Conceptual design and preliminary data analysis for classification of plasma disruption event at Aditya-U tokamak

Disruption prediction and its avoidance/mitigation is an essential part of the tokamak operations, particularly for large size tokamak as the disruptions could produce very large heat loads on diverter targets and other Plasma Facing Components (PFC), and large electromagnetic forces on the Vacuum Vessel (VV) can lead to the structural damages. It directs toward disruption research to develop methods for safe and rapid shutdown of high-power tokamak plasmas. To control or avoid the disruption, data-driven methodology using time series of relevant plasma parameters are useful with sufficient anticipation time. The Ohmically heated circular limiter tokamak ADITYA (R0 = 75 cm, a = 25 cm) has been upgraded to a tokamak named the ADITYA Upgrade (ADITYA-U) with an open diverter configuration. The data driven methodology uses supervised learning techniques for classification to develop accurate automatic classifiers from large set of discharges data which will be used for regression to determine key events of time for disruption. Based on extensive literature survey, preliminary data analysis is performed for ~ 8000 previous plasma shots from large dataset of ADITYA/ADITYA-U tokamak archival. Binary classification model is developed to separate two sets of data using Support Vector Machine (SVM) and Neural Network (NN) for ADITYA TOKAMAK. Both models are compared with test data with previous shots and validated with accuracy performance which may lead towards future development of time series data-driven model for accurate prediction of disruption event for ADITYA-U tokamak.