911³Ô¹Ï

Synopsis

Climate change makes it imperative that we shift towards a sustainable method of producing energy. Some sustainable energy sources, such as solar power, wind power, and hydroelectric power, can negatively impact the environment around them. Nuclear fusion is a promising candidate for producing energy because it would be more efficient than a fission reactor. Furthermore, a fission reactor uses radioactive fuel that would decay into nuclear waste. In contrast, a fusion reactor is fuelled primarily by nonradioactive deuterium, which exists abundantly in seawater. A popular fusion reactor implementation involves a tokamak, which uses magnetic fields to confine plasma. A major challenge is to control the shape and temperature of the plasma. At the Tokamak à Configuration Variable (TCV), a Swiss research fusion reactor at École Polytechnique Fédérale de Lausanne, researchers have applied machine learning to replace traditional control techniques. Through deep reinforcement learning, the authors were able to produce and control a diverse set of plasma configurations on the TCV using magnetic actuator coils. This approach demonstrates the potential of applying reinforcement learning in the research of fusion energy