A Data Transmission Path Optimization Protocol for Heterogeneous Wireless Sensor Networks Based on Deep Reinforcement Learning

Wireless sensor networks had become a hot research topic in Information science because of their ability to collect and process target information periodically in a harsh or remote environment. However, wireless sensor networks were inherently limited in various software and hardware resources, especially the lack of energy resources, which is the biggest bottleneck restricting their further development. A large amount of research had been conducted to implement various optimization techniques for the problem of data transmission path selection in homogeneous wireless sensor networks. However, there is still great room for improvement in the optimization of data transmission path selection in heterogeneous wireless sensor networks (HWSNs). This paper proposes a data transmission path selection (HDQNs) protocol based on Deep reinforcement learning. In order to solve the energy consumption balance problem of heterogeneous nodes in the data transmission path selection process of HWSNs and shorten the communication distance from nodes to convergence, the protocol proposes a data collection algorithm based on Deep reinforcement learning DQN. The algorithm uses energy heterogeneous super nodes as AGent to take a series of actions against different states of HWSNs and obtain corresponding rewards to find the best data collection route. Simulation analysis shows that the HDQN protocol outperforms mainstream HWSN data transmission path selection protocols such as DEEC and SEP in key performance indicators such as overall energy efficiency, network lifetime, and system robustness.