Proof of concept of the efficient design method for high-temperature superconducting magnets employing machine learning regression

We present a newly developed machine learning based optimized design method for high-temperature superconducting (HTS) magnet. Previous optimization design methods required performing thousands to tens of thousands of magnet characteristic calculations repeatedly to evaluate the objective functions and constraints. If the computation time for analyzing magnet characteristics was long, the design process inevitably became very time-consuming. In this research, we introduce a method that uses machine learning regression techniques to achieve similar design performance while significantly reducing computation time. XGBoost algorithm was trained to create a virtual model capable of predicting the actual characteristics of the magnet. By utilizing this predictive model, which allows for much faster calculations, rather than directly computing the characteristics during the optimization process, the design process was significantly enhanced in terms of efficiency. The proposed design method was applied to the design of a 2 T-class HTS magnet, and it was confirmed that similar results to the previous design could be achieved much more quickly.