LAUSR

Abstract In general, modeling the plasma displacement system of a tokamak device based on physical principles requires a lot of assumptions and linearization. To reduce the deviation between the response model and the controlled object, we applied a long short-term memory (LSTM) network that is good at processing time series and has strong memory capabilities to model the plasma horizontal displacement system. In this research, we designed a network topology of the prediction model, then trained and tested it through the historical experimental data of the HL-2A tokamak, and finally obtained a satisfactory fitting effect. The especial nonlinear network model can make accurate predictions of the plasma horizontal displacement. Another work of this paper was to use the network model as the controlled object, and proposed to design an HL-2A plasma horizontal displacement controller using model reference adaptive control (MRAC) algorithm. Simulation results showed that the output response of the control system can quickly track various input reference signals, and the controller can well control the plasma horizontal displacement and has an excellent anti-disturbance capability.