An automatic data cleaning procedure for electron cyclotron emission imaging on EAST tokamak using machine learning algorithm

Abstract

A : A new data cleaning procedure for electron cyclotron emission imaging (ECEI) of EAST tokamak is developed. Machine learning techniques, including Support Vector Machine (SVM) and decision tree, are applied to identifying saturated, zero, and weak signals of ECEI raw data, which not only reduces the effort of researchers for data analysis, but also improves the accuracy of data preprocessing. Proper training sets are sampled using massive raw ECEI data from the EAST tokamak. Optimal window size of temporal signal, kernel function, and other model parameters are obtained by model training. With the optimized parameters, the recognition rates of saturated, zero, and weak signals in raw data are 99.4%, 99.86%, and 99.9%, respectively, which proves the accuracy of this procedure.

A spectral clustering based method for automatic plasma pattern identification in electron cyclotron emission imaging on EAST Tokamak

“…We conduct the experiments on ECEI signals of 5 discharges which are selected from 80 shots in EAST tokamak ECEI database by using a data cleaning procedure introduced in ref. [16]. The number of normal signals in each of the 5 shots is higher than 3/4 of total number of signals in each shot, which indicates high data quality.…”

“…Being independent of specific plasma patterns, our method can be used to identify many kinds of plasma patterns. It should be noted that though the prior data cleaning procedure requires training sets with data quality labels [16], after obtaining the ECEI data with high quality, our method can be used without labeled data or domain knowledge about patterns. It means that our method not only can be used in identifying classic patterns, but also can help discover new patterns.…”

“…We note that the original ECEI data contains a lot of malfunctional, saturated and weak signals which are unusable. To overcome this issue, we firstly use a data cleaning procedure developed by C. Li et al [16] and then obtain 5 shots with high quality signals from 80 shots (shot 42980-43000, 51050-51070, 52310-52330, 52520-52540).…”

Overview of machine learning applications in fusion plasma experiments on J-TEXT tokamak

“…This will make it much easier for tokamak operators. In addition, there are also many other machine learning applications in tokamak diagnostics data processing and plasma simulation [7][8][9][10][11].…”