DeepMind controls fusion reactor with artificial intelligence

DeepMind controls fusion reactor with artificial intelligence

Fusion reactors are one of the most exciting and promising developments of the future. One of the biggest challenges at the moment is super hot plasma To keep stable inside the reactor. it’s about 100 million degrees and currently only a few moments Keep it up

the researchers Swiss Plasma Center (SPC) Ecole Special de Lausanne (EPFL) with google deep mind Accepted. They have discovered new ways to efficiently shape plasma.

AI learns to control Tokamako

Plasma is in a so called tokamak, a donut-shaped device. it generates a strong Magnetic Fieldto hold the ultra-hot plasma in place. TCV (Variable Configuration Tokamak), which is used for research in SPC, can switch between different plasma configurations. In this way, different control methods can be tried.

3D model of TCV with hot plasma surrounded by magnetic coils to hold the plasma

Deepmind analyzes configuration

So that you do not have to go through different ways, you got the help of DeepMind. For this purpose a AI Algorithm Trained to perform specific plasma configurations using simulation of TCV.

To do this, the AI ​​tried various control strategies. Based on the knowledge gained from this, AI can now be automatically find the best wayTo keep the plasma in a certain shape.

DeepMind can automatically adjust settings for desired plasma size

For example, can two different plasma Placed together in the same vessel or plasma is formed in the shape of a triangle or snowflake. The research team developed AI. less tested actual conditions,

AI enables flexible manufacturing methods

research should contribute More flexible and new design To be able to develop more efficient fusion reactors. With the use of a self-learning AI controlling a tokamak’s magnetic field, calculations could be done much faster and new control elements for future fusion reactors could be created, the researchers write.

This means that a separate, individual control system would not need to be developed for each tokamak, but the AI ​​could simply adapt to the device concerned. AI could then do for each device best configuration search for to increase performance And this to maximize energy productionThe paper says that in the journal nature has appeared.

See also  Why Electric Cars Don't Want To Get Any Lighter


Please enter your comment!
Please enter your name here