An Efficient Optimization Method for Large-Solution Space Electromagnetic Automatic Design

In the field of electromagnetic design, it is sometimes necessary to search for the optimal design solution (i.e., the optimal solution) within a large solution space to complete the optimization. However, traditional optimization methods are not only slow in searching for the solution space but are also prone to becoming trapped in local optima, leading to optimization failure. This paper proposes a dual-population genetic algorithm to quickly find the optimal solution for electromagnetic optimization problems in large solution spaces. The method involves two populations: the first population uses the powerful dynamic decision-making ability of reinforcement learning to adjust the crossover probability, making the optimization process more stable and enhancing the global optimization capability of the algorithm. The second population accelerates the convergence speed of the algorithm by employing a "leader dominance" mechanism, allowing the population to quickly approach the optimal solution. The two populations are integrated through an immigration operator, improving optimization efficiency. The effectiveness of the proposed method is demonstrated through the optimization design of an electromagnetic metasurface material. Furthermore, the method designed in this paper is not limited to the electromagnetic field and has practical value in other engineering optimization areas, such as vehicle routing optimization, energy system optimization, and fluid dynamics optimization, etc.