In order to further improve the communication quality of the extremely-low-frequency communication further, based on the traditional improved generalized sidelobe cancellation, a new interference suppression algorithm in the field of extremely-low-frequency communication called generative sidelobe cancellation algorithm was proposed. Generative adversarial nets as one of the hot research topics in artificial intelligence was introduced into generalized sidelobe cancellation, the network structure and relevant hyperparameters of the generative model were designed and optimized, addressing the problem of the residual desired signal existing into the original algorithm effectively, providing more relevant reference information about the interference components in the main channel for the next-stage filtering algorithm of sidelobe cancellation channel, thereby enhancing the estimation accuracy of the interference components in the main channel. In order to validate the effectiveness of the optimized generative model and the suppression ability of the proposed algorithm on different types of interferences, an experimental platform was set up under the laboratory environment and multiple sets of controlled experiments were designed. The experimental results show that the optimized generative model has better generative ability, better robustness and relatively lower computational complexity. Compared with the traditional improved algorithm, the proposed algorithm can further improve the signal-to-interference-plus-noise ratio within the signal bandwidth further.