A classification scheme for edge-localized modes based on their probability distributions.

Area of Science:

• Plasma physics
• Machine learning
• Data analysis

Background:

• Edge-localized modes (ELMs) are crucial plasma instabilities in fusion devices.
• Accurate classification of ELM types (e.g., Type I, Type III) is essential for understanding and controlling fusion plasmas.
• Existing classification methods can be labor-intensive and struggle with parameter uncertainties.

Purpose of the Study:

• To develop an automated, robust classification scheme for plasma phenomena, particularly edge-localized modes (ELMs).
• To handle scenarios with significant parameter uncertainties or stochastic quantities.
• To reduce the manual effort required by experts for ELM type identification.

Main Methods:

• Modeling parameters with probability distributions within a metric space.
• Employing a nearest neighbor approach for classification.
• Applying the framework to classify Type I and Type III ELMs in carbon-wall plasmas at JET.

Main Results:

• Successful implementation of an automated classification scheme for ELM types.
• Demonstrated effectiveness in handling parameters with significant uncertainties.
• Achieved fast and standardized classification of Type I and Type III ELMs.

Conclusions:

• The developed automated classification scheme is efficient and accurate for ELM identification.
• The method's generality allows for application to other plasma phenomena.
• This approach is expected to significantly aid fusion research by streamlining data analysis.