As an important component of the machining system, the influence of fixtures on the machining deformation of the workpiece cannot be ignored. By controlling the clamping force during the machining process is an effective means to suppress or improve the machining deformation. However, due to the dynamic coupling of part geometry, clamping method, manufacturing process and time-varying cutting forces, it is difficult to obtain accurate clamping forces, which hinders the realization of fixture-based deformation control. In this paper, the variation of clamping force is considered as the response of the joint action of cutting force and other working conditions in spatial and temporal terms, and a clamping force prediction method based on deep spatio-temporal network is proposed. The part geometry model is first parameterized based on voxels, after which the cutting forces are dynamically correlated with the clamping forces in spatial and temporal terms. Then, a convolutional network was designed to capture the spatial correlation between the working conditions such as cutting force and clamping force, and a gated recurrent cell network to capture the temporal correlation to predict the clamping force during machining. Finally, an experiment of milling a cylindrical thin-walled part illustrates the effectiveness of the proposed method.