Plasma Profile Reconstruction For EAST Based On Bayesian Inference

High-parameter operation of tokamak requires optimal control of plasma current profile,density profile and temperature profile.These parameters can be given by the corresponding diagnostics,such as current density can be measured by electromagnetic,motion Stark effect,etc.;electron density can be measured by laser interferometer,Thomson scattering and so on.Part of the integral diagnostics even needs to be combined with the magnetic surface to obtain an accurate profile distribution.Compared with the traditional plasma profile reconstruction algorithm based on complex physical models,the plasma profile reconstruction based on Bayesian inference determines the conditional probability of the profile parameters in a probabilistic and statistical manner,and gives the optimal plasma profile based on the diagnostic measurement data.In this paper,it is the first time to reconstruct the plasma current density profile,plasma boundary and electron density based on multiple diagnostics through Bayesian inference for Experimental Advanced Superconducting Tokamak(EAST),and then Bayesian Integrated Data Analysis(BANANA)platform has been established.Firstly,based on multiple EAST typical equilibrium,sets of electromagnetic measurement data were constructed with the random error of 3%.The plasma current density profile reconstruction algorithm adopts Conditional AutoRegressive(CAR)prior,and fits the electromagnetic measurement data to reconstruct the plasma current density profile and plasma configuration equilibrium parameters.All 100 groups of boundary errors from up single-null configuration are all below 1 cm.There are 13 groups of boundary errors from low single-null configuration and 81 groups of boundary errors from double-null configuration greater than 1 cm and less than 2.15 cm.At the same time,it was found that,similar to EFIT,etc.,when relying only on electromagnetic measurement diagnostic data,there is a large relative error between the reconstructed plasma current density distribution and the simulated given value,with a maximum of about 20%.Therefore,referring to the position of the motion Stark effect diagnostics,the magnetic field of 5 points in the core is given to constrain the plasma current.The maximum relative error of the reconstructed plasma current density is reduced to 7%,and the result is significantly improved.Secondly,a new Advanced Squared Exponential(ASE)prior model integrating EAST plasma reference information is developed.Compared with the CAR prior model,the ASE prior model enhances the information of the prior probability.The nonstationary hyperparameters is achieved by integrating the plasma reference information,and the maximum relative error of the plasma current density is reduced to within 10%,In addition,100 groups of boundary errors based on electromagnetic diagnostics with 3%random error from the three configurations are all less than 1 cm,which greatly improves the accuracy and reliability of the plasma current profile reconstruction.For the diversity of plasma reference information selection in the construction of ASE prior model,a plasma reference information selection algorithm based on fully connected neural network model is developed,which realize the automatic adaptation of plasma reference information in ASE prior model.Next,three Bayesian models with different inputs are constructed based on Bayesian inference,which are polarization interferometer diagnostics,far-infrared diagnostics,and a joint probability model of far-infrared and polarization interferometer diagnostics.The phantom data of the electron density profile were constructed by MTANH and polynomial functions,and then the three models were verified.Among them,the joint probability model performances superiority in the reconstruction process,the reconstruction accuracy is the highest,and the maximum reconstruction error is about 5%.By analyzing the uncertainty of the model and observing the fitting of the diagnostics,it can be found that the joint probability model obtains a self-consistent electron density profile after integrating multiple diagnostics,and the fitting errors of the diagnostics are kept low.In addition,the robustness of the model is verified by adding 3%random error to the diagnostics,and the joint probability model shows strong robustness to the anti-jamming performance.Finally,on the basis of the plasma current density and electron density profile reconstruction algorithm,the integrated data analysis is established platform.The integrated analysis of various diagnostics is realized by the mean of joint probability density,and the posterior probability obtained by Bayesian inference is sampled by the Hamiltonian Monte Carlo sampling algorithm.Based on this platform,multiple diagnostic data such as electromagnetic measurement,far-infrared and polarization interferometer are used to reconstruct the plasma current density and electron density profiles.Compared with the CAR current profile reconstruction algorithm that only uses electromagnetic measurement diagnostics,the integrated data analysis algorithm reduces the maximum relative error of the plasma current density to 9.39%,which proves the accuracy and reliability of the integrated data analysis algorithm and system.