I. Introduction
With the advancement of electrification, there is an increasing demand for high-performance power converters, which need to be efficient, power-dense, lightweight, and cost-effective, leading to demanding optimization processes. Magnetic components are among the most important building blocks in power electronic systems, being utilized in a wide range of applications from cellphone chargers to megawatt solid-state transformers. Oftentimes, they consume the most time and computational resources in multi-objective optimization of the whole converter [1] due to their complex nature. To begin with, the modeling process, which is essentially a multi-physics problem, requires knowledge in many different domains. On top of that, the wide design space, including the diversity of the variables to be designed, and the extensive range of some of them, combined with various objectives, makes the optimization more challenging. Therefore, it is important to optimize the magnetic component in a smart way, which is, on the one hand, able to find the optimal design, and on the other hand, is fast and does not require too much computational power. Certain algorithms are proposed [2]–[4], but are usually limited to a specific optimization goal or application. In recent years, the development of artificial intelligence has made it a potentially powerful tool also in the field of power electronics. However, the usage of it requires data for training [5], either through FEM simulation or measurement, which is not easy to obtain. Therefore, this paper proposes a new optimization method, which is based on linear interpolation, therefore straightforward to understand and implement, but achieves very good accuracy and significantly reduces the computation time. The pre-requisite of the optimization is the modeling routine of the magnetic component itself, i.e., the methodology to evaluate the performance of a specific design of the magnetic device, e.g., losses, volume, weight, and cost, based on specifications. The modeling and optimization processes are mostly implemented in a MATLAB-based tool [6]–[8], with the former part briefly introduced in Sec. II, and the latter illustrated in depth in Sec. III. To fairly assess the performance of this algorithm, Sec. IV compares it with the brute-force sweeping in a case study, both in terms of accuracy and speed. Finally, in Sec. V, the conclusions are drawn.