Chamber pressure is a key factor in the core issues of the stress state of equipment and the stability of tunnel face during shield tunneling. The chamber pressure exhibits significant spatial variability, and its formation and evolution originate from the complex coupling effect of geotechnics and mechanism, which is related to multiple parameters such as geological features and tunnelling parameters. Yet, the spatial distribution features or geological features is generally ignored in existing methods for predicting chamber pressure. To probe this problem, this paper proposes a method to predict the chamber pressure field in shield machines based on the deep learning algorithm guided by physical feature function of spatial distribution. This method constructs physical feature function for decoupling the spatial distribution feature of chamber pressure, uses convolutional neural network and gated recurrent unit to extract the spatial features of the historical information of multi-source parameters and the temporal features of feature coefficient, respectively, and combines the real-time information of multi-source parameters to predict the feature coefficient, so as to realize the prediction of chamber pressure field. Taking a section of Changsha Metro Line 4 as a case study, this method was used to accurately predict the measured spatial distribution of chamber pressure with an accuracy of 0.98, which verified the effectiveness of the proposed method. Sensitivity analysis reveals that the main sensitive parameter of the spatial distribution feature coefficient of chamber pressure are basically the same in different strata, but their sensitivities vary significantly with the geological conditions of strata. The result can provide guidance for the refined control of chamber pressure of shield machines in complex strata.