Development of a neural network technique for KSTAR Thomson scattering diagnostics

Area of Science:

• Plasma Physics
• Machine Learning
• Fusion Energy

Background:

• Real-time control of tokamak plasmas is crucial for diagnostics.
• Traditional Thomson scattering analysis (χ² method) is computationally intensive, hindering real-time measurements.

Purpose of the Study:

• To apply a neural network approach to Thomson scattering diagnostics.
• To calculate electron temperature and compare results with the traditional χ² method.

Main Methods:

• Implementation of a neural network for data analysis.
• Training the neural network with 10³ cycles and eight hidden layer nodes.
• Comparison of neural network results against the established χ² method.

Main Results:

• The neural network approach achieved good agreement with the χ² method.
• The neural network significantly accelerated calculations, performing them twenty times faster.
• Optimal performance was observed with 10³ training cycles and eight hidden layer nodes.

Conclusions:

• Neural networks offer a viable and efficient alternative for real-time electron temperature determination in fusion plasma diagnostics.
• The developed neural network method enhances the speed of Thomson scattering analysis without compromising accuracy.
• This advancement supports improved real-time control and measurement of plasma parameters in fusion devices.