In this paper, dynamic neural networks are proposed to predict the plasma disruptions in a nuclear fusion device. Disruptions are critical events where the plasma, which is magnetically confined in a vacuum vessel, becomes unstable, cools down and the confinement is suddenly destroyed. These events may damage to the vessel, so they have to be foreseen well in advance in order to take mitigating action. Dynamic neural networks act as filters, which predict one step ahead the value of diagnostic signals acquired during a plasma pulse. The prediction error of the neural network depends on the regularity of signals. For this reason, an increasing prediction error reveals that the plasma operative conditions are changing, hence a disruption could be imminent. In this work, different diagnostic approaches, network adapting parameters, and diagnosis thresholds have been tested in order to determine the best performance in terms of prediction capability.
|Publication status||Published - 2007|
|Event||10th International Conference on Engineering Applications of Neural Networks, EANN 2007 - , Greece|
Duration: 1 Jan 2007 → …
|Conference||10th International Conference on Engineering Applications of Neural Networks, EANN 2007|
|Period||1/1/07 → …|
All Science Journal Classification (ASJC) codes
- Computer Science(all)
Cannas, B., Fanni, A., Montisci, A., Murgia, G., Sonato, P., & Zedda, M. K. (2007). Dynamic neural networks for prediction of disruptions in tokamaks. Paper presented at 10th International Conference on Engineering Applications of Neural Networks, EANN 2007, Greece.