Scholarlink

Critical Infrastructure Wireless Sensor Networks (CI-WSNs) are used to provide critical events in contexts of industrial automation, water distribution networks, and smart grids. There is a need to feel changes over time in a steady, dependable way to ensure operational security. However, the energy availability of the sensor nodes in such networks is the most critical, and thus synchronising node wake-ups is a crucial factor in the long-term sustainability of the network. Traditional approaches to scheduling have been based on either static duty cycling or centralised control dynamics, which cannot scale to dynamic conditions with changing event criticality and random node failures. The outcome of these restrictions is ineffective energy use, reduced network lifespan, and the risk of missing important events, thereby jeopardising the safety of critical infrastructure. We aim to address these challenges in the system by proposing a Distributed Reinforcement Learning-based Multi-Agent Scheduling mechanism, DReaM-Sched, that enables sensor nodes to determine optimal wake-up and sleep schedules individually. A node is an intelligent agent with imperfect knowledge of its environment and learns an adaptive scheduling policy via multi-agent reinforcement learning (MARL). To solve this, an interaction reward mechanism that balances energy efficiency, sensing accuracy, and network resiliency will be used. A salient point about the nodes is that, by simply sharing little information with neighbours, they can make local decisions that result in global optimisation, without a centralised controller. In addition, a hierarchical meta-controller implemented at critical nodes refines network-wide policies to maintain high reliability even under adversarial conditions, such as jamming or partial node failures. Widespread simulations of CI-WSN smart grid scenarios have shown that DReaM-Sched can increase network lifetime by 38% compared to static duty cycling, while achieving 97% event detection accuracy. It reduces the average energy usage per node to 31% and maintains more than 90% of performance even with 10% of nodes failing. The results show that distributed learning-based scheduling is an energy-efficient, resilient, and highly scalable approach for deploying mission-critical wireless sensor networks.

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

Distributed Reinforcement Learning for Optimised Node Wake-up Scheduling in Critical Infrastructure WSNs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ashu Nayak,
  • Debarghya Biswas

摘要

Critical Infrastructure Wireless Sensor Networks (CI-WSNs) are used to provide critical events in contexts of industrial automation, water distribution networks, and smart grids. There is a need to feel changes over time in a steady, dependable way to ensure operational security. However, the energy availability of the sensor nodes in such networks is the most critical, and thus synchronising node wake-ups is a crucial factor in the long-term sustainability of the network. Traditional approaches to scheduling have been based on either static duty cycling or centralised control dynamics, which cannot scale to dynamic conditions with changing event criticality and random node failures. The outcome of these restrictions is ineffective energy use, reduced network lifespan, and the risk of missing important events, thereby jeopardising the safety of critical infrastructure. We aim to address these challenges in the system by proposing a Distributed Reinforcement Learning-based Multi-Agent Scheduling mechanism, DReaM-Sched, that enables sensor nodes to determine optimal wake-up and sleep schedules individually. A node is an intelligent agent with imperfect knowledge of its environment and learns an adaptive scheduling policy via multi-agent reinforcement learning (MARL). To solve this, an interaction reward mechanism that balances energy efficiency, sensing accuracy, and network resiliency will be used. A salient point about the nodes is that, by simply sharing little information with neighbours, they can make local decisions that result in global optimisation, without a centralised controller. In addition, a hierarchical meta-controller implemented at critical nodes refines network-wide policies to maintain high reliability even under adversarial conditions, such as jamming or partial node failures. Widespread simulations of CI-WSN smart grid scenarios have shown that DReaM-Sched can increase network lifetime by 38% compared to static duty cycling, while achieving 97% event detection accuracy. It reduces the average energy usage per node to 31% and maintains more than 90% of performance even with 10% of nodes failing. The results show that distributed learning-based scheduling is an energy-efficient, resilient, and highly scalable approach for deploying mission-critical wireless sensor networks.