Search

CN-122019953-A - Joint adaptive parallel annealing source item inversion method for determining nuclear radiation source

CN122019953ACN 122019953 ACN122019953 ACN 122019953ACN-122019953-A

Abstract

The invention discloses a joint self-adaptive parallel annealing source item inversion method for determining a nuclear radioactive source, which comprises the following steps of obtaining a nuclear pollutant concentration actual measurement value and meteorological data of each monitoring point, establishing an SRS matrix by using a reverse calculation diffusion process of the meteorological data by adopting an accompanying method, establishing a self-adaptive inversion model, modeling a posterior probability density function of source item parameters and determining a sampling method, wherein the sampling method adopts an enhanced MCMC method, the enhanced MCMC method adopts an MCMC method and fuses a joint self-adaptive jump function and a parallel annealing algorithm, the joint self-adaptive jump function is a weighted set comprising an AM algorithm, a SCAM algorithm and a DE algorithm, taking the SRS matrix and the pollutant concentration actual measurement value as input data of the algorithm, carrying out calculation in the inversion model, and carrying out inversion and calculating nuclear radioactive source information. The method and the device reconstruct the source item rapidly and stably, and provide an effective calculation means for source item determination and accident result evaluation.

Inventors

  • LI QINGYUN
  • GUO HUAN
  • NIU YANJING
  • SONG YU
  • XU XIANGJUN
  • Lv Xindong
  • TIAN ZHIJIE
  • ZHANG JUNFANG
  • LIAN BING
  • LIU LIYE
  • LV MINGHUA
  • LI MINGYE
  • YAO RENTAI
  • ZHAO DUOXIN
  • ZHAO DAN

Assignees

  • 中国辐射防护研究院

Dates

Publication Date
20260512
Application Date
20251226

Claims (10)

  1. 1. A method for joint adaptive parallel annealing source item inversion for determining nuclear radiation sources, comprising the steps of: Acquiring the actual measurement value and meteorological data of the nuclear pollutant concentration of each monitoring point; adopting an accompanying method, and establishing an SRS matrix by utilizing a reverse calculation diffusion process of the meteorological data; establishing a self-adaptive inversion model, wherein the self-adaptive inversion model comprises modeling a posterior probability density function of a source item parameter and determining a sampling method, the sampling method adopts an enhanced MCMC method, the enhanced MCMC method adopts an MCMC method and fuses a combined self-adaptive jump function and a parallel annealing algorithm, and the combined self-adaptive jump function is a weighted set comprising an AM algorithm, a SCAM algorithm and a DE algorithm; and taking the SRS matrix and the actual measurement value of the pollutant concentration as input data of an algorithm, and carrying out calculation in the inversion model, and carrying out inversion to calculate nuclear radioactive source information.
  2. 2. A method for joint adaptive parallel annealing source term inversion for nuclear radiation source determination according to claim 1, wherein the posterior probability density function of the source term parameters is modeled by Bayesian method, namely: wherein As a function of the prior-art, As a function of the likelihood that the time of day, As a function of posterior probability density.
  3. 3. A method for determining a joint adaptive parallel annealing source term inversion of a nuclear radiation source as defined in claim 2, wherein said prior function uses a uniform prior and said likelihood function uses a log Gaussian likelihood function Wherein y is a monitor vector, x is a source vector, H is a conversion matrix, yt and cref are super parameters, , 。
  4. 4. The method for determining nuclear radiation source joint adaptive parallel annealing source item inversion according to claim 1, wherein the weight relation of AM algorithm, SCAM algorithm and DE algorithm in said joint adaptive jump function is Wherein w AM 、w SCAM 、w DE is the weight of the AM algorithm, the SCAM algorithm and the DE algorithm, and P AM 、P SCAM 、P DE is the stable acceptance rate of the AM algorithm, the SCAM algorithm and the DE algorithm.
  5. 5. The method for determining the joint adaptive parallel annealing source item inversion of a nuclear radiation source according to claim 1, wherein the weight ratio of the AM algorithm, the SCAM algorithm and the DE algorithm is 2:2:5.
  6. 6. A method for determining a combined adaptive parallel annealing source term inversion of a nuclear radiation source as defined in claim 1, wherein said parallel annealing algorithm is configured with parallel MCMC sampling chains of different temperatures, the high temperature chain flattening the probability density function such that the hopping function is not locally optimized and the probability density function at temperature T Wherein Representing the original probability density distribution.
  7. 7. A method for joint adaptive parallel annealing source term inversion for nuclear radiation source determination as in claim 6, wherein said parallel annealing algorithm employs 8 MCMC sampling chains of different temperatures, 7 of which are annealing chains.
  8. 8. The method for determining the inversion of a joint adaptive parallel annealing source term of a nuclear radiation source according to claim 7, wherein the temperature scheme of the parallel annealing algorithm adopts an exponential scheme, the geometric spacing is set to be e, and the highest temperature is set to be e 7 .
  9. 9. The method for determining the inversion of the joint adaptive parallel annealing source item of the nuclear radiation source according to claim 8, wherein the sampling step number of the MCMC sampling chain is set to 20000 steps, and the average value of sampling results of 5000 steps is taken as the final result obtained by inversion.
  10. 10. A method for joint adaptive parallel annealing source item inversion for nuclear radiation source determination as in claim 1, wherein the SRS matrix is established by reverse calculation using FLEXPART.

Description

Joint adaptive parallel annealing source item inversion method for determining nuclear radiation source Technical Field The invention relates to the technical field of nuclear accident source item inversion, in particular to a joint self-adaptive parallel annealing source item inversion method for determining a nuclear radiation source. Background More than a nuclear accident occurred in the 80 s of the 20 th century. There is a general public concern about environmental and health problems caused by atmospheric release of radionuclides. In recent years, a plurality of monitored events of abnormal radioactive isotopes occur, and extensive research on inversion of large-scale long-distance unknown position source items by the academy is initiated. On the other hand, there is also a need for a large scale inversion method that can estimate the radioactive aerosol release location and release amount to determine possible nuclear trials as well as nuclear threats. Although nuclear explosions can be monitored by infrasound, underwater sound, electromagnetic pulse, etc., inversion by atmospheric contaminant diffusion is the only effective means to determine the amount of nuclide released. Therefore, the development of a technology for inversion estimation of unknown release source items based on the mesoscale and the large scale of monitoring site data is very important, and the technology can form quick and effective traceability estimation on release of a large amount of radionuclides caused by accidents such as accidental release of nuclear power stations and isotope production plants, further form a consequence evaluation and emergency scheme quickly, and is beneficial to guaranteeing life and property safety of masses around nuclear accidents. Since the detection of the abnormal isotope I-131 and Ru-106 events in two, which occurred in 2011 and 2017, respectively, multiple countries have studied inversion of source terms at unknown positions, e.g., tichy et al propose a method to perform least squares calculations at all possible release positions and determine the possible release positions by determining the outcome of the loss function on each grid. Saunier et al propose a method of calculating a loss function using a finite-memory quasi-newton method solver and jointly determining the possible release locations by comparing the performance of the integrated FAC2 parameter results for the different release locations. The above method has difficulty in giving the probability distribution of the possibility of inversion results, and all the discretized spatial solution domains need to be traversed. Therefore, in recent years, scholars are more focused on the inversion method of probability model modeling based on a Bayesian method, such as Dumont et al, which performs inversion research on the release position of Ru-106 in 2017 based on the Metropolis-Hasting (MH) algorithm of Markov chain Monte Carlo, de Meutter et al, which performs inversion research on the event by adopting a multi-chain Markov chain Monte Carlo method, and performs inversion research in a mode of using a numerical weather forecast set and an atmospheric diffusion mode set in order to increase the robustness of the model. Such studies, while capable of giving probability density distribution of model inversion results, involve a large number of parameter settings that need to be adjusted for different cases and are difficult for a user to simply implement. The related work of the source item inversion of unknown positions in China is very little, most of researches are concentrated on small-scale researches, the applicable scene is usually close to a factory or is developed under ideal simulation experiment conditions in small scale, the adopted method is relatively simple, and the problems that a plurality of parameters are to be adjusted according to different cases and the operation difficulty for users is high are also faced. The current MCMC (Markov chain Monte Carlo) method is widely applied and has obvious effects in the field of inversion of large-scale unknown position source items. In the prior art, the MCMC method still has a lot of dilemma in practical application, and the selection of the jump function needs to be suitable for the posterior probability density function of the accident modeling itself to realize effective sampling convergence of sampling, which has a certain requirement on experience of a designer, and can ensure that the model can jump out of local optimum to find global optimum, and can describe distribution of the local probability density function sufficiently finely. At the same time, how to quickly realize convergence of the MCMC in Gao Weijie domain space to meet the requirement of emergency response is also a main difficulty of the algorithm. Both single adaptive algorithms and general parallel multi-chain MCMC have been applied in the bayesian inversion problem. Both have certain problems. For a sin