Load forecasting-based congestion control algorithm for delay-Tolerant networks

Delay-Tolerant Networks (DTN) use 'carrystorage-forward' mechanism to deal with high latency and frequent interruption in the extreme network environment. It requires sufficient storage of nodes. As network load increases, buffer overload and network congestion of some essential nodes have been resistance to DTN development. For the issue of congestion control in DTN, this article proposes a congestion control algorithm based on load forecasting. As traffic conditions are complex and non-linear, we selected back propagation neural networks to predict the future load. First, a forecast function is trained based on the historical buffer information. And then forecast function is used to predict short-Term future buffer occupancy. Finally, the prediction result is broadcasted to surrounding nodes. When neighbor nodes have data to transmit, buffer occupancy forecast serves as reference to avoid that packets are transmitted to congestion nodes, thus reducing traffic input when node's buffer is almost full. Simulation results indicate that this algorithm performs well according to network indicators such as buffer occupancy and delivery ratio, which means it can effectively alleviate network congestion and improve network efficiency.