This website requires JavaScript.

Deep Reinforcement Learning for Wireless Scheduling in Distributed Networked Control

Wanchun LiuKang HuangDaniel E. QuevedoBranka VuceticYonghui Li
摘要
In the literature of transmission scheduling in wireless networked controlsystems (WNCSs) over shared wireless resources, most research works havefocused on partially distributed settings, i.e., where either the controllerand actuator, or the sensor and controller are co-located. To overcome thislimitation, the present work considers a fully distributed WNCS withdistributed plants, sensors, actuators and a controller, sharing a limitednumber of frequency channels. To overcome communication limitations, thecontroller schedules the transmissions and generates sequential predictivecommands for control. Using elements of stochastic systems theory, we derive asufficient stability condition of the WNCS, which is stated in terms of boththe control and communication system parameters. Once the condition issatisfied, there exists at least one stationary and deterministic schedulingpolicy that can stabilize all plants of the WNCS. By analyzing and representingthe per-step cost function of the WNCS in terms of a finite-length countablevector state, we formulate the optimal transmission scheduling problem into aMarkov decision process problem and develop a deep-reinforcement-learning-basedalgorithm for solving it. Numerical results show that the proposed algorithmsignificantly outperforms the benchmark policies.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答