Speaker
Description
Beam emission spectroscopy (BES) is an active plasma diagnostic employed for plasma density measurements. BES synthetic diagnostics are computationally expensive and comprehensive modelling suites designed to provide a better understanding of the diagnostic’s perception of underlying plasma phenomena. RENATE-OD is an advanced BES synthetic diagnostic relying on a rate-equation solver to derive the beam emission for given input plasma profiles.
Due to the resource intensiveness of the calculation, the capabilities of RENATE-OD are severely limited in three-dimensional modelling, where hundreds of thousands of calculations can be needed.
In this work we examined a subset of neural networks called Extreme Learning Machines (ELMs) for the problem described above. We generated a robust artificial dataset consisting of pairs of realistic plasma density profiles (containing a wide range of plasma fluctuations) and temperature profiles. The corresponding linear emission density profiles were calculated with RENATE-OD. This dataset was used to train and evaluate the ELM models.
We aimed to harness the advantages of both worlds, the precision of the classical numerical calculations done by RENATE-OD and the efficiency and scalability of machine learning models. We created a model that can significantly speed up 3D modelling by predicting the solutions of the underlying linear differential equation system. Obtaining the predictions this way was found to be faster by roughly three orders of magnitude.
We also coupled our artificial dataset with existing turbulence models and added plasma turbulence to the inputs of our models to test their robustness and applicability to realistic scenarios, while using the results to further validate our work with the existing plasma physics models.
