Researchers from DeepMind and EPFL have released a very interesting article called "Magnetic Control of Tokamak Plasmas Through Deep Reinforces,". She is about how to use RL to effectively control coils in Tokamak. But let's in order:
Nuclear synthesis: what is it and why it is needed
A nuclear reaction is the process of interaction between the atomic nucleus with another atomic core or elementary particle. The reaction will often be accompanied by a change in the composition and structure of the nucleus. In this case, the kinetic energy of the obtained particles can be much higher than the original, i.e. Energy is released.
A controlled thermonuclear synthesis (UTS) is a type of nuclear reaction in which heavier nuclei forms from lighter atomic nuclei. In this case, energy is also released. UTS, unlike many types of nuclear reactions, is controlled: that is, this process can be used in the industry for the synthesis of a huge amount of energy. One of the areas that requires this type of energy is the space industry: it seems that only with the help of TCC can energy be obtained enough to perform long -term cosmic flights.
They are implemented most often with the help of magnetic retention of plasma, or "magnetic trap". To obtain thermonuclear synthesis, the reacting nuclei must have high speeds corresponding to the high plasma temperature. Plasma with such a high temperature cannot be stored in an ordinary vessel, since any existing material will instantly evaporate when in contact with plasma.
The magnetic trap is one of the ways of long -term preservation of the plasma in a stable state without its contact with the surface of the capacity in which it is contained. To create a magnetic trap, a tor -shaped vessel is most often used - Tokamak (1 picture to post).
Although the UTS is called "controlled", in fact it is quite difficult to control this process: it is difficult to achieve its stability. The main problem is to maintain the high plasma temperature inside the Tokamak vessel. In Tokamak, temperature regulation occurs using coils that create the right magnetic field for plasma stability inside the device (2 picture to post). Magnetic controller controls the coils. For different processes of TCT with different types of plasma, different configurations of the magnetic field of coils are needed - this means that for each experiment a new controller needs to be designed. At the same time, the design of even one controller for one specific type of tokamak and plasma is a difficult matter.
What does RL have to do with it:
Researchers from DeepMind and EPFL using RL have taught the Tokamak controller to control coils to stabilize plasma. The results of the experiments show that the controller with the help of RL successfully learns to control the coils so that the plasma remains stable. At the same time, the approach turned out to be universal: it works for many different plasma configurations. In general, the authors say that their approach will lead to a noticeable reduction in design efforts to create new plasma configurations.
Well, let's fly soon to open new galaxies🤓
Link to the article in Nature
#machinelearning #artificialintelligence #ai #datascience #programming #Deechnology #Deeuplearning #bigData #bigdata