The traditional geological method relies too much on the resolution of seismic reflection and the quality of well data for the determination of geological profiles. In view of the fact that the number of well data that can be used for calibration in the early stage of development is very small, and the traditional geological modeling method generates geological profiles. The efficiency is low and it is difficult to support model establishment and frequent updating. A geological profile generation method based on improved Pix2Pix network was proposed. Firstly, the initial three-dimensional data was sliced. Based on the comprehensive analysis of deep learning network, a Pix2Pix network model based on residual and multi-scale discriminator was constructed. The residual mechanism was introduced in the generator part to improve the learning ability of the network to geological features, and a multi-scale discriminator was set for the model to enhance the discriminant performance of the network. The real seismic reflection data and geological profile data of the oilfield were used to train the model. The experimental results show that the performance of the network model is significantly improved after the introduction of residual mechanism and multi-scale discriminator. The SSIM (structural similarity) score of the generated results and the real geological profile can reach 91.89 %, and the geological features in the generated results are highly fitted with the actual situation.