WSN中基于强化学习的能效优化任务处理机制

Energy Efficiency Optimization Task Processing Mechanism Based on Reinforcement Learning in WSN

以提高无线传感器网络中任务处理的能效为目标,提出了一种近似最优化的任务处理机制,无线传感器节点可根据任务缓存区的任务数量、信道条件,动态地实现任务向边缘服务器的卸载以及本地处理。将任务处理机制建模为马尔可夫决策过程,因为无线传感器节点不知道此过程的状态转移概率,所以采用A3C算法以实现在环境参数未知情况下的探索和学习,从而得到近似最优的任务处理策略。仿真结果表明,与其他机制相比,所提任务处理机制能提高节点能效,且收敛速度更快。

Aiming at improving the energy efficiency of task processing in wireless sensor networks,a nearly optimal task processing mechanism is proposed,in which wireless sensor nodes can dynamically unload tasks to edge servers and perform local processing according to the number of tasks in the task cache and channel conditions.The task processing mechanism is modeled as a Markov decision process.Since the wireless sensor node does not know the state transition probability of this process,the A3C algorithm is used to realize exploration and learning under unknown environmental parameters,so as to obtain the approximately optimal task processing strategy.Under certain buffer conditions and channel conditions,the optimal task quantity,modulation level and transmission power are selected by this strategy,and the average task processing energy efficiency is improved.Simulation results show that compared with other mechanisms,the proposed task-processing mechanism can improve node energy efficiency and has faster convergence speed.