Parameterization of Pedestal Values in Fusion Plasma-ParaPED Using Machine Learning
||Parameterization of Pedestal Values in Fusion Plasma-ParaPED Using Machine Learning|
||SNIC Small Compute|
||Rongzhen Chen <email@example.com>|
||Chalmers tekniska högskola|
||2018-07-19 – 2020-08-01|
Fusion power is one of very few long-term solutions for a sustainable electricity production. Fusion power is highly sensitive to the pedestal height and a reliable model of the pedestal is therefore essential. A predictive model for the pedestal height called Europed is being developed. However, a single calculated data point for Europed takes about 1 hr wall time. Therefore, it prohibits the direct inclusion of the calculated results from Europed into the predictive transport codes. Hence a model, which predicts the pedestal height accurately and encapsulates the non-linear behaviour of physical model, would be exceedingly useful.
In this project, the parameterization will be predicted and analysed by machine learning and deep neural network methods. This will require much computation time, especially for the deep neural network method. Moreover, GPU architecture in Triolith can be beneficial for this project as well.