Extending Wireless Sensor Networks’ Lifetimes Using Deep Reinforcement Learning in a Software-Defined Network Architecture

Routing packets in a Wireless Sensor Network (WSN) is a challenging task, according to the limited resources available on the nodes of these networks, especially their energy sources. The use of Machine Learning (ML) techniques in a Software-Defined Network (SDN) topology has shown a good potential toward solving such a complex task. However, existing techniques emphasize finding the shortest paths to deliver the packets, which can overload certain nodes in the network, depending on their positioning. In this study, a new method is proposed to extend the lifetime of the WSN by balancing the loading on the nodes, using a Deep Reinforcement Learning (DRL) approach. By emphasizing on the lifetime of the network, the proposed method has been able to discover and use alternative routes to deliver the packets, avoiding the use of nodes with low energy. Hence, the average number of hops the packets travel through has been increased but the time required for the first node to exhaust its energy has been significantly increased.