Shield load is an important parameter of shield machine, and accurate prediction of shield load is very important to ensure the safe construction of shield tunnel. In this paper, a new load prediction model (CGA), combining convolutional neural network (CNN), gate recurrent unit neural network (GRU) and attention mechanism (Attention), is proposed based on the shield machine cross existing station at close range. The CNN-Attention model is first used to extract the high-dimensional spatial features of the data and distinguish the importance of different features. Then the GRU model is used to extract the temporal characteristics of the data, followed by the attention mechanism to extract the important time node information. Finally, the prediction results are obtained. To verify the prediction performance of the proposed model, four existing algorithms are selected for comparison. The results show that the proposed model in this paper outperforms other models in three evaluation metrics, and the proposed model can also provide reference for predicting researches on shield tunneling tool wear, surface and structural deformation, etc.