Scholarlink

<p>The Internet of Things (IoT)-driven smart grid paradigm has greatly improved the reliability and flexibility of traditional power systems, with increasing IoT devices deployed for grid monitoring. To achieve reliable, low-latency communication under constrained spectrum and energy resources, we investigate a cognitive radio (CR) and non-orthogonal multiple access (NOMA) enabled smart grid scenario, which consists of an energy harvesting (EH) secondary device and multiple primary devices. We formulate a long-term throughput maximizing problem for the secondary device, and decompose it into two sub problems. A two-layered hybrid algorithm is proposed, which combines convex optimization with a Dual Attention Feedforward Policy network combined Deep Deterministic Policy Gradient (DAFPDDPG) method. Convex optimization is adopted to optimize transmit power and time allocation coefficients. By further introducing dual policy networks and attention mechanisms in Q networks, the DAFPDDPG scheme enables the secondary device to learn optimal decision-making policies efficiently. Simulation results demonstrate that the proposed DAFPDDPG outperforms baseline scheme including DDPG, random and greedy policies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A dual-attention-inspired DRL framework for throughput enhancement in smart grid

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月2日

查看全文
  • Jian Wu,
  • Junwei Ma

摘要

The Internet of Things (IoT)-driven smart grid paradigm has greatly improved the reliability and flexibility of traditional power systems, with increasing IoT devices deployed for grid monitoring. To achieve reliable, low-latency communication under constrained spectrum and energy resources, we investigate a cognitive radio (CR) and non-orthogonal multiple access (NOMA) enabled smart grid scenario, which consists of an energy harvesting (EH) secondary device and multiple primary devices. We formulate a long-term throughput maximizing problem for the secondary device, and decompose it into two sub problems. A two-layered hybrid algorithm is proposed, which combines convex optimization with a Dual Attention Feedforward Policy network combined Deep Deterministic Policy Gradient (DAFPDDPG) method. Convex optimization is adopted to optimize transmit power and time allocation coefficients. By further introducing dual policy networks and attention mechanisms in Q networks, the DAFPDDPG scheme enables the secondary device to learn optimal decision-making policies efficiently. Simulation results demonstrate that the proposed DAFPDDPG outperforms baseline scheme including DDPG, random and greedy policies.