Researchers at the Princeton Plasma Physics Laboratory (PPPL), have developed advanced AI models that could transform the future of fusion energy by vastly improving plasma heating predictions.
Fusion energy, often considered the ultimate clean and abundant power source, relies heavily on controlling plasma. These new AI-driven models not only enhance the speed of predictions by an astounding 10 million times but also maintain accuracy.
“With our intelligence, we can train the AI to go even beyond the limitations of available numerical models,” explained Álvaro Sánchez Villar, the study’s lead author and a research physicist at the U.S. Department of Energy’s PPPL.
Traditional numerical models used for predicting plasma heating in fusion experiments have struggled with anomalies, particularly erratic outliers that cause spikes in heating profiles. These outliers, which cannot be explained by physical behavior, posed a significant challenge.
“We observed a parametric regime in which the heating profiles featured erratic spikes in arbitrary locations,” Sánchez Villar noted.
These random spikes complicated the accuracy of simulations, making it difficult to predict plasma behavior reliably. However, the AI models developed by the team were not only able to identify these anomalies but also eliminated them from the training data to improve accuracy. By filtering out the problematic spikes, the AI accurately predicted the physics involved without being affected by the outliers.
“We biased our model by eliminating the spikes in the training dataset, and we were still able to predict the physics,” added Sánchez Villar.
In their research, the team discovered inherent limitations in their numerical models. They performed extensive analysis to classify outliers, ultimately deciding whether to include or exclude them from AI training. When they ran the modified version of the code, the AI models produced results that perfectly aligned with earlier predictions—achieving accurate outcomes, even in the most challenging scenarios.
One of the most remarkable aspects of the study was the drastic improvement in computation time. The AI reduced the time required for simulating ion cyclotron range of frequency (ICRF) plasma heating from 60 seconds to a mere 2 microseconds. This reduction enables faster simulations while maintaining accuracy, increasing the efficiency of the entire process.
This breakthrough could accelerate the development of fusion energy by providing scientists with faster, more accurate tools to understand and control plasma behavior. The ability to predict plasma heating with precision is critical for optimizing fusion reactors and ultimately achieving sustainable and clean energy.
“With intelligent use, AI can help us solve problems not only faster but better than before and overcome our human constraints,” Sánchez Villar emphasized.