The Jurassic tight channel sand bodies are widely developed in the western Sichuan area of the Sichuan Basin, which is an important area for tight gas exploration and development. The seismic prediction of gas-bearing sandstone reservoir has always been a key and difficult point in the development process. In the paper, the reservoir prediction technology based on deep learning is adopted to realize the quantitative prediction of gas saturation of high-quality sandstone reservoir using data from the western Sichuan area. The workflow includes logging data correction based on deep feedforward neural network (DFNN), seismic processing optimization focused on AVA-preserving, and pre-stack DFNN gas saturation quantitative prediction. Through this workflow, the AVA characteristics of seismic data can be maintained to the maximum extent, the optimization of well logging curves can be quickly realized, the quality of seismic pre-stack inversion can be improved, and a better gas saturation deep learning model can be established, so as to obtain good results of gas saturation seismic prediction. The practical application effect in the western Sichuan area shows that the gas saturation blind well coincidence rate can reach more than 80%, and the post-well testing prediction error is within 5%. This is the first time to carry out the gas saturation quantitative prediction for channel sand bodies based on DFNN in the Sichuan Basin. The effect is good and this workflow could have a good application prospect in the region.