Bayesian Methods for Magnetic and Mechanical Optimization of Superconducting Magnets for Fusion

Stellarators as compact fusion power sources have incredible potential to help combat climate change. However, the task of making that a reality faces many challenges. This work uses Bayesian optimization, (BO) which is a method that is well suited to black-box optimizations, to address the complicated optimization problem inherent by stellarator design. In particular it focuses on the mechanical optimization necessary to withstand the Lorentz forces generated by the magnetic coils. This work leverages surrogate models that are constructed to integrate as much information as possible from the available data points, significantly reducing the number of required model evaluations. It showcases the efficacy of Bayesian optimization as a versatile tool for enhancing both magneto-static and mechanical properties within stellarator winding packs. Employing a suite of Bayesian optimization algorithms, we iteratively refine 2D and 3D models of solenoid and stellarator configurations, and demonstrate a 15% increase in optimization speed using multi-fidelity Bayesian optimization. For fusion technology to progresses from experimental stages to commercial viability, precise and efficient design methodologies will be essential. By emphasizing its modularity and transferability, our approach lays the foundation for streamlining optimization processes, facilitating the integration of fusion power into a sustainable energy infrastructure.

仿星器（Stellarators）作为紧凑型聚变电源，具备助力应对气候变化的巨大潜力。然而，将这一愿景变为现实仍面临诸多挑战。本研究采用贝叶斯优化（Bayesian Optimization, BO）——一种适配黑箱优化（black-box optimizations）的经典方法——来解决仿星器设计中固有的复杂优化问题，尤其聚焦于抵御磁线圈产生的洛伦兹力（Lorentz forces）所需的力学优化环节。本研究依托旨在从现有数据点中整合尽可能多信息的代理模型（surrogate models），大幅减少了所需的模型评估次数。本研究验证了贝叶斯优化作为通用工具的有效性，可同时优化仿星器绕组组件的静磁与力学性能。通过采用一系列贝叶斯优化算法，我们迭代优化了螺线管（solenoid）与仿星器构型的二维及三维模型，并证明多保真度贝叶斯优化可将优化速度提升15%。若要推动聚变技术从实验阶段迈向商业实用化，精准高效的设计方法必不可少。本研究强调方法的模块化与可迁移性，为简化优化流程、助力聚变电源融入可持续能源基础设施奠定了基础。