Task allocation based multi-agent reinforcement learning for LoRa nodes in gas wellhead monitoring service

Abstract

This paper investigates a new alternative approach to handle the tasks allocation problem that associate with numerous Long Range (LoRa) nodes in the High-Pressure High-Temperature (HPHT) gas wellhead monitoring service. A Multi-Agent Reinforcement Learning approach is proposed in this paper to overcome this problem with the Proximal Policy Optimization (PPO) is chosen as the policy gradient method. An action space is the spreading factor and other parameters such as frequency and transmission power has been kept constant. The reward function for the training process will be determined by two parameters which are the acknowledge flag (ACK) and collision between packets. Each node will be distributed across a defined disc radius. Each node will be represented as an agent. Each agent will undergo packet transmission and the packet will be evaluated according to the reward function. The results show that PPO with Multi Agent Reinforcement Learning was able to determine the optimal configuration for each LoRa node. The total reward value corresponds to the total number of nodes. Furthermore, since this study also implements the use of CUDA, the training was able to done in 200 steps and 45 minutes.