodiment of the present application;

[0048] Figure 6 is a structural schematic diagram of a terminal provided by an embodiment of the present application.

[0049] Reference signs:

[0050] 10000, acquisition module, 20000, training module, 30000, prediction module, 40000, evaluation module, 50000, determination module;

[0051] 1001, processor, 1002, communication bus, 1003, user interface, 1004, network interface, 1005, memory. DETAILED DESCRIPTION

[0052] The following description and drawings are illustrative of specific embodiments of the present application and are not intended to limit the generality of the application.

[0053] It should be clear that the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0054] The following description refers to the accompanying drawings. Unless otherwise noted, same or similar components in different drawings have same or similar reference numerals. The implementations described in the following exemplary embodiments are not meant to represent all implementations consistent with the present application. Rather, they are simply examples of systems and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] In the description of the present application, it should be understood that the terms "first", "second" and the like are used to describe various elements, but not to indicate or imply relative importance. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances. In addition, in the description of the present application, "a plurality of" means two or more, unless otherwise specified. The association relationship of the associated objects is described, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after it.

[0056] The following will be described in conjunction with the accompanying drawings Figure 1 -Appendix Figure 4 A CNN-LSTM-based radionuclide atmospheric diffusion prediction method provided by the embodiments of the present application will be described in detail.

[0057] Please refer to Figures 1-4 The method of the embodiments of the present application can include the following steps:

[0058] Convolutional Neural Network (CNN) is a deep learning model specially designed for processing data with grid structure. The core principle of CNN is to use convolution kernel to perform local perception on input feature data through convolution layer to extract potential data features in input feature data. The convolution layer extracts low-level features step by step by performing sliding window operation on input feature data, and reduces feature dimension through pooling layer to retain the most important information. The role of CNN is to automatically learn features and extract higher level abstract features, i.e. potential data features of the present application, through multi-layer structure, which is widely used in image classification, target detection, speech recognition and natural language processing tasks.

[0059] Long Short-Term Memory (LSTM) is an enhanced version of RNN, which introduces a memory module to solve the problem of gradient disappearance. Through the introduction of forgetting gate, input gate and output gate mechanisms, it can effectively control the propagation and memory of information, i.e. potential data features in time series. It is specially designed to process and model time series data, especially those with long time dependence.

[0060] However, traditional single models have some limitations. For example, the CNN model may ignore the correlation between local features and the whole in the pooling operation, leading to the training result being easily trapped in a local minimum value instead of a global optimal solution. The LSTM model consumes a large amount of computing resources due to its complex calculation structure when processing long time series.

[0061] The CNN model is a CNN network module of the present application, and the LSTM model is an LSTM network module of the present application.

[0062] The present application proposes a radionuclide atmospheric diffusion prediction method based on CNN-LSTM, which improves the prediction accuracy and speed of the trained CNN-LSTM model by combining different deep learning algorithms. Specifically as follows:

[0063] S100, obtaining a test set and a training set for radionuclide atmospheric diffusion prediction, comprising:

[0064] Obtaining radionuclide diffusion related data for radionuclide atmospheric diffusion prediction;

[0065] Preprocessing the radionuclide diffusion related data to determine the test set and the training set for the radionuclide atmospheric diffusion prediction.

[0066] In the embodiments of the present application, the radionuclide historical concentration data and historical meteorological data of different monitoring stations at each time step are obtained, the radionuclide diffusion related data is preprocessed, supervised learning data suitable for model training is constructed, and the supervised learning data is divided into a training set and a test set. The historical meteorological data is collected by the monitoring station closest to the radionuclide release point, and the historical meteorological data includes air pressure, temperature, wind speed and humidity data.

[0067] Specifically, since the radionuclide diffusion related data obtained by actual monitoring is difficult to obtain and the training of the neural network needs a large amount of reliable data, the radionuclide diffusion time series data, that is, the radionuclide diffusion related data, is simulated and generated by using the CALPUFF program. The embodiment of the application uses a nuclear power plant site as a simulation area. Among them, the model calculation area is 20km x 20km in horizontal direction, 40 grids in east-west and north-south directions, and the spatial resolution is 500m x 500m with the release point as the center. The vertical direction is divided into 20m, 40m, 80m, 160m, 320m, 640m, 1200m, 2000m, 3000m and 10 layers. The terrain elevation data of the simulation area is used. The program uses the annual meteorological data of the nuclear power plant, and the historical meteorological data includes: ground meteorological data and upper air meteorological data, and the elements of the historical meteorological data are as follows: the ground meteorological data of the ground station per hour includes: wind speed, wind direction, air temperature, relative humidity, total cloud amount, cloud base height, station pressure and precipitation type; the upper air meteorological data of the sounding station twice a day includes: atmospheric pressure, sea level height, atmospheric temperature, wind direction and wind speed. The accident source term adopts the average release rate of nuclear accident I-131, which is about 6.08E10+11 Bq / s, in the form of continuous point source release, and the simulation time is 1 year and the sampling time is 1 hour, that is, 3600 seconds.

[0068] In order to accurately simulate the atmospheric diffusion of the radionuclide, the integral puff method of the CALPUFF model is used in the embodiment of the application, because the segmentation change of the volume of the puff in the sampling step is usually small, and the integral puff can meet the calculation requirements for the mesoscale distance transmission.

[0069] The basic concentration equation of the integral puff method of the CALPUFF model is:

[0070]

[0071] Wherein, E is the ground radionuclide concentration, the unit is Bq / m 2 ; Q is the source intensity, that is, the radionuclide release rate, the unit is Bq / s; σ x , σ y , σ z are the diffusion parameters of the wind direction, the crosswind direction and the vertical direction respectively; g is the vertical formula of the Gaussian equation; d a is the downwind distance, the unit is m; d c is the vertical wind direction distance, the unit is m; He is the effective release height, the unit is m; J is the mixing layer height, the unit is m; K is an integer, which is used to consider the multiple reflections of the pollutants on the reflection surface.

[0072] The diffusion data of radionuclide I-131 is generated by the CALPUFF model simulation, wherein the time step of data collection is 3600 seconds (1 hour), the total simulation time is 8760 hours, and finally 8760 groups of data are obtained. The data characteristics include the historical concentration data of radionuclides and air pressure, temperature, humidity and wind speed of 16 monitoring points. Part of the data generated by the CALPUFF model is shown in Table 1 below.

[0073] Table 1 Radionuclide diffusion related data

[0074]

[0075] The prediction target point is selected, the input time step M of the CNN-LSTM model is set, and the output time step P, i.e. the radionuclide concentration prediction value of the target point after P hours in the future, is output. The historical concentration data of radionuclides and historical meteorological data of the three monitoring stations closest to the target point in the Euclidean distance are selected as the input features of the CNN-LSTM model. The target point is the radionuclide release point.

[0076] The radionuclide diffusion related data, i.e. the time series data, is normalized to 0-1, the time series data is converted into supervised learning data, and sample data is created; the sample data includes: input feature vector E = {E t-M , E t-M+1 , ···, E t}, radionuclide concentration true value V t+P , wherein E t-M represents the radionuclide concentration at time t M hours ago, and V t+p means the radionuclide concentration true value at time t P hours later. The supervised learning data is divided into training set and test set in the ratio of 8:2.

[0077] The above normalizes the time series data to 0-1, which accelerates the convergence speed of the subsequent CNN-LSTM model in the training process; the calculation formula of normalization is as follows:

[0078]

[0079] , wherein x2 is the normalized input feature data including the historical concentration data of radionuclides and historical meteorological data, x1 is the input feature data including the historical concentration data of radionuclides and historical meteorological data before normalization, x min represents the minimum value of the time series data, and x max represents the maximum value of the time series data.

[0080] S200, training the CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction, comprising:

[0081] The training set is input into the CNN network module for training, convolution operation is performed in the time dimension, and potential data features for radionuclide atmospheric diffusion prediction are extracted. Specifically, the CNN network module has one convolution layer, one-dimensional convolution is adopted, the number of convolution kernels is r, the size is set to k, and the sliding step is 1. The convolution kernel slides along the time domain of the input feature data, and gradually convolves the input feature data. Each time the convolution kernel slides, the convolution kernel and the local region of the input feature data are multiplied point by point to obtain a convolution output value. This process is repeated on the entire input data until the convolution kernel traverses the entire training set. The number of convolution kernels is r, and each convolution kernel generates an output feature map. Finally, all output feature maps form all convolution output values of the convolution layer, capturing local hidden features in the data, i.e. potential data features. The one-dimensional CNN convolution operation is shown in Figure 2 As shown in the figure, the local window in the input sequence is multiplied point by point with the convolution kernel and then summed to obtain a feature value in the output sequence. Subsequently, the convolution kernel slides along the input sequence at a certain step, and the above operation is repeated to finally generate the entire output feature vector. The red area represents the part covered by the convolution kernel at present, and a feature value 1 in the output sequence generated by its calculation. Among them, the input sequence is a sequence representing the input feature data, and the input feature data can be an input one-dimensional array [1, 1, 0, 0]; the output sequence is a sequence representing the entire output feature vector, and the output feature vector can be a convolution output value [1, 1]; the input one-dimensional array [1, 1, 0, 0] is subjected to two sliding convolution operations with a step of 1 by the convolution kernel [1, 0, 1] to obtain the convolution output value [1, 1].

[0082] The potential data features are input into the LSTM network module as time series vectors for training to extract time features for radionuclide atmospheric diffusion prediction. Specifically, the LSTM network module includes an LSTM network layer. The LSTM network layer has a cell state and a hidden state. The cell state is the core component of the LSTM network layer, which is used to store and transmit long-term memory to alleviate gradient disappearance. The hidden state is the result output by the LSTM network layer at each time step, which is used for short-term memory of the network state at the current time step and is calculated from the cell state and the input potential data features, and is used for prediction or generation at the next time step. The principle of the LSTM network layer is shown in Figure 3 As shown in the figure, three core gating mechanisms, i.e. forgetting gate, input gate and output gate, are introduced to control the flow of information to solve the long sequence dependence problem. In the forgetting gate, the forgetting gate activation value ft =σ(W f ·[h t-1 ,X t ]+b f ), where h t-1 It is the hidden state of the previous time step t-1, X, at the current time step t. t W is the input at the current time step t. f and b f These are the weights and biases of the forget gate, and σ is the sigmoid activation function that restricts the result to the range [0,1]. The forget gate applies to the memory cell C at the previous time step t-1. t-1 The system determines the information to be forgotten. At the input gate, the input gate activation value i... t =σ(W i ·[h t-1 ,X t ]+b i ), where W i and b i These are the weights and biases of the input gate. Candidate memory. Among them W c and b c The input gate and candidate memory are the weights and biases, and tanh is used as the activation function to restrict the results to the range [-1, 1]. Finally, by combining the input gate and candidate memory, the memory unit at the current time step t is obtained. In the output gate, the output gate activation value is o. t =σ(W o ·[h t-1 ,X t ]+b o ), where W o and b o These are the weights and biases of the output gates. Finally, the hidden state h of the current time step t is obtained based on the memory cell state of the current time step t. t =o t ·tanh(C t ).

[0083] The time features are input into the FC network module for training to obtain the predicted radionuclide concentration value for atmospheric diffusion prediction. Specifically, the FC network module includes two fully connected layers: the first fully connected layer has N neurons and the second fully connected layer has 1 neuron, used to output the predicted radionuclide concentration value.

[0084] After training, the trained CNN-LSTM model for predicting atmospheric diffusion of the radionuclides is obtained.

[0085] The application is based on a CNN-LSTM deep learning algorithm, and the CNN-LSTM model comprises a CNN network module, an LSTM network module and an FC network module. The CNN-LSTM model is also referred to as a CNN-LSTM network module or a CNN-LSTM-based radionuclide atmospheric diffusion model.

[0086] The network structure parameter settings of the network modules of the CNN-LSTM model in the embodiments of the application are shown in Table 2.

[0087] Table 2: Parameter settings of the CNN-LSTM model

[0088]

[0089] In the radionuclide concentration prediction, the embodiments of the application set M = 5 and P = 10, that is, the radionuclide diffusion-related data in the previous 5 hours at this moment are used to predict the radionuclide concentration prediction value in the future 10th hour.

[0090] S300: Using the trained CNN-LSTM model to predict the test set to obtain the radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction.

[0091] S400: Performing an inverse normalization operation on the radionuclide concentration prediction value to determine the final prediction value of the radionuclide atmospheric diffusion prediction.

[0092] In the embodiments of the application, the inverse normalization calculation formula of the inverse normalization operation is as follows:

[0093] Y2 = Y1 x (Y max -Y min )+ Y min

[0094] wherein Y2 is the final prediction value after inverse normalization, Y1 is the radionuclide concentration prediction value before inverse normalization, Y max is the maximum value of the radionuclide historical concentration data in the test set, and Y min is the minimum value of the radionuclide historical concentration data in the test set.

[0095] The method described in the embodiments of the application further comprises:

[0096] Comparing and analyzing the trained CNN-LSTM model and the LSTM model to comprehensively evaluate the performance of the trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction.

[0097] In the embodiments of the present application, the LSTM model adopts the Adam optimizer for parameter updating in the global training process, and selects the mean absolute error (MAE) as the loss function and the mean square error (MSE) as the evaluation index; wherein the expressions of MAE and MSE are as follows:

[0098]

[0099] In the formula, n represents the total number of samples, i represents the ith sample, V p represents the predicted value of the LSTM model, V t represents the true value of the radionuclide concentration at time t.

[0100] In the embodiments of the present application, the coefficient of determination R 2 , the mean square error (MSE), the mean absolute percentage error (MAPE) and the mean absolute error (MAE) are used to comprehensively evaluate the performance of the trained CNN-LSTM model.

[0101] Table 3 Comparison of prediction performance

[0102]

[0103] From the above results, it can be seen that the trained CNN-LSTM model proposed in the present application can well fit the overall change of the radionuclide concentration over time. Compared with the single LSTM model, better results are achieved in the MAPE and R 2 and other indicators.

[0104] Part of the comparison between the final predicted value of the trained CNN-LSTM model, the predicted value of the LSTM model and the true value of the radionuclide concentration is shown in Figure 4

[0105] ​In the present application, the radionuclide atmospheric diffusion prediction method based on the CNN-LSTM model comprises the following steps: obtaining a test set and a training set for radionuclide atmospheric diffusion prediction; training a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction; predicting the test set using the trained CNN-LSTM model to obtain a radionuclide concentration prediction value for radionuclide atmospheric diffusion prediction; and performing an inverse normalization operation on the radionuclide concentration prediction value to determine a final prediction value for radionuclide atmospheric diffusion prediction. The present application fully considers the influence of radionuclide diffusion related data on radionuclide diffusion, improves the accuracy and prediction speed of the trained CNN-LSTM model, and has important significance for promoting the research of deep learning methods in the field of nuclear accident atmospheric diffusion. The trained CNN-LSTM model of the present application has a significantly enhanced prediction speed and high accuracy compared to traditional physical models, and can provide more reference information for nuclear emergency decision-making. The trained CNN-LSTM model of the present application combines multiple deep learning models including a CNN network module, an LSTM network module, and an FC network module, and is improved based on traditional artificial neural networks, effectively solving the problems of gradient explosion or information loss that may occur when traditional methods process radionuclide diffusion related data. While maintaining a low algorithm complexity, the model significantly enhances the information processing capability. The present application integrates and optimizes the information processed by the CNN network module and the LSTM network module, aiming to maximize the comprehensive utilization of radionuclide diffusion related data. By improving the utilization rate of data, the trained CNN-LSTM model reduces the risk of being unable to predict unknown situations during the prediction process, thereby improving the reliability of the prediction performance of the trained CNN-LSTM model.

[0106] The following is an embodiment of the device of the present application, which can be used to perform the method embodiment of the present application. For details not disclosed in the device embodiment of the present application, please refer to the method embodiment of the present application.

[0107] Please refer to Figure 5 which shows a structure schematic diagram of a radionuclide atmospheric diffusion prediction device based on a CNN-LSTM model according to an exemplary embodiment of the present application. The device comprises an acquisition module 10000, a training module 20000, a prediction module 30000, a determination module 40000, and an evaluation module 50000.

[0108] The acquisition module 10000 is configured to acquire a test set and a training set for radionuclide atmospheric diffusion prediction.

[0109] The training module 20000 is configured to train a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction.

[0110] a prediction module 30000, configured to perform prediction on the test set by using the trained CNN-LSTM model to obtain a radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction.

[0111] a determination module 40000, configured to perform inverse normalization operation on the radionuclide concentration prediction value to determine a final prediction value of the radionuclide atmospheric diffusion prediction.

[0112] According to a preferred embodiment, the acquisition module 10000 is specifically configured to:

[0113] acquire radionuclide diffusion related data of the radionuclide atmospheric diffusion prediction;

[0114] perform preprocessing on the radionuclide diffusion related data to determine a test set and a training set of the radionuclide atmospheric diffusion prediction.

[0115] According to a preferred embodiment, the CNN-LSTM model comprises a CNN network module, an LSTM network module and an FC network module.

[0116] According to a preferred embodiment, the training module 20000 is specifically configured to:

[0117] input the training set into the CNN network module to perform training and extract potential data features of the radionuclide atmospheric diffusion prediction;

[0118] input the potential data features into the LSTM network module to perform training and extract time features of the radionuclide atmospheric diffusion prediction;

[0119] input the time features into the FC network module to perform training and obtain a radionuclide concentration prediction value of the radionuclide atmospheric diffusion prediction;

[0120] after the training, a trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction is obtained.

[0121] According to a preferred embodiment, the method further comprises:

[0122] an evaluation module 50000, configured to compare the trained CNN-LSTM model and the LSTM model and comprehensively evaluate the performance of the trained CNN-LSTM model of the radionuclide atmospheric diffusion prediction.

[0123] It should be noted that the CNN-LSTM based radionuclide atmospheric diffusion prediction device provided in the above embodiment is only used as an example to illustrate the division of the above functional modules when the CNN-LSTM based radionuclide atmospheric diffusion prediction method is performed, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the CNN-LSTM based radionuclide atmospheric diffusion prediction device provided in the above embodiment and the CNN-LSTM based radionuclide atmospheric diffusion prediction method embodiment belong to the same concept, and the implementation process is embodied in the method embodiment, which will not be repeated here.

[0124] In the present application, the CNN-LSTM based radionuclide atmospheric diffusion prediction device obtains a test set and a training set for radionuclide atmospheric diffusion prediction; trains a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction; uses the trained CNN-LSTM model to predict the test set to obtain a radionuclide concentration prediction value for radionuclide atmospheric diffusion prediction; and performs an inverse normalization operation on the radionuclide concentration prediction value to determine a final prediction value for radionuclide atmospheric diffusion prediction. The present application fully considers the influence of radionuclide diffusion related data on radionuclide diffusion, improves the accuracy and prediction speed of the trained CNN-LSTM model; compared with traditional physical models, the trained CNN-LSTM model of the present application has significantly enhanced prediction speed and high precision, which is of great significance for promoting the research of deep learning methods in the field of nuclear accident atmospheric diffusion, and can also provide more reference information for nuclear emergency decision-making. The trained CNN-LSTM model of the present application combines multiple deep learning models including CNN network module, LSTM network module and FC network module, and is improved on the basis of traditional artificial neural network, effectively solving the problems of gradient explosion or information loss that may occur when traditional methods process radionuclide diffusion related data; while keeping the algorithm complexity low, the model's information processing capability is significantly enhanced. The present application integrates and optimizes the information processed by the CNN network module and the LSTM network module, aiming to maximize the comprehensive utilization of radionuclide diffusion related data. By improving the utilization rate of data, the risk caused by the inability to predict unknown situations during the prediction process of the trained CNN-LSTM model is reduced, thereby improving the reliability of the prediction performance of the trained CNN-LSTM model.

[0125] The present application also provides a computer readable medium having program instructions stored thereon, which, when executed by a processor, implement the CNN-LSTM based radionuclide atmospheric diffusion prediction method provided by each of the above method embodiments.

[0126] The application further provides a computer program product comprising instructions which, when executed on a computer, cause the computer to perform the radionuclide atmospheric diffusion prediction method based on the CNN-LSTM of each method embodiment described above.

[0127] Please refer to Figure 6 A terminal structure schematic diagram is provided for the embodiments of the application. The terminal can include at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0128] The communication bus 1002 is configured to realize the connection and communication between the components.

[0129] The user interface 1003 can include a display screen (Display) and a camera (Camera). Optionally, the user interface 1003 can further include a standard wired interface and a wireless interface.

[0130] The network interface 1004 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0131] The processor 1001 can include one or more processing cores. The processor 1001 connects various parts in the terminal through various interfaces and lines, executes various functions of the terminal and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 1005, and calling data stored in the memory 1005. Optionally, the processor 1001 can be realized in at least one of the following hardware forms: digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 1001 can be integrated with a combination of one or more of the following: central processing unit (CPU), graphics processing unit (GPU), and modem. The CPU is mainly used to process operating systems, user interfaces, and application programs. The GPU is responsible for rendering and drawing the content to be displayed on the display screen. The modem is used for processing wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 1001, but can be realized by a separate chip.

[0132] The memory 1005 can include a random access memory (RAM) and can also include a read-only memory (ROM). Optionally, the memory 1005 includes a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, codes, code sets, or instruction sets. The memory 1005 can include a program storage area and a data storage area, where the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area can store data involved in the various method embodiments described above, etc. The memory 1005 can also be at least one storage device located away from the aforementioned processor 1001. As shown in Figure 6 The memory 1005 as a computer storage medium can include an operating system, a network communication module, a user interface module, and a CNN-LSTM based radionuclide atmospheric diffusion prediction application program.

[0133] In the terminal shown in Figure 6 The user interface 1003 is mainly used to provide an interface for user input and obtain user input data; the processor 1001 can be used to call the CNN-LSTM based radionuclide atmospheric diffusion prediction application program stored in the memory 1005 and specifically perform the following operations:

[0134] Obtain a test set and a training set for radionuclide atmospheric diffusion prediction;

[0135] Train a CNN-LSTM model according to the training set to obtain a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction;

[0136] Use the trained CNN-LSTM model to predict the test set to obtain a radionuclide concentration prediction value for radionuclide atmospheric diffusion prediction;

[0137] Perform an inverse normalization operation on the radionuclide concentration prediction value to determine a final prediction value for radionuclide atmospheric diffusion prediction;

[0138] Compare the trained CNN-LSTM model and the LSTM model to comprehensively evaluate the performance of the trained CNN-LSTM model for radionuclide atmospheric diffusion prediction.

[0139] In one embodiment, the processor 1001 specifically performs the following operations when performing the test set and the training set of the radionuclide atmospheric dispersion prediction:

[0140] Obtain radionuclide diffusion related data of the radionuclide atmospheric dispersion prediction;

[0141] Preprocess the radionuclide diffusion related data to determine the test set and the training set of the radionuclide atmospheric dispersion prediction.

[0142] In one embodiment, the processor 1001 specifically performs the following operations when performing the training of the CNN-LSTM model according to the training set to obtain the trained CNN-LSTM model of the radionuclide atmospheric dispersion prediction:

[0143] Input the training set into the CNN network module for training to extract the potential data features of the radionuclide atmospheric dispersion prediction;

[0144] Input the potential data features into the LSTM network module for training to extract the time features of the radionuclide atmospheric dispersion prediction;

[0145] Input the time features into the FC network module for training to obtain the radionuclide concentration prediction value of the radionuclide atmospheric dispersion prediction;

[0146] After training, the trained CNN-LSTM model of the radionuclide atmospheric dispersion prediction is obtained; the CNN-LSTM model includes a CNN network module, an LSTM network module, and an FC network module.

[0147] In the present application, the radionuclide atmospheric diffusion prediction method and device based on CNN-LSTM, a test set and a training set for radionuclide atmospheric diffusion prediction are obtained; a CNN-LSTM model is trained according to the training set, and a trained CNN-LSTM model for radionuclide atmospheric diffusion prediction is obtained; the trained CNN-LSTM model is used to predict the test set, and a radionuclide concentration prediction value for radionuclide atmospheric diffusion prediction is obtained; and the radionuclide concentration prediction value is subjected to an inverse normalization operation to determine the final prediction value for radionuclide atmospheric diffusion prediction. The present application fully considers the influence of radionuclide diffusion related data on radionuclide diffusion, improves the accuracy and prediction speed of the trained CNN-LSTM model; the trained CNN-LSTM model of the present application has a significantly enhanced prediction speed and high precision compared to traditional physical models, has important significance for promoting the research of deep learning methods in the field of nuclear accident atmospheric diffusion, and can also provide more reference information for nuclear emergency decision-making. The trained CNN-LSTM model of the present application combines various deep learning models including CNN network modules, LSTM network modules and FC network modules, and is improved on the basis of traditional artificial neural networks, effectively solving the problems of gradient explosion or information loss that may occur when traditional methods process radionuclide diffusion related data; while keeping the algorithm complexity low, the information processing capacity of the model is significantly enhanced. The present application integrates and optimizes the information processed by the CNN network module and the LSTM network module, aiming to maximize the comprehensive utilization of radionuclide diffusion related data. By improving the utilization rate of data, the risk caused by the inability to predict unknown situations during the prediction process of the trained CNN-LSTM model is reduced, thereby improving the reliability of the prediction performance of the trained CNN-LSTM model.

[0148] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium and can include the processes of the above-mentioned embodiments when executed.

[0149] The above only discloses preferred embodiments of the present application, and of course cannot limit the scope of the rights of the present application, so equivalent changes made according to the claims of the present application are still within the scope of the present application.

Claims

1. A method for predicting atmospheric diffusion of radionuclides based on CNN-LSTM, characterized in that, Includes the following steps: Obtain test and training sets for predicting the atmospheric diffusion of radionuclides; The CNN-LSTM model is trained based on the training set to obtain the trained CNN-LSTM model for predicting the atmospheric diffusion of radionuclides. The trained CNN-LSTM model is used to predict the concentration of radionuclides on the test set to obtain the predicted concentration of radionuclides in the atmospheric diffusion prediction. The predicted values ​​of radionuclide concentrations are denormalized to determine the final predicted values ​​of atmospheric diffusion of radionuclides.

2. The method for predicting atmospheric diffusion of radionuclides based on CNN-LSTM according to claim 1, characterized in that, The test set and training set for obtaining the atmospheric diffusion prediction of radionuclides include: Obtain data related to the atmospheric diffusion of radionuclides for prediction; The radionuclide diffusion-related data are preprocessed to determine the test set and training set for the atmospheric diffusion prediction of the radionuclide.

3. The method for predicting atmospheric diffusion of radionuclides based on CNN-LSTM according to claim 1, characterized in that, The CNN-LSTM model includes: a CNN network module, an LSTM network module, and an FC network module.

4. The method for predicting atmospheric diffusion of radionuclides based on CNN-LSTM according to claim 3, characterized in that, The step of training a CNN-LSTM model based on the training set to obtain the trained CNN-LSTM model for predicting the atmospheric diffusion of radionuclides includes: The training set is input into the CNN network module for training to extract potential data features for the atmospheric diffusion prediction of radionuclides. The potential data features are input into the LSTM network module for training to extract the temporal features of the atmospheric diffusion prediction of the radionuclide. The time features are input into the FC network module for training to obtain the predicted values ​​of radionuclide concentrations for atmospheric diffusion prediction of radionuclides. After training, the trained CNN-LSTM model for predicting atmospheric diffusion of the radionuclides is obtained.

5. The method for predicting atmospheric diffusion of radionuclides based on CNN-LSTM according to claim 1, characterized in that, Furthermore, it also includes: The trained CNN-LSTM model and the LSTM model are compared to comprehensively evaluate the performance of the trained CNN-LSTM model for predicting atmospheric diffusion of radionuclides.

6. A device for predicting the atmospheric diffusion of radionuclides based on CNN-LSTM, characterized in that, include: The acquisition module is used to acquire the test set and training set for predicting the atmospheric diffusion of radionuclides; The training module is used to train a CNN-LSTM model based on the training set to obtain a trained CNN-LSTM model for predicting the atmospheric diffusion of the radionuclide. The prediction module is used to predict the test set using the trained CNN-LSTM model to obtain the predicted value of the radionuclide concentration in the atmospheric diffusion prediction of the radionuclide. The determination module is used to perform an inverse normalization operation on the predicted values ​​of the radionuclide concentration to determine the final predicted value of the atmospheric diffusion of the radionuclide.

7. The CNN-LSTM-based radionuclide atmospheric diffusion prediction device according to claim 6, characterized in that, The acquisition module is specifically used for: Obtain data related to the atmospheric diffusion of radionuclides for prediction; The radionuclide diffusion-related data are preprocessed to determine the test set and training set for the atmospheric diffusion prediction of the radionuclide.

8. The CNN-LSTM-based radionuclide atmospheric diffusion prediction device according to claim 6, characterized in that, The CNN-LSTM model includes: a CNN network module, an LSTM network module, and an FC network module.

9. The CNN-LSTM-based radionuclide atmospheric diffusion prediction device according to claim 8, characterized in that, The training module is specifically used for: The training set is input into the CNN network module for training to extract potential data features for the atmospheric diffusion prediction of radionuclides. The potential data features are input into the LSTM network module for training to extract the temporal features of the atmospheric diffusion prediction of the radionuclide. The time features are input into the FC network module for training to obtain the predicted values ​​of radionuclide concentrations for atmospheric diffusion prediction of radionuclides. After training, the trained CNN-LSTM model for predicting atmospheric diffusion of the radionuclides is obtained.

10. The CNN-LSTM-based radionuclide atmospheric diffusion prediction device according to claim 6, characterized in that, Furthermore, it also includes: The evaluation module is used to compare the trained CNN-LSTM model and the LSTM model to comprehensively evaluate the performance of the trained CNN-LSTM model for predicting atmospheric diffusion of radionuclides.

Images