农村微电网作为支撑“双碳”目标落地、赋能乡村振兴战略实施的重要载体，通过整合分布式能源与传统发电资源，能够有效提升系统整体的运行效率。为此，提出了一种基于改进双延迟深度确定性策略梯度（twin delayed deep deterministic policy gradient，TD3）的农村微电网源网荷储优化运行策略。首先，综合考虑运行成本和可再生能源利用率，构建了含沼气的农村微电网源网荷储优化运行模型。其次，构建了基于复合评分机制的注意力加权评论家网络（CASA-Critic），动态提取并聚焦关键状态特征以提升价值网络的评估精度。然后，采用动态感知策略，基于训练阶段在时间进程与环境状态的不确定性特征动态调节动作探索强度，改善算法在探索与利用之间的平衡。最后，基于典型的农村微网源网荷储运行场景进行仿真分析，仿真结果表明，相较于其他基于深度强化学习的优化运行策略，所提方法具有更好的收敛性和寻优效果，在提升系统运行经济性、促进可再生能源消纳方面具有明显的优越性。

As a crucial carrier for supporting the implementation of the "dual carbon" goals and empowering rural revitalization, rural microgrids can effectively enhance overall system operational efficiency by integrating distributed energy with traditional power generation resources. To this end, this paper proposes a optimal operation strategy for source-grid-load-storage in rural microgrids based on an improved twin delayed deep deterministic policy gradient (TD3) algorithm. Firstly, an optimal operation model for rural microgrids incorporating biogas is constructed by comprehensively considering operating costs and renewable energy utilization. Secondly, a compound assessment scored attention-weighted critic(CASA-Critic)is designed to dynamically extract and focus on key state features, thereby enhancing the evaluation accuracy of the value network. Then, a dynamic perception strategy is introduced to adaptively adjust the intensity of action exploration based on the temporal progression and environmental state uncertainties during training, improving the balance between exploration and exploitation. Finally, simulation analysis is conducted based on a typical rural microgrids source-grid-load-storage operation scenario. The results demonstrate that, compared to optimization strategies based on other deep reinforcement learning, the proposed method exhibits better convergence and optimization performance, while also showing significant superiority in improving system operational economy and promoting renewable energy accommodation.