Tiding up the chaos with Genetic Algorithms: examples in Magnetically Confined Nuclear Fusion

Magnetically Confined Nuclear Fusion (MCNF) devices produce massive amounts of data, frequently redundant, affected by noise and sometimes corrupted by measurement errors. That data is used to create models in order to detect or to predict specific physical phenomena. Genetic Algorithms (GAs) can be applied to identify and select only the most relevant parameters to be included in these models. By this way, simple equations that summarize the main physics, and therefore the main causal relations, involved in these phenomena can be extracted. In this work, 2 examples of relevant events that occur in the most relevant MCNF device in operation in the world (the Joint European Torus) are addressed: Confinement Regime Classification and Disruption Prediction. With the combination of Support Vector Machines and GAs, linear equations, helpful to provide a simplified landscape of each one of these complex and extremely non-linear phenomena, are reached.