Multi-agent based DRL with federated learning for data transmission in mobile sensor networks

Abstract

There has been a flurry of activity in the field of wireless sensor networks, or WSNs, as of late. Because packets need to be transported from source nodes to the destination nodes as quickly and energy effectively as feasible in various application areas, packet routing is one of core difficulties in WSNs. A plethora of routing options have been suggested to tackle this problem. The proposed method distributed and designed to run on a network of interconnected routers. Different from most of its competitors, the proposed results frame the routing problem as a reinforcement learning problem with several agents. To optimize more complicated cost functions, such as the time it takes for bags to be delivered and the amount of energy used in a baggage handling system, it is possible to model every router as a deep neural network. The proposed MA-DRL attains latency of 2.41, energy consumption of 26J has superior efficiency compared to the existing methods. However, the MA-DRL has minimized latency and lower energy consumption. This way, each router may take into consideration different types of data about its surroundings. Based on four metrics latency, and energy consumption the simulation results show that this architecture performs well.

Keywords

• Multi-agent learning
• deep reinforcement learning
• wireless sensor networks
• packet routing