The combination of a neural network (NN) and a genetic algorithm (GA) achieves the autonomous structural design of a toroidal coil to optimize its resistance (R) and radiated magnetic field noise (B). Instead, of the finite element method, NNs are used to calculate R, inductance (L), and B. The NNs are trained using the relationship between four structural parameters and the R, L, and B data set. The toroidal coil structures are optimized under a constraint condition of L while effectively using trained NNs and a GA. As a result, R is reduced by 10%, the power supply efficiency is improved by 1.3%, and B is reduced by 17.6 dB.