Author(s)

Kaili Shao, Jifeng Qin

Corresponding Author

Jifeng Qin

Abstract

With the continuous development of current internet technology, the requirements for wireless sensor network signal performance are gradually increasing. Affected by external and internal noise factors, sensor network signals are prone to significant fluctuations, resulting in reduced stability and weakened network performance. Therefore, this article conducts research on the method of constructing region sensitive models based on deep RLnetworks, improving the convergence speed of the network, reducing routing load, and providing routing control methods for each connection. RL(Reinforcement Learning) refers to learning a suitable set of actions for selection, with the aim of maximizing the reward value. Given the continuous interaction with the external environment, the algorithm proposed in this article is based on a set of RLagents to learn the most suitable spectrum allocation scheme, in order to maximize the given reward in each access network. The research results indicate that the algorithm proposed in this paper reduces the scale of long convergence events in the network. The convergence times of AGA (Adaptive Genetic Algorithm) and AR (Augmented Reality) are both very short, generally around 0.58 seconds. However, the convergence time of the algorithm proposed in this paper is the shortest, only around 0.43 seconds. The experimental results verify that the algorithm proposed in this paper effectively improves the efficiency of network convergence time.

Keywords

Deep reinforcement learning; Network construction; Regional sensitivity; Model