Abstract: To improve the performance of vibrating screening operations when the materials are dynamic feeding conditions, a novel horizontal attitude control method for vibrating screen surface is proposed using the double deep Q-Network (DDQN) deep reinforcement learning framework. Taking the vibrating screening of rice grains as the research object, a multi-degree-of-freedom vibrating screening dynamics model was established using the discrete element method (DEM), and the materials transportation on the screen surface and passing through the screen apertures processes were obtained through DEM simulations. The evaluation index for the uniformity of material distribution on the screen surface was proposed, and the correlations between the materials distribution state on the screen surface and the screening performance were obtained. By analysing the influence of the material feeding distribution state and the horizontal attitude angle of the screen surface on the screening performance, the reasonable range of the uniformity coefficient of the material distribution on the screen surface was determined. Some reasonable regions below the vibrating screen surface were determined for monitoring the materials passing through the screen apertures, and a BP neural network was established to achieve real-time prediction of the uniformity of materials distribution on the vibrating screen surface. After training, the established neural network demonstrated excellent convergence and generalization performance. Under the DDQN framework, the action space and state space of the vibrating screening operations were determined, a new reward function composed of dense and sparse terms was constructed. Then, the main hyper-parameters of the DDQN model were optimized, and a horizontal attitude control model of the vibrating screen surface was constructed. The established DDQN model was trained using the DEM simulation dataset, and the model demonstrated excellent convergence. The trained DDQN model was applied to the multi-degree-of-freedom hybrid vibrating screening control system, and the validation tests were carried out on a Multi-DOF vibration screening test rig under different materials feeding conditions. During the validation tests, the horizontal attitude angle of the screen surface was limited within a certain range. When the materials were fed onto the vibrating screen surface uniformly, the ideal horizontal attitude angle of the screen surface was 0°, therefore, there was no significant difference in loss rate between the proposed control method and the traditional method with a fixed horizontal attitude angle. When the material feeding coefficient increasing from 0 to 0.375, the loss rate under the fixed horizontal attitude angle increased rapidly from 0.59% to 0.98%. The proposed control method could adaptively adjust the horizontal attitude angle of the vibrating screen surface, thereby changing the movement direction of the materials on the screen surface. This can effectively improve the uniformity of materials distribution, and the loss rate only increased to approximately 0.68%. The loss rate decreased by approximately 30.6 %. The results of the comparative tests verified the feasibility and advantages of deep reinforcement learning in the control system design of the vibrating screen attitude. The proposed method provides an effective solution for increasing the materials screening performance under dynamic feeding conditions, and has broad application prospects in the field of intelligent control for agricultural screening equipment.