MicroRNA(miRNA) is endogenous small non-coding RiboNucleic Acid (RNA), approximately 21~25 nucleotides in length, which plays an important role in gene expression via binding to the 3'-UnTranslated Region (UTR) of their mRNA target genes for translational repression or degradation of target messenger RNA. To improve the accuracy of gene regulatory network, a Rough Set based hybrid Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) method (PSO-GA-RS) was proposed. Firstly, features of sequence information were extracted, and then using rough set dependence as a fitness function, an optimal feature subset was selected through hybrid PSO and GA. At last, Support Vector Machine (SVM) was used to establish the model to predict the unknown regulatory relationships. The experimental results show that, compared with Feature Selection based on Rough Set and PSO (PSORSFS) and Rosetta algorithm, the accuracy, F measure and Receiver Operating Characteristic (ROC) curve area of PSO-GA-RS was improved at most 5% on Arabidopsis thaliana, and improved at most 8% on Oryza sativa dataset. The proposed method achieves an improved performance in identifying true connections between miRNA and their target genes.