Virtual University Journals
Search
Search
Route Optimization with Reinforcement Learning
Atif Manzoor
Abstract:
The Internet of Things is changing different industries to improve the interactions between devices to facilitate automation, monitoring, and analysis. However, routing within IoT networks is not a problem that can be easily solved due to some factors such as the dynamic nature of the networks, congestion, flooding and resource wastage. Efficient routing methods are necessary to guarantee connection availability, longer network duration, and rational resource usage. Reinforcement learning is one of the subfields of machine learning that can potentially address these challenges because it allows for learning in decision-making processes. In this regard, the flooding-controlled adaptive reinforcement learning-based route optimization model, called FARLRO, is introduced to mitigate the problem of network flooding as well as make the most of the routing decisions within the networks. The parameters such as residual energy level, available bandwidth, mobility pattern, traffic condition and topological arrangements are incorporated into the state space of the model, and it uses reinforcement learning to adapt the routing decisions. The Q-learning model continuously improves the state variables and optimizes routes to reduce the cases of flooding and enhance the network’s efficiency. It also uses the Bellman equation for assessing future rewards, thus making it a forward-looking method of route optimization. Extensive experiments have shown that the model provides significant improvements in several critical performance metrics, such as a smaller flooding ratio, a lower network congestion index, less frequent broadcast storms, a lower packet drop ratio resulting from flooding, an increased network lifetime, a higher Mobility Aware Packet Delivery Ratio, and higher Resource Utilization Efficiency. As compared to the conventional routing protocols, the proposed model outperforms various state-of-the-art ad adhoc routing schemes. Extensive experiments have been performed to show that the proposed model decreases the flooding ratio, less overhead. Both of the above parameters are critical to ensure longer network lifetime compared with the other approaches in high-density and high-mobility environments. Another advantage of the model is its effective stability in solving route optimisation problems in IoT networks, which provides a great improvement over traditional routing algorithms.
Keywords:
IoT, RL, FARLRO, ML
Full Text:
PDF