To address challenges such as reliance on manual expertise and low computational efficiency in the parameter design of traditional power electronic converters, this paper proposes an intelligent parameter optimization method using Bayesian optimization and neural network surrogate models, with the totem-pole power factor correction (PFC) circuit as a case study. First, detailed loss models for inductors, capacitors, and switching devices are developed. Based on design specifications, feasible parameter ranges are derived. Within these bounds, optimization is carried out iteratively. During the process, a neural network surrogate model predicts the system loss for any given parameter combination, while the expected improvement (EI) acquisition function guides the selection of the next evaluation point. This loop continues until convergence is reached. Simulation results demonstrate that the proposed method efficiently converges to the minimal-loss region across a wide range of input voltages. Compared with conventional approaches, it significantly reduces computational time, improves design efficiency, and offers strong adaptability and generalization in engineering applications.