Aiming at the limitation of the traditional Prony harmonic detection method under noise conditions, The article proposes a strategy for accurate harmonic detection under large noise conditions. By adaptively VMD decomposition of the sampled signal, we use the average periodic energy to select and reject the noise mode. Recombine the residual modes to obtain a stationary signal suitable for the Prony algorithm, and then obtain the Prony algorithm on the stationary signal. Finally, cyclic filtering and GSO optimization for harmonic characteristic information, and then we can get the harmonic characteristic information. Using this method to carry out harmonic detection simulation experiment. The simulation results show that the proposed method can effectively improve the recognition accuracy of the Prony algorithm under large noise conditions. It is adaptive and can automatically filter the real frequency components. It’s also efficient and accurate.