Acquisition (SCADA) data in wind farms, aiming to enhance data integrity. AT-SGAIN simplifies the GAIN model structure, significantly improves computational efficiency, and adopts a dual discriminator structure for distinguishing between real data and generated data, ensuring the maintenance of interpolation accuracy during the speed improvement process. This model integrates a Transformer encoder, enhancing the ability to capture time series features of wind power data. Through an adaptive dual branch attention mechanism, it accurately adjusts channel and spatial attention weights, improving the network"s sensitivity to local information. The experimental results show that the root mean square error and average absolute error of the algorithm after interpolation reach 0.047 and 0.043, respectively. Compared with the original GAIN network, the interpolation accuracy is improved by 58.31%, the interpolation speed is 2776 milliseconds, and the processing speed is improved by 19.34%. In multiple comparative tests, it is significantly better than existing classical methods.