The timely and accurate quench detection system is an important guarantee for the safe operation of fusion devices. For the superconducting magnets with rapidly changing current in the fusion device, the core challenge is to compensate for the complex induced voltage to obtain the quench voltage signal. EAST adopts a two-stage induced voltage compensation method of Co-wound Wire primary compensation and Inductive Noise Realtime Calculation (MIK) secondary compensation for quench detection. However, even after compensation, the induced voltage noise may still exceed the quench detection threshold, resulting in system misjudgment. In order to meet the demands of increasingly complicated operating environments of fusion devices and the application of High Temperature Superconductivity, we proposed a new neural network-based induced voltage compensation method IntelliMIK. It is used to replace MIK for secondary compensation. IntelliMIK has been trained using historical data from EAST and then applied on EAST. In the recent actual operation of 2000 shots (shot 141001 - shot 143000), IntelliMIK demonstrated significant advantages over MIK. The results show that IntelliMIK significantly decreases the induced voltage noise of the quench signal, can accurately detect the quench, and can effectively avoid quench misjudgment caused by peak noise.
This paper presents a new neural network-based method called IntelliMIK for detecting quenches in superconducting magnets used in fusion devices. IntelliMIK improves upon the existing MIK method by better compensating for induced voltage noise, allowing for more accurate quench detection and avoiding false alarms.