Random forest model driven by CLT-simulated data for predicting the growth rate of internal kink mode in tokamak

Abstract Real-time prediction of the growth rate of internal kink modes and the timely implementation of control measures are crucial for the stable operation of tokamak devices. In this study, machine learning techniques are combined with the three-dimensional toroidal nonlinear magnetohydrodynamic code CLT (Ci-Liu-Ti) to enable real-time prediction of the growth rate of internal kink modes. The CLT code is first utilized to generate a dataset of internal kink mode growth rates with 15 input features. Based on this dataset, four machine learning algorithms are applied to predict the growth rate of internal kink modes within a timescale of less than 25 microseconds. Experimental results indicate that Random Forest model achieves the highest coefficient of determination (R 2 ) at 95.07%. Additionally, dataset size was progressively expanded, starting from 50 data points and increasing by 10 points at a time, resulting in a total of 16 datasets of varying sizes. Random Forest and other machine learning models are trained on each of these 16 datasets. The results suggest that the current dataset contains a sufficient volume of data, and the four machine learning models trained exhibit adequate generalization capabilities, making the predictions of the internal kink mode growth rate reliable.