Optimisation of reinforcement learning algorithm applied to dynamic network

Abstract

Mobile ad-hoc network represent a class of dynamic networks where nodes are non-stationary and the topology is changing rapidly. The routing algorithms for these types of networks play an important role. Traditional routing algorithms do not perform optimally under congestion and high-speed situations. Reinforcement learning-based routing algorithm SAMPLE holds a lot of potential to these types of problems but performs worse as compared to AODV and DSR under the Random Waypoint Mobility Model. To overcome the bottleneck of SAMPLE in terms of performance in a pure ad-hoc environment and enable its optimal performance under dynamic scenarios, we propose a mobility model to characterise group mobility under real city like scenario. The proposed model will be tested for its performance using suitable mobility metrics, verified statistically and then applied to a reinforcement learning algorithm SAMPLE to study its impact.