| The neutron distribution in the neutron field of reactor core is closely related to the economy and safety of reactor.Due to the limitation of reactor core operating environment and reactor type design,it is impossible to use the in-core detector to monitor neutron field in some cases.Therefore,it is a development trend to reconstruct the core neutron field based on the data from the detector outside the reactor core.In this paper,based on the spatial response principle of the ex-core detector and using the neural network technology,the reconstruction method of the in-core neutron field based on the outer core detector is studied.The main research contents,results and innovations of this paper are as follows:An ex-core detector response construction method based on response relay is proposed.The method was based on the space of the detector response theory,deduce the response relation between the adjacent phase space reactor neutron field.Then,the setup strategy of the reactor response relay region is designed,and the response calculation method between adjacent response relay areas is developed.Finally,the multiple response matrix from the ex-core detector to the in-core neutrons is obtained.The proposed method was tested based on BN600 benchmark model,and the results show that the multiple response matrix constructed by the proposed method can accurately characterize the dynamic response relationship between the ex-core detector and the in-core neutrons.In this method,the reactor data set for neural network training is proposed by using the dynamic response relationship between the ex-core detector and the neutrons in the in-core phase space,which solves the problem of insufficient network training data because a large amount of real data inside and outside the reactor are not readily available.Aiming at the neutron field reconstruction under different model complexity,the reconstruction architecture based on artificial neural network(ANN)was firstly designed.The shallow network architecture was used to realize the reconstruction of low-resolution neutron field under simple model.Secondly,a"coarse-to-fine" neutron field hierarchical reconstruction architecture is designed based on deep learning technology to achieve high-resolution neutron field reconstruction under complex model.To verify that the off-situ reconstruction method developed in this paper is feasible and stable,the CLEAR-1 reactor model and H.B.Robinson(HBR)reactor benchmark model was used to test the proposed reconstruction method.The average relative deviation ARDTotal between the neutron field reconstructed by network model and the neutron field calculated by Monte Carlo forward transport could be kept within 2%under six core change scenarios of each reactor type.The test results show that the proposed method can reconstruct the in-core neutron field data stably and accurately,which further indicates that the current off-situ neutron field reconstruction method has the prospect of engineering application in the Chinese fast lead cooled reactor CLEAR-1 and the HBR-2. |
