Happy Nice Mall

In nuclear fusion, the atomic nuclei of hydrogen atoms are forced together to form heavier atoms such as helium. This produces a lot of energy compared to a small amount of fuel, making it a very efficient power source. It is much cleaner and safer than fossil fuels or conventional nuclear energy, created by fission that forces the nuclei to separate. It is also the process that powers the stars.

However, nuclear fusion on Earth is difficult to control. The problem is that atomic nuclei repel each other. Putting them together in a reactor can only be done at extremely high temperatures, often reaching hundreds of millions of degrees, that is, higher than the center of the sun. At these temperatures, matter is neither solid, liquid, nor gas. It enters a fourth state known as plasma: a choppy, superheated soup of particles.

The task is to hold the plasma inside a reactor together long enough to draw energy from it. Inside stars, plasma is held together by gravity. On Earth, researchers use a variety of tricks, including lasers and magnets. In a magnet-based reactor known as a tokamak, the plasma is held in an electromagnetic cage, forcing it to hold its shape, preventing the plasma from touching the reactor walls, which will cool it and damage the reactor.

Controlling the plasma requires constant monitoring and manipulation of the magnetic field. The team trained the reinforcement learning algorithm to do this in a simulation. After learning how to control and change the shape of the plasma inside a virtual reactor, the researchers gave him control of magnets in the Variable Configuration Tokamak (TCV), an experimental reactor in Lausanne. They found that the artificial intelligence was able to control the actual reactor without any additional tweaks. In total, the AI ​​only controlled the plasma for two seconds – but that’s as long as it could run before the TCV reactor got too hot.

quick reactions

Ten thousand times a second, the trained neural network takes 90 different measurements that describe the shape and position of the plasma, and in response adjusts the voltage at 19 magnets. This feedback loop is much faster than previous reinforcement learning algorithms have had to deal with. To speed things up, the AI ​​was split into two neural networks. A large network called the Critics learned by trial and error how to control the reactor within the simulation. The critic’s ability was then encoded in a smaller, faster network, called the actor, running inside the reactor itself.

“It’s an incredibly powerful method,” says Jonathan Citrin of the Netherlands Institute for Fundamental Energy Research, who was not involved in the study. “It’s an important first step in a very exciting direction.”

The researchers believe that using artificial intelligence to control plasma will make it easier to experiment with different conditions inside reactors, help them understand the process, and potentially accelerate the development of commercial nuclear fusion. The AI ​​has also learned how to control the plasma by adjusting magnets in ways humans haven’t tried before, suggesting there may be new reactor configurations to explore.

“With this type of control system, we can take risks we wouldn’t otherwise dare to take,” says Ambrogio Fasoli, director of the Swiss Plasma Center and head of the Eurofusion Consortium. Human operators generally do not want to force the plasma beyond certain limits. “There are events that we should definitely avoid because they damage the device,” he says. “If we are confident that we have a control system that gets us closer to the limit but not beyond it, then we can explore more possibilities. We can accelerate research.”