Study of the energy deposition of helicon plasmas driven by machine learning algorithms

Abstract

To find a fast and reliable way predicting the energy deposition of helicon plasmas, this work focuses on machine learning algorithms. Data generation model and the distribution property of the source data are studied, then the classical algorithms and deep neural network (DNN) are built, and these algorithms are studied to test the performance on the energy deposition datasets. Both decision tree classifier (DTC) and support vector machine (SVM) find the electron temperature is the noise feature, and when it is included in the feature vector, the performance will degrade. Therefore feature selection needs be done to obtain high accuracy. For DNN, by directly changing the numbers of the hidden layers and units, test accuracy exceeds 0.95 when hidden layer is greater than three. The feature selection is automatically performed and the learning process is simpler in DNN. When the splitting ratio is varied, the generalization performance of DTC and SVM fluctuates, whereas DNN exhibits no evident change. Compared with the classical algorithms, DNN shows better stability when the source data are changed. The calculations suggest machine learning technique is a promising choice to predict the energy deposition of helicon plasmas.

Optimization of uniformity in plasma ashing process using genetic programming

“…In the field of fusion plasma, several studies applying ML to a plasma ion source were reported [7][8][9]. The optimization studies of plasma density in helicon plasmas were also reported [10][11][12]. In these cases, algorithms such as the deep neural network, decision tree classifier and support vector machine have been applied.…”

Special issue: Machine learning methods in plasma physics

“…Similarly, in Narita et al, [9] a neural network is used to compute diffusive and non-diffusive transport parameters of tokamak fusion plasmas. Cheng et al [10] compare different machine-learning algorithms for predicting properties of helicon plasmas and conclude that their deep neural networks outperform other approaches.Neural networks have very fast response times, which opens up the possibility of using them for real-time control or online monitoring diagnostic settings. Tang et al [11] describe the implementation of a recurrent neural network as a disruption predictor into a control system intended to gracefully shut down the device before a damaging disruption can occur.…”

“…Similarly, in Narita et al, [ 9 ] a neural network is used to compute diffusive and non‐diffusive transport parameters of tokamak fusion plasmas. Cheng et al [ 10 ] compare different machine‐learning algorithms for predicting properties of helicon plasmas and conclude that their deep neural networks outperform other approaches.…”