The Design And Implementation Of Real-Time Disruption Prediction And Avoidance System On J-TEXT

Tokamak is one of the most promising ways to achieve controlled nuclear fusion.Plasma disruption is an unstable phenomenon during Tokamak discharge,which will lead to uncontrolled shut down and damage to devices and instruments.Disruption will reduce the safety and power generation efficiency for the future fusion reactors.Therefore,research and development of disruption prediction and avoidance system is an indispensable part of Tokamak study.In order to avoid disruption,it is necessary to predict disruption in advance.However,the physical description of a disruption is very complicated and the present studies are not able to provide a satisfactory model to describe all the characteristics of disruptions.Thus,using a data-driven machine learning method is a promising way to establish a disruption model to predict and avoid disruptions.And in order to avoid disruption,the corresponding real-time diagnosis and control system is also indispensable.This paper designed and implemented various real-time disruption prediction and avoidance system on J-TEXT.For the mode locking disruptions caused by tearing mode(TM),based on the theory that applying the resonant magnetic perturbations(RMP)in appropriate TM island phase ranges can stabilize the TM.This paper developed the feedback control system of RMP for suppressing the TM.In this paper,the feedback control system implemented on FPGA can calculate the island phase and output the control signal in real-time,the phase is obtained by acquire multi-channel of Mirnov signals.The control period is 500 ns,and the accuracy of the calculated phase is within 2 degrees by optimizing the real-time algorithm.Thus ensure the RMP can be applied in appropriate phase ranges and feedback control can be completedFor density limit disruption,this paper developed a real-time density limit disruption prediction and avoidance system.The executor of the system is the density feedback control system.Therefore,firstly,this paper developed a new real-time density feedback system based on FPGA and data of the J-TEXT newly deployed polarimeter-interferometer(POLARIS).The new density feedback system not only can integrate with the disruption prediction system to avoid the density limit disruption,but also its performance far exceeds the former interferometer-based density feedback system.The new real-time density feedback control system uses more accurate POLARIS signals and optimizes the control algorithm.The control period is reduced from 5 ms to 1 ms,and the stability,maintainability,and expansibility of the system are improved.For the density limit disruption prediction,this paper proposed and designed a time series neural network structure for density prediction.The system uses an original two-stage hybrid neural network,analyses and selects the most related signals as input.It achieved the performance of successful alarm rate more than 90%,false alarm rate less than 10%,and the average warning time more than 30 ms.In this paper,the neural network is implemented on a real-time platform and integrated with the density feedback system to form a real-time density limit disruption prediction and avoidance system.A series of experiments were designed and carried out on the J-TEXT to predict and avoid the density limit disruption successfully.This paper mainly uses LabVIEW FPGA technology to achieve high-speed data acquisition,real-time data processing and real-time algorithms,also tested and validated the effectiveness,instantaneity and reliability of these control systems in several J-TEXT campaign.