To meet the requirements of low-cost and low-sidelobe level for space-borne phased array antennas， an optimization method for sparse arrays based on deep learning is proposed. For the optimization problem of large-scale sparse arrays， traditional genetic algorithms face issues such as high computational complexity and time-consuming fitness evaluation during the optimization process. In this paper， a deep-learning model is introduced to predict the sidelobe level， replacing the time-consuming simulation calculations， and significantly reducing the computational complexity. Experimental results show that， compared with traditional genetic algorithms， the method proposed in this paper shows some improvement in optimization effect， exhibits superior computational efficiency particularly suited for high-frequency engineering applications and can effectively solve the optimization problem of large-scale sparse arrays.