Communication Optimization Method for Power Internet of Things Based on Multi-agent Deep Reinforcement Learning

Abstract

To solve the bottleneck of power Internet of Things terminals in remote areas in terms of communication interruption, energy depletion and insufficient computing power, a Space-Air-Ground Integrated Power Internet of Things communication and computing collaborative optimization framework is constructed, and a communication optimization method based on Multi-Agent Distributed Deep Deterministic Policy Gradient (MAD3PG) is proposed. Firstly, the energy replenishment and task offloading problem is modeled as a multi-agent Markov decision-making process, and the Actor-Critic structure of centralized training and distributed execution is used to jointly optimize cross-agent trajectories, power allocation, user association and other strategies. Secondly, a hybrid offloading strategy and a shared result cache pool are introduced to reduce repeated calculations, and wireless information and energy simultaneous transmission technology are combined to collaboratively execute task transmission and energy replenishment. The results showed that when the MAD3PG was deployed in a 10km² area, the average communication throughput reached 28.65Mbps, the minimum user rate remained at 4.12Mbps, the total system energy consumption was only 4.82kJ, the task offloading success rate reached 99.1%, and the edge computing resource utilization rate was 85.2%. The research method can provide a set of high-efficiency, low-energy consumption and adaptive communication optimization solutions for the communication-empowered power Internet of Things, which is of great value in ensuring the long-term stable operation of new power systems with new energy as the main body.