A database for developing machine learning based disruption predictors

Abstract

Machine learning based disruption prediction methods have exhibited good prediction performance with higher success rate, lower false alarm rate and earlier warning time than physical based methods. One important thing pushed recent advances in machine learning field is high-quality training data. So a database with rich set of accurate disruption related information is crucial to the development of a high performance disruption predictor. In order to develop machine learning based disruption predictors fast and iteratively on J-TEXT, a database dedicated for machine learning disruption prediction has been built. This database provides not only disruption related labels, interfaces for querying the data based on user proved filter and auto generation of training and test data, but also interface for benchmark the predictors. Its modular design allows us to plug-in various shot analysis modules which process diagnostic signals and generate different labels automatically. These modules can be scheduled and run parallel on a cluster which will speed up the shot analysis process. The generated labels are inserted into a MongoDB NoSQL database for later querying. But one major hurdle for machine learning disruption prediction is they perform un-acceptably poor on devices other than they are trained on. It requires data from many different tokamaks to possibly develop a cross machine predictor, so this database also has a data import module which reads diagnostic data from different data sources such as MDSplus and store them as HDF5 files with unified data structure on a parallel file system. It is easy to import data from different machines and provide a unified data access interface.