TRIAM-1Mプラズマ位置制御のためのニューラルネットワークによるプラズマ位置予測

概要

論文の詳細を見る
Methods of improving the prediction of plasma horizontal and vertical positions are discussed for a neural network. Performance of the prediction is evaluated in terms of root mean square error between true and predicted positions (RMS error), and hitting ratio of up-and-down trend. Input signals are chosen appropriatedly and the number of input steps are increased in order to improve the performance. Due to the former, the RMS error of the predictions of the horizontal and vertical positions is improved. The improvement of the hitting ratio is insignificant. Due to the latter, both the RMS error and the hitting ratio of the prediction of the horizontal position are improved, although the improvement of the RMS error and the hitting ratio of the prediction of the vertical position are insignificant. Using signals of calculating step and the difference between the signal at the calculating step and signals at past steps, the RMS error can be reduced.
社団法人プラズマ・核融合学会の論文
2001-02-25