Abstract: This paper presents a bi-level capacity planning method based on deep reinforcement learning to enhance the integration and synergy efficiency of rural integrated energy systems (RIES) against the background of carbon neutrality targets and multi-energy coupling. The upper level optimizes equipment capacities with the objective of minimizing the total economic cost over the planning horizon, while the lower layer minimizes daily operating costs across typical scenarios. To address the high-dimensional and non-convex nature of the planning problem, an efficient solution is achieved through the specific design of the agent state, action space, and reward function. Simulation results indicate that the DDPG algorithm reduces investment costs by 14.44% and carbon emissions by 3.83% compared with the DQN algorithm. Consequently, this research provides an economically efficient framework for the low-carbon transition of RIES.