Scholarlink

<p>A reinforcement learning-based energy–security co-optimization strategy for clusters of MANET is designed in this paper. Currently available routing strategies aim to optimize energy efficiency or security but ignore other factors, resulting in poor performance in dynamic environments. In order to fill the gap, this work designs a novel constrained Markov Decision Process formulation in which energy efficiency, security performance, throughput, and delay are balanced via Q-learning algorithm used in the cluster heads. In addition, the method uses dynamic cryptographic scaling and dynamic key management techniques according to residual energy, traffic loads, and the intrusion levels. The simulations carried out using MATLAB environment and involving 1000 nodes show that this approach ensures rapid convergence within 100 iterations, lowering the overall energy expenditure by about 35%. In comparison with the existing approaches under similar simulation conditions, whereby the reduction is based on the interval average of the total energy across all simulation stages. It provides up to 15% higher security performance than hybrid clustering or deep reinforcement learning routing strategy. Furthermore, the approach proves robust to ± 15% changes in the parameters and works equally well in various mobility scenarios. However, the approach assumes consistent cluster-head availability and is evaluated under controlled simulation scale constraints.</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

Reinforcement Learning-based Energy–security Co-optimization in Clustered Manets

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • R. Mohandas,
  • A. Emmanuel Peo Mariadas,
  • Elangovan Muniyandy,
  • Ayman A. AbuBaker

摘要

A reinforcement learning-based energy–security co-optimization strategy for clusters of MANET is designed in this paper. Currently available routing strategies aim to optimize energy efficiency or security but ignore other factors, resulting in poor performance in dynamic environments. In order to fill the gap, this work designs a novel constrained Markov Decision Process formulation in which energy efficiency, security performance, throughput, and delay are balanced via Q-learning algorithm used in the cluster heads. In addition, the method uses dynamic cryptographic scaling and dynamic key management techniques according to residual energy, traffic loads, and the intrusion levels. The simulations carried out using MATLAB environment and involving 1000 nodes show that this approach ensures rapid convergence within 100 iterations, lowering the overall energy expenditure by about 35%. In comparison with the existing approaches under similar simulation conditions, whereby the reduction is based on the interval average of the total energy across all simulation stages. It provides up to 15% higher security performance than hybrid clustering or deep reinforcement learning routing strategy. Furthermore, the approach proves robust to ± 15% changes in the parameters and works equally well in various mobility scenarios. However, the approach assumes consistent cluster-head availability and is evaluated under controlled simulation scale constraints.