Abstract: To address the issues of insufficient prediction accuracy of tunnel boring machine (TBM) operation parameters and small-sample data dilemma,a joint prediction framework integrating deep learning and generative adversarial network (GAN) is proposed.Taking the TBM construction data of the Jilin Central City Water Diversion from Songhua River Project as the object,a binary discriminant function screened valid data segments.The box-plot method removed out-liers,and multi-source heterogeneous data were normalized by Z-Score.Three prediction models—deep neural network (DNN),long short-term memory network (LSTM),and convolutional neural network (CNN)—were constructed.Network parameters were optimized using stochastic gradient descent (SGD),adaptive learning rate AdaGrad,and Adam algorithms,with activation functions like ReLU and Leaky ReLU enhancing non-linear fitting.GAN-generated high-quality augmented data coupled with deep-learning models to form GAN-DNN,GAN-LSTM,and GAN-CNN architectures.Comparative analysis reveals that the GAN-CNN model excels in prediction accuracy (RMSE of 0.041,R-squared of 0.912) and training efficiency.Engineering verification in the Jilin Water Diversion and Qingdao Metro Projects shows an 18%~22% increase in tunneling efficiency,offering a reliable intelligent solution for TBM parameter optimization under complex geology.