| Major plasma disruption is one of the most serious problems when running the plasma discharge experiments on the tokamak, and its physical essence is the result of rapid evolution of the plasma global MHD instability. Electromagnetic forces, heat loads, and runaway electrons during plasma disruption can cause tremendous damage to the first wall materials, this disruption consequences will influence the mechanical components of the devices and the safe operation of fusion reactor in the future. Therefore the development of disruption prediction and disruption mitigation is a key topic of fusion research. However, plasma disruption database is necessary for data analysis and the establishment of the disruption databases is the basis of the above work. This thesis research work based on domestic and foreign research, we carry out the following work on EAST:In this thesis, the basic process of disruption mechanism is briefly introduced, the typical disruption prediction algorithms and disruption mitigation experiments around the world are overviewed. The plasma equilibrium theory and the MHD instability theory (especially VDEs) are introduced.The main physical reasons relevant to major disruptions are listed.We adopted a machine learning method, Back-Propagation (BP) neural network, to analyze the magnetic diagnostic data. The database has been generated by means of a specially adapted version of an MHD equilibrium code EFIT with reference to the EAST geometry and stored in the EAST mdsplus database. The network uses external magnetic measurements as input data and the selected plasma parameters as output data to train and test. A novel strategy is implemented for the selection of the optimum location of a limited number of magnetic probes based data analysis of the network. The average accuracy of the identification procedure is quite good (e.g., the maximum relative error is 0.260% of internal inductance), with a constrast of the computation results of EFIT as desired output. It has been shown that the degradation of the performance is rather small (e.g., RMS error of minor radius vary from 4.307% to 4.765%) when the number of magnetic probes is reduced by nearly half.To detect the disruption shots, the program generating the disruption database requires six conditions to establish disruption database and disruption warning database for EAST tokamak. The disruption database, based on Structured Query Language (SQL), comprises 41 disruption parameters, which include current quench characteristics, EFIT equilibrium characteristics, kinetic parameters, halo currents, vertical motion. Present most disruption databases are based on plasma experiments of non-superconducting tokamak devices. The purposes of EAST database are to find disruption characteristics and disruption statistics to fully superconducting tokamak EAST, to elucidate the physics underlying tokamak disruptions, to explore the influence of disruption to superconducting magnets. Then we analyzed several plasma parameters and disruption rate, we found that 26.8% of plasma discharges were disruption shots in the last 2012 campaign on EAST. Considering EFIT does not converge during disruptions, we use a fixed set of toroidal current filaments to represent currents flowing in the system to obtain plasma shape during disruption. We review the criteria of selected parameters on JT-60U and ASDEX-U, we choose plasma parameters for disruption warning database according to the selected method of NSTX-U. The database containing disruption-relevant parameters at many different time slices on each plasma shot (115000 time slices data), and for both disruptive and non-disruptive shots (3000 shots), would be very meaningful for disruption prediction. To analyze the parameters between different disruption wming database from different devices (EAST, C-Mod and DIII-D), The prediction parameter Ip_error is used to test. The result shows that the parameter prediction performance of EAST is better than that of DIII-D, and DIII-D is better than C-Mod under the condition of the same prediction time 10 ms. |
