The rapid development of artificial intelligence technology provides new tools to optimize the design and application of site-specific integrases and drive innovation in this field. In this study, a site-specific integrase generation model based on artificial intelligence was designed. The learning effect of the model to generate site-specific integrase is improved by mining sequence data of site-specific integrase with feature selection and discretization, and then using generative adversarial network as a framework to extract the detail information of protein sequences by using convolutional layer, and extracting the global features of sequences by using self-attention layer. In addition, to address the degradation problem during training, a residual structure module is constructed and spectral normalization is used to ensure training stability. Meanwhile, Gumbel Softmax Trick is used to solve the problem of non-returnable gradient of discrete data generated by the model. The sequence of the site-specific integrase generated by the model showed 92% identity with the training set, which has better sequence quality. In terms of amino acid composition, the Pearson value with the natural amino acid composition was greater than 0.8, and the two were highly correlated. The site-specific integrase can increase the expression of bax protein and decrease the expression of bax-2 protein and Ki67 protein in lung cancer patients, which is favorable for patient treatment. It can up-regulate the expression of ovarian STAR, CYP11A1, CTP19A1, and 3β-HSD genes and promote steroidogenesis in ewes. The alkane content of the group of strains incorporating site-specific integrase was 57.25%~63.00% lower than that of those without the enzyme in a high concentration of petroleum pollution environment.