Dynamic collaborative optimization of end-to-end delay and power consumption in wireless sensor networks for smart distribution grids

Wireless sensor network (WSN) technology is poised to be widely adopted in the smart distribution grid (SDG), which has strict requirements regarding communication delays. However, delays in WSNs are easily affected by dynamic interference factors (such as channel access competition, transmitting power, and node failure), and these dynamic characteristics make the traditional offline optimization methods unsuitable. Besides, the reinforcement learning (RL) based online optimization methods have dimension explosion and convergence problems. In this paper, we propose a dynamic collaborative optimization of the end-to-end delay and power consumption of the WSNs based on grouped RL. In particular, we first build an environment model for evaluating the values of the optimization objective. Those values are used to calculate the rewards for the RL algorithm. To accelerate the convergence of RL swamped by the dimensions of the action space, a novel grouped RL is proposed. Then iterative learning is performed to balance the end-to-end delay and power consumption by adjusting the transmitting power of each node. The simulation results show that the proposed algorithm is able to meet the SDG delay requirements with low power consumption when the communication is dynamically affected. The developed algorithm achieves a maximum end-to-end delay reduction of 20.3% and a computational cost reduction of 6.2% to 52.7% compared with the other two RL algorithms.