Researchers from the Princeton Plasma Physics Laboratory (PPPL) and Harvard University are using a powerful new version of Artificial Intelligence (AI) to forecast damaging disruptions that have hampered efforts to develop fusion energy.

These disruptions involve sudden losses of confinement of plasma particles and energy that cause bursts of extreme heat that can mar the sensitive and expensive fusion reactors.

The researchers are using a deep learning code called Fusion Recurrent Neural Network (FRNN) to analyzes a huge database provided by two major fusion facilities, one in California and one in the UK. The code has been able to predict these disruptions with high accuracy because, unlike traditional software, FRNN learns from its mistakes, constantly adjusting itself by sifting through a set of alternatives to come to a desired output.

The researchers believe the next step will be to move from prediction to control – that is, to steer the plasma away from regions of instability and avoid disruptions altogether.

This research promises to speed the development of fusion as a viable and clean source for generating electricity. Nuclear fusion – the reaction that powers the sun – is seen by many as a way to produce safe, clean and virtually limitless energy.