Classifier based on support vector machine for JET plasma configurations.

Area of Science:

• Fusion Energy Research
• Plasma Physics
• Machine Learning Applications

Background:

• The last flux surface is a key indicator of plasma configuration in fusion devices.
• Automated recognition of plasma configurations is crucial for efficient fusion reactor operation.
• Previous methods for identifying configurations were less efficient or required manual input.

Purpose of the Study:

• To develop an automated system for recognizing plasma configurations in Joint European Torus (JET) discharges.
• To leverage machine learning, specifically support vector machines, for this classification task.
• To achieve high accuracy in identifying diverse plasma configurations.

Main Methods:

• Utilized support vector machines (SVMs) for a multiclass classification problem.
• Employed the one-versus-the-rest strategy to handle multiple plasma configurations simultaneously.
• Described each plasma configuration using 12 distinct geometrical parameters derived from the last flux surface.

Main Results:

• The developed system achieved a success rate close to 100% for eight simultaneous plasma configurations.
• Demonstrated the effectiveness of SVMs in classifying complex plasma geometries.
• Validated the use of geometrical parameters from the last flux surface for configuration identification.

Conclusions:

• Automated recognition of JET plasma configurations using SVMs is highly effective and accurate.
• The developed system offers a robust solution for real-time plasma control and analysis.
• This machine learning approach significantly advances the operational efficiency of fusion devices.