AI Breakthrough Boosts "Artificial Sun" Fusion Energy

A Chinese research team has developed a data-driven control system for tokamak plasmas, a significant step towards achieving sustainable fusion energy. This advancement could bring the process, often described as the "ultimate clean power," closer to reality. The findings were published in Communications Physics, a journal under the Nature portfolio.

Credit: VCG

Fusion energy aims to replicate the sun's power source by fusing light atomic nuclei, which releases vast amounts of energy without carbon emissions. The tokamak, a doughnut-shaped device, is considered the most promising design for building an "artificial sun" worldwide, using powerful magnetic fields to confine ultra-hot plasma.

The main challenge for tokamaks involves keeping the plasma stable and precisely shaped long enough for fusion reactions to produce more energy than they consume. Traditional plasma control relies on computationally demanding physics models and first-principles simulators, making it difficult to efficiently train advanced controllers.

To address this, researchers from the Southwestern Institute of Physics collaborated with Zhejiang University and Zhejiang Lab. They constructed a high-fidelity, data-driven model using historical experimental data from the Huanliu-3 (HL-3) tokamak, China's most advanced magnetic confinement fusion device.

This model incorporates modern artificial intelligence techniques, including long short-term memory (LSTM) networks, self-attention mechanisms, and scheduled sampling. LSTM networks are a type of recurrent neural network capable of learning long-range dependencies in sequential data. These innovations enable accurate prediction of key plasma parameters, such as current and shape, over time.

The new approach also prevents the accumulation of errors common in traditional simulations. The team has deployed this advanced agent within the HL-3’s real-world plasma control system.

Researchers reported that the system maintained stability and adaptability even under unfamiliar conditions. It demonstrated strong robustness and "zero-shot" generalisation. Reviewers said these results represent a major step toward faster, more efficient training of intelligent controllers for future devices like ITER and for commercial fusion reactors.

---

• Chinese scientists developed a new AI method for controlling tokamak plasma.

• This data-driven system could accelerate the development of sustainable fusion energy.

• The model was trained on historical data from China's HL-3 tokamak, showing strong robustness.

Source: CGTN