Tearing modes are harmful magnetohydrodynamic instabilities in tokamaks that reduce plasma confinement by altering the magnetic field and forming magnetic islands. Traditional research methods, mainly relying on theories, numerical simulations or experiments, face challenges due to the enormous data, complexity of experiments and high computational costs. Machine learning, particularly deep learning algorithms, has shown significant potential in various plasma applications, including disruption prediction and tokamak control. In this study, we developed a toolbox called Deep Learning for Tearing Mode, which incorporates two machine learning methods, Tearing-Mode Neural Network and Tearing-Mode Operator Neural Network, to simulate the distributions of perturbation in the linear phase of tearing modes and predict linear growth rates and nonlinear saturated kinetic energy. Through appropriate training strategies and network structures, we can enhance simulation accuracy and prediction reliability, offering faster and more reliable results compared to traditional methods. The goal of this research is to explore the feasibility of developing a surrogate model for tearing mode prediction that can be applied in complex experimental environments.
This paper presents a deep learning-based approach to simulate and predict tearing modes, which are harmful instabilities in tokamak plasmas. The proposed method, called DLTM, uses neural networks to accurately simulate the distribution of perturbations and predict the growth rate and nonlinear saturation of tearing modes, offering faster and more reliable results than traditional methods.