Scholarlink

<p>The emergence of novel application scenarios within the Industrial Internet of Things (IIoT) necessitates a reevaluation of resource distribution strategies within Computility Networks. Existing researches show that allocating computational resources in alignment with user intent has been demonstrated to improve the efficiency of resource distribution significantly. Existing research fails to adequately address the dynamics and heterogeneity of nodes, as well as the ambiguity of user intent in Computility Networks, leading to suboptimal resource allocation results. To address these challenges, this research introduces a novel Soft Actor-Critic Long Short-Term Memory (SAC-LSTM) framework, which seamlessly translates user intents into concrete resource allocation strategies. Specifically, the proposed framework integrates natural language processing (NLP) techniques for intent modeling with deep reinforcement learning (DRL) for decision-making, enabling an end-to-end optimization of intent-driven resource allocation. To mitigate the scarcity of sufficiently annotated corpora in various scenarios, we employ large-scale unsupervised corpora for pre-training. The experiments are conducted on the EdgeCloudSim simulation platform. For intent detection, the alpha and CAMPI datasets are utilized. Resource allocation experiments are carried out within both the EdgeCloudSim and Mininet simulation environments. Experimental results indicate that the proposed framework not only accurately identifies user intents but also showcases outstanding performance in managing resource allocation tasks within both continuous and discrete action spaces. The efficacy of the framework is corroborated by tests on diverse datasets, which confirm its potential to optimize resource distribution in the dynamic IIoT environment.</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

Sac-lstm: optimal resource allocation based on user intent in computility networks

  • 期刊论文

  • 发表日期：

    2025年8月15日

查看全文
  • Yingying Zheng,
  • Ningjiang Chen,
  • Yin Yin,
  • Zizhan Huang,
  • Jingwei Wang,
  • Weijing Wang,
  • Xinghui Gan

摘要

The emergence of novel application scenarios within the Industrial Internet of Things (IIoT) necessitates a reevaluation of resource distribution strategies within Computility Networks. Existing researches show that allocating computational resources in alignment with user intent has been demonstrated to improve the efficiency of resource distribution significantly. Existing research fails to adequately address the dynamics and heterogeneity of nodes, as well as the ambiguity of user intent in Computility Networks, leading to suboptimal resource allocation results. To address these challenges, this research introduces a novel Soft Actor-Critic Long Short-Term Memory (SAC-LSTM) framework, which seamlessly translates user intents into concrete resource allocation strategies. Specifically, the proposed framework integrates natural language processing (NLP) techniques for intent modeling with deep reinforcement learning (DRL) for decision-making, enabling an end-to-end optimization of intent-driven resource allocation. To mitigate the scarcity of sufficiently annotated corpora in various scenarios, we employ large-scale unsupervised corpora for pre-training. The experiments are conducted on the EdgeCloudSim simulation platform. For intent detection, the alpha and CAMPI datasets are utilized. Resource allocation experiments are carried out within both the EdgeCloudSim and Mininet simulation environments. Experimental results indicate that the proposed framework not only accurately identifies user intents but also showcases outstanding performance in managing resource allocation tasks within both continuous and discrete action spaces. The efficacy of the framework is corroborated by tests on diverse datasets, which confirm its potential to optimize resource distribution in the dynamic IIoT environment.