Machine learning prediction of electron density and temperature from He I line ratios

Area of Science:

• Plasma Physics
• Atomic and Molecular Physics
• Machine Learning Applications

Background:

• Accurate measurement of electron density (ne) and temperature (Te) is crucial for understanding plasma behavior.
• Traditional diagnostic methods like Langmuir probes can be intrusive or limited in certain plasma environments.
• Helium I line intensity ratios offer a non-intrusive optical method for plasma characterization.

Purpose of the Study:

• To develop and validate machine learning models for predicting electron density and temperature from Helium I line intensity ratios.
• To establish a data-driven approach for plasma diagnostics.
• To compare the accuracy of machine learning predictions with established Langmuir probe measurements.

Main Methods:

• Utilizing support vector machine regression for predictive modeling.
• Training models with measured He I line ratios as input and Langmuir probe data (ne and Te) as output.
• Validating model performance on separate evaluation datasets not used during training.

Main Results:

• Machine learning models accurately predicted electron density in the range of 0.28 × 10^18 to 3.8 × 10^18 m^-3.
• Electron temperature predictions were accurate within the range of 3.2 to 7.5 eV.
• The models successfully reproduced both absolute values and radial profiles of probe-measured ne and Te.

Conclusions:

• Machine learning, specifically support vector machine regression, provides a robust and accurate method for determining electron density and temperature from He I line ratios.
• The developed models demonstrate excellent agreement with Langmuir probe measurements, offering a promising non-intrusive diagnostic tool.
• This approach has the potential to enhance plasma characterization in various scientific and industrial applications.