Prediction of ICRH coupling resistance for J-TEXT

To predict the coupling efficiency of ion cyclotron resonance heating (ICRH), a data-driven prediction model for ICRH coupling resistance based on a long short-term memory (LSTM) network is developed. The model is trained using multi-diagnostic time-series data acquired from the ICRH system, the far-infrared three-wave polarimeter-interferometer (FIR), and other plasma diagnostic systems. The dataset is constructed from selected J-TEXT experimental shots with ICRH durations longer than 50 ms, including 60 shots for training and 12 shots for testing, ensuring that the data mainly correspond to time intervals beyond the initial transient phase of ICRH operation. The prediction performance of the LSTM-based model is evaluated for different numbers of input features and is compared with that of a multi-layer perceptron (MLP) model under identical prediction settings. The results show that the LSTM-based model provides improved prediction accuracy and stability compared with the MLP model, demonstrating a stronger capability to capture the temporal evolution of coupling resistance. The best prediction performance is achieved when 23 input signals are used with a prediction time of 1 ms. These results demonstrate the effectiveness of the LSTM-based model for short-term prediction of ICRH coupling resistance under steady operating conditions .