<p>For deep space missions, optimizing flyby sequences is essential to reach target destinations with sufficient fuel for scientific objectives. However, identifying optimal flyby sequences is challenging. Conventional global optimization algorithms typically demand a fixed number of design variables, which constrains their applicability for such complex problems. Even with a predefined number of flybys, the extensive design space makes the optimization process computationally expensive. To address this issue, the current study proposes a vectorized dynamic-size genetic algorithm (VDSGA) designed to rapidly generate optimal flyby sequences. To further enhance computational efficiency, the proposed VDSGA method is integrated with a neural network Lambert’s approximator (NNLA), enabling high-speed, accurate approximations of Lambert’s problem. This incorporation improves computational efficiency by leveraging GPU-assisted trajectory optimization without the need for implementation-level modifications, thereby reducing reliance on CPU processing. This approach also provides several different suboptimal flyby sequences, thus broadening the solution space for a given transfer. The accuracy and feasibility of these preliminary design solutions are validated through interplanetary mission examples in this paper.</p>