Disruption Predictor Based on Convolutional Autoencoder and Its Cross-tokamak Deployment Strategy Study

Abstract

Deploying data-driven disruption predictors across different tokamak devices is challenging due to their inherent differences. This study addresses some challenges of cross-tokamak deployment for data-driven plasma disruption predictors. Current predictors primarily employ supervised learning methods and require a balanced dataset of disruption and non-disruption shots. However, obtaining disruption shots for future tokamaks is extremely costly, resulting in imbalanced training datasets. Imbalance leads to reduced or even ineffective performance of supervised learning predictors. To solve this problem, we propose the Enhanced Convolutional Autoencoder Anomaly Detection (E-CAAD) predictor, which can be trained using disruption precursor samples when disruption shots occur. This model not only overcomes the sample imbalance problem in supervised learning predictors but also overcomes the inefficient dataset utilization faced by traditional anomaly detection predictors that cannot use disruption precursor samples for training, making it more suitable for the unpredictable datasets of future tokamaks. Compared to traditional anomaly detection predictor, the E-CAAD predictor performs better in disruption prediction and is deployed faster on new devices. Additionally, we explore strategies to accelerate deployment of E-CAAD predictor on the new device by using data from existing devices, with a focus on achieving good performance using minimal new device data during the initial operation stages. Two deployment strategies are presented: mixing data from existing devices and fine-tuning the predictor trained on existing devices. Our comparisons indicate that rational utilization of data from existing devices expedites deployment. Notably, the fine-tuning strategy yields the fastest deployment on new device among the proposed strategies.