Neural Network Application for Plasma Position Detection
A neural network was successful in the detection of the vertical plasma position (Zp) with a high accuracy during disruptions.
Application of this new concept to a real-time plasma position control will be effective in protecting the device from the Vertical Displacement Event (VDE).
- Position detection using an ordinary regression formula gives an insufficient Zp prediction during disruption, resulting in VDE. The neural network can predict Zp with a high accuracy (+ 3 cm) even for a transient plasma (Fig. 1).
- This neural network consists of 4-layer perceptron and has been trained with 3000 cases of JT-60U's disruption data (Fig. 2).