A method and system for locating a radioactive leakage source
By constructing an atmospheric diffusion simulation sample library and using the XGBoost algorithm to fit the temporal characteristics of monitoring data, the problems of initial value dependence and network sparsity in the estimation of radioactive leak source location and leakage rate were solved, and rapid and accurate leak source location was achieved.
Patent Information
- Application Number
- CN202210607187.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-31
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-05-31
AI Technical Summary
Existing technologies are easily affected by initial values in estimating the location and leakage rate of radioactive leak sources. They rely on the density and distribution of monitoring networks, and inaccurate atmospheric diffusion patterns or significant noise in monitoring data can lead to inaccurate location results.
By constructing an atmospheric diffusion simulation sample library, extracting the temporal characteristics of monitoring data, and using the XGBoost algorithm for fitting, a leakage source localization model is established, reducing the dependence on monitoring networks and initial values, and improving data utilization and localization accuracy.
Without relying on the leakage rate, it can quickly and accurately estimate the location of the leakage source, reduce the impact of data noise and atmospheric diffusion pattern errors, and improve the accuracy and adaptability of the positioning.
Smart Images

Figure CN114925615B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method and system for locating a radioactive leakage source, belonging to the technical field of pollutant source tracing, and in particular to the technical field of radioactive pollutant source tracing. Background Art
[0002] In the Fukushima nuclear accident, because the tsunami and earthquake destroyed the power system of the plant, the information of the radioactive leakage source could not be directly measured by the detection devices in the nuclear facilities, resulting in great uncertainty in the prediction of the transport and diffusion of radioactive nuclides. Therefore, it is necessary to reconstruct the release source term (i.e., leakage rate) of the radioactive leakage source based on environmental radiation monitoring data and atmospheric diffusion models. In the radionuclide leakage incident in Europe in 2017, only the increase in the content of ruthenium-104 isotope was monitored in the European atmosphere, and the actual leakage location was actually unknown. Therefore, the location of the radioactive leakage source is also a core issue of human nuclear emergency. In most nuclear accident situations, the leakage source location and leakage rate (collectively referred to as leakage source parameters) are often unknown, and many existing technologies are studied for the estimation of these two parameters. The estimation method of leakage rate has become increasingly mature, but the estimation technology of leakage source location is still in the stage of continuous exploration and trial. Most methods estimate the leakage rate and leakage source location at the same time.
[0003] Existing methods for estimating leakage source parameters include optimization methods and Bayesian methods. (1) Optimization method: aims to find leakage source parameters that minimize the cost function. The cost function is used to quantify the difference between the atmospheric diffusion simulation results and the environmental monitoring results. The specific form of the cost function determines the convergence speed and accuracy of the optimization method. It mainly adopts an iterative process, and continuously iterates through parameter update rules until the algorithm converges, thereby minimizing the cost function. Common optimization methods include least squares method, regularization method, quasi-Newton optimization method, simulated annealing algorithm and genetic algorithm. (2) Bayesian method: The input and model form of the algorithm are specified by probability density function, which contains information such as the uncertainty of input data and atmospheric diffusion pattern, randomly samples the leakage source parameter space, estimates the posterior probability distribution of the parameters, and generates leakage source parameter estimation results with confidence levels. Common Bayesian methods include Markov chain Monte Carlo sampling method (MCMC), sequential Monte Carlo sampling method (SMC) and differential evolution Monte Carlo method (DEMC).
[0004] The above optimization methods are easily affected by the initial values of the given leakage source parameters, thus falling into local optimality. Many technologies use hybrid algorithms to simultaneously absorb the advantages of global and local searches, but leakage source parameter estimation is an ill-posed problem. Data disturbances may have a great impact on parameter estimation results, especially positioning results. The likelihood function of the above Bayesian method is difficult to select in actual nuclear emergency situations. Although many technologies can correct the form of the likelihood function by combining the distribution of observation errors and simulation errors, they are often only applicable to a single modeling scenario. Moreover, both optimization methods and Bayesian methods have a high dependence on the monitoring network, and the inaccuracy of the atmospheric diffusion model or the presence of monitoring data noise will greatly affect the parameter estimation results. Summary of the invention
[0005] In view of the above problems, the purpose of the present invention is to provide a method and system for locating a radioactive leakage source, which does not rely on the initial value of the leakage source position, uses the time series characteristics of radiation monitoring data to improve the utilization rate of monitoring data information, improves the problem of insufficient information provided by the monitoring network, and can estimate the leakage source position when the leakage rate is unknown.
[0006] In order to achieve the above-mentioned purpose, the present invention proposes the following technical scheme: a method for locating a radioactive leakage source, comprising the following steps: determining the magnitude of a unit leakage rate according to the atmospheric diffusion simulation results of a boundary point, and obtaining a unit leakage rate vector according to the magnitude of the unit leakage rate; using different grid positions in a calculation domain as simulated leakage sources, inputting the unit leakage rate vector as a source term into an atmospheric diffusion model for simulation, and constructing an atmospheric diffusion simulation sample library according to the response results of each monitoring point; preprocessing sample data in the atmospheric diffusion simulation sample library; extracting the time series features (including time domain features and frequency domain features) of the preprocessed sample data, and fusing the time series features with the corresponding simulated leakage source positions to form a training data set; establishing a leakage source positioning model by fitting the data in the training data set; extracting the time series features of the environmental radiation monitoring data, and inputting the time series features into the fitted leakage source positioning model to complete the estimation of the leakage source position.
[0007] Furthermore, the method for determining the magnitude of the unit leakage rate based on the results of atmospheric diffusion simulation at boundary points is as follows: establishing grid coordinates, inputting the calculation domain, determining the coordinates of the grid boundary points based on the calculation domain, and using the grid boundary points as leakage sources; establishing an atmospheric diffusion model, inputting meteorological data and known unit leakage rate data of different magnitudes, and simulating using the atmospheric diffusion model; and determining the magnitude of the unit leakage rate in the actual monitoring data by comparing the difference between the simulated data and the actual data.
[0008] Furthermore, the unit leakage rate vector q e T×1 for:
[0009] q e T×1 =1×10 λ *ones(T,1)
[0010] Among them, ones(T,1) represents a T×1 matrix of all ones, where T refers to the number of simulation time steps; λ is the magnitude of the unit leakage rate.
[0011] Furthermore, the preprocessing includes the following steps: performing sliding mean filtering on the sample data; selecting data corresponding to the time period of "obvious response" in the filtered sample data, "obvious response" means that at the same time step, the observed data is greater than or equal to the average value of all data (different leakage source locations) simulated using a unit leakage rate; by maximizing the correlation coefficient, filtering the sample library from the data obtained in the previous step; adding Gaussian random noise to the sample library filtered in the previous step to complete the preprocessing of the atmospheric diffusion simulation sample library.
[0012] Furthermore, the time domain features include volatility, mean, median and sample entropy; the frequency domain features are obtained by performing fast Fourier transform on the preprocessed sample data to obtain power spectrum density, and then statistically analyzing the amplitude statistical features and shape statistical features of the power spectrum density.
[0013] Furthermore, the training data set generated by fusing the time domain features and frequency domain features with the corresponding simulated leakage source positions is normalized and outlier zero-filled.
[0014] Furthermore, the leakage source localization model is fitted by the XGBoost algorithm based on Bayesian hyperparameter optimization and feature screening.
[0015] Furthermore, the fitting method is as follows: the data of the training data set is input into the XGBoost algorithm for model fitting, and the hyperparameters are optimized based on the Bayesian optimization method. The results of the five-fold cross validation of the training data set (determination coefficient R 2 The average value of ( ) is used as the evaluation index to iteratively update the training hyperparameters. When the maximum number of iterations is reached, the optimization is stopped to obtain the best hyperparameter combination. Then, the features are screened by the recursive feature elimination method with cross-validation to obtain the optimal leakage source location model.
[0016] Furthermore, the specific method for completing the estimation of the leakage source location is as follows: first, the radiation monitoring data is processed by sliding mean filtering, the time series features of the filtered data are extracted, and the time series features are fused to form a verification data set, and the verification data set is input into the fitted leakage source positioning model to obtain the leakage source positioning result.
[0017] The present invention also discloses a radioactive leakage source positioning system, comprising: a leakage rate magnitude acquisition module, used for determining the magnitude of the unit leakage rate according to the atmospheric diffusion simulation result of the boundary point, and obtaining the unit leakage rate vector according to the magnitude of the unit leakage rate; a sample library establishment module, used for inputting the unit leakage rate vector into the atmospheric diffusion model for simulation, and constructing the atmospheric diffusion simulation sample library according to the response results of each monitoring point; a preprocessing module, used for preprocessing the sample data in the atmospheric diffusion simulation sample library; a feature extraction module, used for extracting the time series characteristics (including time domain characteristics and frequency domain characteristics) of the preprocessed sample data, and fusing the time series characteristics with the corresponding simulated leakage source position to form a training data set; a model fitting module, used for establishing a leakage source positioning model, and fitting the leakage source positioning model with the data in the training data set; a result output module, used for extracting the time series characteristics of the environmental radiation monitoring data, and inputting the time series characteristics into the fitted leakage source positioning model to complete the estimation of the leakage source position.
[0018] The present invention adopts the above technical solution, which has the following advantages:
[0019] 1. The solution of the present invention improves the utilization rate of monitoring data information by using the time series characteristics (including time domain characteristics and frequency domain characteristics) of radiation monitoring data in atmospheric radioactive leakage, which can make up for the defects of sparse or poorly distributed monitoring networks in the prior art.
[0020] 2. The solution of the present invention does not need to know the time series variation of the leakage rate. It only needs to determine the magnitude of the leakage rate through the boundary conditions of the calculation domain, and then perform forward simulation to construct data samples based on the same unit leakage rate. In addition, a simulation sample library can be constructed in advance at the location that needs to be monitored, without the need to re-perform a large number of forward simulations when a nuclear accident occurs, which is conducive to improving the rapidity of nuclear emergency response.
[0021] 3. The scheme of the present invention performs data preprocessing on both monitoring data and simulation data, thereby improving the consistency between monitoring data and simulation data and, to a certain extent, weakening the impact of data noise and inaccurate atmospheric diffusion patterns on the leakage source location results.
[0022] 4. Most of the parameters involved in data preprocessing in the present invention, all training parameters in the XGBoost algorithm and feature screening are driven by data, without the need for manual selection, and are suitable for various complex radioactive leakage scenarios with greater universality.
[0023] 5. The present invention screens the data set used to train the XGBoost model, quantifies the difference between the atmospheric diffusion pattern prediction results and the radiation monitoring results through the correlation coefficient, and then screens out a part of samples with a larger correlation coefficient for extracting time series features and constructing the XGBoost training data set, thereby improving the accuracy of XGBoost training and accelerating the training speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a flow chart of a method for locating a radioactive leakage source in one embodiment of the present invention. DETAILED DESCRIPTION
[0025] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail through specific embodiments. However, it should be understood that the provision of specific embodiments is only for a better understanding of the present invention, and they should not be understood as limitations of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be understood as indicating or implying relative importance.
[0026] Existing leak source location methods often have four defects: (1) The initial value of the leak source position has a great impact on the location result; (2) It is highly dependent on the density and distribution of the monitoring network. A sparse or poorly distributed monitoring network will greatly affect the accuracy of leak source location; (3) The leak source position and leakage rate are often estimated simultaneously, which makes the uncertainty of leak source parameter estimation affect each other. This is because leak source parameter estimation is an ill-posed problem, that is, a small data change will cause a drastic change in the calculation result. Therefore, the larger the parameter space, the greater the impact of data disturbance on the parameter estimation result; (4) Due to the inaccuracy of the atmospheric diffusion model or the presence of noise in the monitoring data, the atmospheric diffusion simulation results and the actual monitoring results are quite different, and it is impossible to accurately use the similarity between the atmospheric diffusion simulation data and the monitoring data to locate the leak source.
[0027] In view of the defects of the prior art, the present invention proposes a method and system for locating a radioactive leakage source. First, the atmospheric diffusion model is used to simulate the monitoring point response results generated by leakage at different locations in the calculation domain under the condition of unit leakage rate, and an atmospheric diffusion simulation sample library is constructed; then, the sample library is preprocessed to extract the time series characteristics (including time domain characteristics and frequency domain characteristics) of all monitoring point data, and fused to form a training data set; then, the XGBoost algorithm is used to fit the training data set, and after inputting the time series characteristics of the radiation monitoring data, the estimated result of the leakage source location can be obtained. Among them, the XGBoost training hyperparameters are optimized based on the Bayesian method, and the data features are screened by the recursive feature elimination method with cross-validation, which improves the training accuracy while avoiding overfitting, and the parameters are all driven by data, without manual participation, and the method has strong universality. The present invention requires that the location of the leakage source does not depend on the initial value selection of the leakage source position, but only obtains the result through environmental radiation monitoring data and atmospheric diffusion model; it makes full use of the time series characteristics of radiation monitoring data to improve the utilization rate of monitoring data information, and makes up for the problem of insufficient information provided by the monitoring network; it weakens the spatiotemporal coupling relationship between leakage source location and leakage rate estimation to a certain extent, that is, the leakage position is estimated using the unit leakage rate when the leakage rate is unknown; it uses the connection between radiation monitoring data and atmospheric diffusion simulation data, and proposes an appropriate preprocessing method to reduce the difference between the atmospheric diffusion model and the actual diffusion, thereby improving the accuracy of leakage source location. The scheme of the present invention is described in detail below with reference to the accompanying drawings through embodiments.
[0028] Embodiment 1
[0029] This embodiment discloses a method for locating a radioactive leakage source. Figure 1 As shown, the following steps are included:
[0030] S1 determines the magnitude of the unit leakage rate according to the atmospheric diffusion simulation results at the boundary points, and obtains the unit leakage rate vector according to the magnitude of the unit leakage rate.
[0031] Specifically, network coordinates are established according to the experimental site conditions, the calculation domain is input, the coordinates of the grid boundary points are determined according to the calculation domain, and the grid boundary points are used as leakage sources; an atmospheric diffusion model is established, and meteorological data and known unit leakage rate data of different magnitudes are input to simulate the atmospheric diffusion model; then, the magnitude of the unit leakage rate in the actual monitoring data is determined by comparing the difference between the simulated data and the actual data. It is worth noting that the "unit leakage rate" here does not refer to 1Bq / s, but is defined as 1×10 λ Bq / τ. Therefore, the unit leakage rate vector q e T×1 The calculation formula is:
[0032] q eT×1 =1×10 λ *ones(T,1)
[0033] Among them, ones(T,1) represents a T×1 matrix of all ones, where T refers to the simulation time step; λ is the magnitude of the unit leakage rate. The upper right corner of the vector T×1 represents the dimension size, assuming that the magnitude of the leakage rate is approximately constant within the time interval τ.
[0034] S2 inputs the unit leakage rate vector into the atmospheric diffusion model for simulation and builds an atmospheric diffusion simulation sample library based on the response results of each monitoring point.
[0035] Specifically, grid points are uniformly sampled in the computational domain, and the grid coordinates of the sampling points are used as the coordinates of the simulated leakage source. Assuming that there are N sampling points, the unit leakage rate vector q constructed in S1 is e T×1 As the leakage source item and meteorological data are input into the atmospheric diffusion model for simulation, the time series simulation results of leakage sources at different locations are obtained at the monitoring point. Assuming there are n monitoring points, the simulation results of a single sampling point are expressed as Indicates that Similarly, the monitoring results of all n monitoring points are given by Indicates that
[0036] S3 preprocesses the sample data in the atmospheric diffusion simulation sample library.
[0037] The preprocessing includes the following steps:
[0038] S3.1 performs sliding mean filtering on the sample data.
[0039] right Perform sliding mean filtering, and the sliding window step size is TL=5. It is worth noting that this step is unified with the processing of radiation monitoring data, and the purpose is to reduce the positioning error caused by the time evolution component of the leakage rate, thereby weakening the spatiotemporal coupling relationship between the leakage source positioning and the leakage rate estimation. The implementation method is: do not change the first TL-1 data of the original data, and modify each number starting from the TLth data to the mean of the first TL numbers including it (j=TL, TL+1,...,T), and after filtering The result is as follows:
[0040]
[0041] S3.2 selects data from the filtered sample data during the time period of “obvious response”.
[0042] In actual situations, since the radionuclide plume has not reached the monitoring sensor network, the data obtained has more 0 elements in the initial time period, which means "no obvious response". However, in the simulated data, the monitoring station always responds, so it is necessary to select the monitoring data at the time step with "obvious response" for leak source location. The standard for "obvious response" at a single time step is: the observed data is greater than or equal to the average value of all N groups of data simulated with unit leakage rate, that is:
[0043]
[0044] in, Represents the concentration data at the jth time step of the i-th set of simulation data.
[0045] Because the location of the leak source requires the characteristics of time series data, it is necessary to ensure that the selected time period is continuous. Therefore, the time period is selected by selecting the starting time step T 0 and the end time step T end To determine the starting time step T 0 is the first time step with an “obvious response”, T end is the last time step with an “obvious response”. In order not to lose effective data information, the data at different monitoring points are judged separately. 0 and T end , and finally select the smallest T 0 and the largest T end , that is,
[0046]
[0047]
[0048] [T 0,min ,T end,max ] The data within the time period is used as the final data for locating the leakage source.
[0049] S3.3 Filter the sample library from the data obtained in the previous step by maximizing the correlation coefficient.
[0050] Many leakage source location technologies use the method of minimizing the cost function to screen the sample library, where the cost function is used to describe the difference between atmospheric diffusion simulation data and radiation monitoring data. Although this method is not very accurate, using its positioning results as prior information for subsequent methods is very helpful in narrowing the positioning area. In this embodiment, the cost function is selected as the correlation coefficient multiplied by -1, that is, the cost function minimization problem is converted into a correlation coefficient maximization problem. The cost function between the response data generated by the leakage source at the monitoring point and the actual environmental monitoring data in a single sample is as follows:
[0051]
[0052] in, Represents the column vector a=(a 1 ,a 2 ,…,a i ,…a n ) takes the average value of the elements, that is, In the above formula
[0053] According to needs, a part of samples with the largest correlation coefficient can be selected for subsequent feature extraction and fused into a training data set, which can effectively improve the training accuracy and calculation speed. However, the number of samples cannot be too small. Too small a number of samples can easily lead to overfitting of the model and reduce the accuracy of leak source locating.
[0054] S3.4 adds Gaussian random noise to the sample library screened in the previous step to complete the preprocessing of the atmospheric diffusion simulation sample library.
[0055] After S3.1 to S3.3, the difference between the atmospheric diffusion simulation data and the radiation monitoring data is significantly reduced. However, in order to further simulate the radiation monitoring response in real situations and improve the accuracy of leak source location, it is very necessary to add Gaussian random noise to the simulation data. The mean of Gaussian random noise is μ = 0, and the variance is σ i =|σ oi -σ mi |,i=1,2,…,n, where σ oi is the standard deviation of the monitoring data at a single monitoring point, σ mi is the standard deviation of the simulated data at a single monitoring point, so the probability density function of Gaussian noise is expressed as:
[0056]
[0057] The formula for adding random Gaussian noise is as follows:
[0058]
[0059] in,
[0060]
[0061] Random noise is added to the simulated data of all monitoring points corresponding to all simulated leakage sources to complete the preprocessing of the atmospheric diffusion simulation sample library.
[0062] S4 extracts the time series features (including time domain features and frequency domain features) of the preprocessed sample data, and fuses the time series features with the corresponding simulated leakage source positions to form a training data set.
[0063] The time domain features include volatility (wave_rate), average (average), median (median) and sample entropy (SampEn); the frequency domain features are obtained by performing fast Fourier transform (FFT) on the preprocessed sample data to obtain the power spectrum density, and then statistically analyzing the amplitude statistical features (fft_mean) and shape statistical features (fft_shape_mean) of the power spectrum density.
[0064] The training data set generated by fusing the time domain features and frequency domain features with the corresponding simulated leakage source positions is normalized and outlier zero-filled.
[0065] Simulated concentration data of a single sample and a single measurement point For example, the calculation formulas for each feature are as follows:
[0066] (1) Volatility (wave_rate)
[0067] The calculation method of volatility is as follows: First, Normalize and obtain the 90th and 10th percentile values of the data. The difference between the two is the volatility of the column data, that is:
[0068]
[0069] Among them, quantile means to obtain the quantile value, norm means to normalize the data. Volatility can reflect the fluctuation of the data amplitude.
[0070] (2) Average and median
[0071] The mean and median reflect the central moment characteristics of the data and belong to the statistical characteristics of time series data. The calculation formula is as follows:
[0072]
[0073]
[0074] (3) Sample entropy (SampEn)
[0075] for The sample entropy is calculated as follows:
[0076] The first step is to convert the time series The 1st to the Tth data are extracted in order every t data to form a vector, so a group of vectors with t as the dimension is obtained: in
[0077] Step 2: Definition For any two vectors in the vector group and The distance between the two vectors. Subtract the elements at the same position in the two vectors, and then take the maximum absolute value to get the distance between the vectors. The calculation method is as follows:
[0078]
[0079] The third step is to give a given C mi ,statistics The number of j (1≤j≤T-t+1,j≠i) is recorded as B i For 1≤i≤T-t+1, define:
[0080] And there are:
[0081]
[0082] Step 5: Increase the dimension to t+1 and count The number of j (1≤j≤Tt,j≠i) is recorded as A i For 1≤i≤Tt, define And there are:
[0083]
[0084] So, B (t) (r) is the probability that two sequences match t points under similarity tolerance r, and A (t+1) (r) is the probability that two sequences match t+1 points within the similarity tolerance r. So the sample entropy is defined as:
[0085]
[0086] When T is a finite value, it can be estimated using the following formula:
[0087]
[0088] Specifically, std represents the standard deviation of the data. The sample entropy feature is mainly used to measure the complexity of the time series of the monitoring point response data generated by the atmospheric diffusion simulation. This complexity is determined by A (t+1) (r) and B (t) The smaller the sample entropy, the smaller the probability of the sequence generating a new pattern, which means the higher the similarity of the sequence itself and the lower the randomness. Otherwise, the more complex the sequence is, the greater the randomness is.
[0089] (4) Frequency domain features: First, The power spectrum density is obtained by performing fast Fourier transform (FFT), and then the amplitude statistical characteristics and shape statistical characteristics of the power spectrum density are counted. The data expression after FFT transformation is as follows:
[0090]
[0091] where c mk is the frequency amplitude value of the kth window, the number of windows is the time step T, and the amplitude statistical characteristics (fft_mean) and shape statistical characteristics (fft_shape_mean) of the power spectrum density are calculated respectively. The calculation formula is as follows:
[0092]
[0093]
[0094] The training data set generated by fusing the time domain features and frequency domain features with the corresponding simulated leakage source positions is normalized and outlier zero-filled.
[0095] S5 establishes a leakage source location model and fits the leakage source location model through the data in the training data set.
[0096] The leak source location model is fitted by the XGBoost algorithm based on the Bayesian optimization method and feature screening. The fitting method is: input the data of the training data set into the XGBoost algorithm for model fitting, optimize the hyperparameters based on the Bayesian optimization method, and use the results of the five-fold cross validation of the training data set (determination coefficient R 2 The average value of the training hyperparameters is used as the evaluation index to iteratively update the training hyperparameters. When the maximum number of iterations N is reached, max When N max Set to 2000, N max The larger the value, the better the optimization effect, but the calculation time will be longer. Feature selection is based on recursive feature elimination with cross-validation. These two parts are completely data-driven and do not require manual parameter selection.
[0097] S6 extracts the time series characteristics of the environmental radiation monitoring data, inputs the time series characteristics into the fitted leakage source positioning model, and completes the estimation of the leakage source position.
[0098] The specific method is: first, the radiation monitoring data Sliding mean filtering is helpful to weaken the influence of the time evolution component of the leakage rate on the location of the leakage source. The result is shown in the following formula:
[0099]
[0100] Among them, j = TL, TL+1, ..., T. The time series features of the filtered data are extracted, and the time series features are fused to form a verification data set, and the verification data set is input into the fitted leakage source location model to obtain the leakage source location result.
[0101] Embodiment 2
[0102] Based on the same inventive concept, this embodiment discloses a radioactive leakage source positioning system, including:
[0103] A leakage rate magnitude acquisition module is used to determine the magnitude of the unit leakage rate according to the atmospheric diffusion simulation results at the boundary point, and obtain the unit leakage rate vector according to the magnitude of the unit leakage rate;
[0104] The sample library establishment module is used to input the unit leakage rate vector into the atmospheric diffusion model for simulation, and to build an atmospheric diffusion simulation sample library according to the response results of each monitoring point;
[0105] A preprocessing module, used for preprocessing sample data in the atmospheric diffusion simulation sample library;
[0106] A feature extraction module is used to extract the time series features (time domain features and frequency domain features) of the preprocessed sample data, and fuse the time series features with the corresponding simulated leakage source positions to form a training data set;
[0107] A model fitting module is used to establish a leakage source location model and fit the leakage source location model through the data in the training data set;
[0108] The result output module is used to extract the time series characteristics of the environmental radiation monitoring data, input the time series characteristics into the fitted leakage source positioning model, and complete the estimation of the leakage source location.
[0109] In summary, the present invention can make full use of the time series characteristics of radiation monitoring data in the time domain and frequency domain to fit the XGBoost model, and then input the monitoring data to obtain the leakage source location results. The Belgian site release experiment was used for verification. The estimated leakage source location is very different from the true value, which is smaller than the preset minimum grid unit, breaking through the limitation of physical resolution. In particular, it does not need to determine the specific size of the leakage rate in advance, but only needs to know the magnitude to accurately locate the leakage source, which can provide assistance for the decoupling and reconstruction of the leakage source parameters.
[0110] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0111] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0112] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0113] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not deviate from the spirit and scope of the present invention should be included in the protection scope of the claims of the present invention. The above content is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A method for locating a radioactive leakage source, characterized in that: The following steps are involved: Determine the magnitude of the unit leakage rate according to the atmospheric diffusion simulation result at the boundary point, and obtain the unit leakage rate vector according to the magnitude of the unit leakage rate; The unit leakage rate vector is input into the atmospheric diffusion model for simulation, and an atmospheric diffusion simulation sample library is constructed according to the response results of each monitoring point; Preprocessing the sample data in the atmospheric diffusion simulation sample library; Extracting time series features of the preprocessed sample data, and fusing the time series features with the corresponding simulated leakage source positions to form a training data set; Establishing a leakage source positioning model, and fitting the leakage source positioning model with the data in the training data set; Extracting the time series characteristics of the environmental radiation monitoring data to be tested, inputting the time series characteristics into the fitted leakage source positioning model, and completing the estimation of the leakage source position; The method for determining the magnitude of the unit leakage rate according to the atmospheric diffusion simulation results at the boundary point is: Establishing network coordinates, inputting a calculation domain, determining the coordinates of grid boundary points according to the calculation domain, and using the grid boundary points as leakage sources; Establishing an atmospheric diffusion model according to the location of the leakage source, inputting meteorological data and known unit leakage rate data of different magnitudes to simulate the atmospheric diffusion model; By comparing the difference between the simulated data and the actual data, the magnitude of the unit leakage rate in the actual monitoring data is determined; The unit leakage rate vector for: = in, express An all-1 matrix, where refers to the simulation time step; is the order of magnitude of the unit leakage rate.
2. The method for locating a radioactive leakage source according to claim 1, characterized in that: The pre-processing comprises the following steps: Performing sliding mean filtering on the sample data; Selecting data of a time period with an obvious response from the filtered sample data, wherein the obvious response refers to that the observed data is greater than or equal to the average value of different leakage source locations simulated by using a unit leakage rate; By maximizing the correlation coefficient, the sample library is screened from the data obtained in the previous step; Gaussian random noise is added to the sample library screened in the previous step to complete the preprocessing of the atmospheric diffusion simulation sample library.
3. The method for locating a radioactive leakage source according to claim 1, characterized in that: The time series features include time domain features and frequency domain features. The time domain features include volatility, mean, median and sample entropy; the frequency domain features are obtained by performing fast Fourier transform on the preprocessed sample data to obtain power spectrum density, and then statistically analyzing the amplitude statistical features and shape statistical features of the power spectrum density.
4. The method for locating a radioactive leakage source according to claim 3, characterized in that: The training data set generated by fusing the time domain features and frequency domain features with the corresponding simulated leakage source positions is normalized and outlier zero-filled.
5. The method for locating a radioactive leakage source according to claim 1, characterized in that: The leakage source positioning model is fitted by an XGBoost algorithm based on Bayesian hyperparameter optimization and feature screening methods.
6. The method for locating a radioactive leakage source according to claim 5, characterized in that: The fitting method is as follows: inputting the data of the training data set into the XGBoost algorithm for model fitting, optimizing the hyperparameters based on the Bayesian optimization method, taking the result of the five-fold cross validation of the training data set as the evaluation index, iteratively updating the training hyperparameters, and stopping the optimization when the maximum number of iterations is reached to obtain the best hyperparameter combination.
7. The method for locating a radioactive leakage source according to claim 1, characterized in that: The specific method for completing the estimation of the leakage source position is: first, the radiation monitoring data is subjected to sliding mean filtering, the time series features of the filtered data are extracted, and the time series features are fused to form a verification data set, and the verification data set is input into the fitted leakage source positioning model to obtain the leakage source positioning result.
8. A radioactive leak source location system, characterized in that: include: A leakage rate magnitude acquisition module, used to determine the magnitude of the unit leakage rate according to the atmospheric diffusion simulation result of the boundary point, and obtain the unit leakage rate vector according to the magnitude of the unit leakage rate; A sample library building module is used to input the unit leakage rate vector into the atmospheric diffusion model, simulate under different leakage source locations, and build an atmospheric diffusion simulation sample library according to the response results of each monitoring point; A preprocessing module, used for preprocessing the sample data in the atmospheric diffusion simulation sample library; A feature extraction module, used to extract the time series features of the preprocessed sample data, and fuse the time series features with the corresponding simulated leakage source positions to form a training data set; A model fitting module, used for establishing a leakage source positioning model, and fitting the leakage source positioning model with the data in the training data set; A result output module is used to extract the time series characteristics of the environmental radiation monitoring data, input the time series characteristics into the fitted leakage source positioning model, and complete the estimation of the leakage source position; The method for determining the magnitude of the unit leakage rate according to the atmospheric diffusion simulation results at the boundary point is: Establishing network coordinates, inputting a calculation domain, determining the coordinates of grid boundary points according to the calculation domain, and using the grid boundary points as leakage sources; Establishing an atmospheric diffusion model according to the location of the leakage source, inputting meteorological data and known unit leakage rate data of different magnitudes to simulate the atmospheric diffusion model; By comparing the difference between the simulated data and the actual data, the magnitude of the unit leakage rate in the actual monitoring data is determined; The unit leakage rate vector for: = in, express A matrix of all 1s, where refers to the simulation time step; is the order of magnitude of the unit leakage rate.
