Throughput and Energy Efficient Routing using Reinforcement Learning for Wireless Sensor Networks

In this work, Throughput and Energy Efficient Routing using Reinforcement Learning (TEER-RL) strategy for Clustering under Noisy Conditions (RLC-NC) that considers the weighted probabilities of the Residual Energy (RE) and the Packet loss (PL) as the parameters under noisy conditions is proposed. The model is tested for its throughput for different number of transmitted packets under varying energy and packet loss conditions. This model differs from most of the earlier models that consider the Residual Energy and distance in hops or other factors as separate independent weighted parameters. Also, these models assume perfect conditions and hence do not consider the packet loss. However, the proposed model is more realistic in that the packet loss, as well as the residual energy is affected by the number of transmitted packets. An initial permissible packet loss ratio is defined for the nodes. Packets are transmitted only if the packet loss ratio of a node is less than a preset threshold and when the residual energy of a transmitting node reaches zero, it is added to the list of dead nodes. It was observed that the throughput converges to a global maximum for a particular combination of weights for RE and PL using Reinforcement Learning with varying number of nodes. In summary, the proposed model maximizes the network lifetime by choosing the next node for transmission that maximizes the throughput while considering its Residual Energy, in contrast to existing models that generally choose the next node simply based on the distance (hop count) to the cluster head and the Residual Energy. For evaluation, the proposed method is compared with an existing Energy-Efficient Routing based on Reinforcement Learning (EER-RL). Experimentation has been carried out with varying number of nodes, for different values of weights p and q. For n=100 nodes, with p=0.1, q=0.9, the average consumed energy over 500 rounds for the proposed approach is 0.040567J as compared to 0.041389J in EER-RL. The average number of live nodes was observed to be 32 for the proposed scheme when compared to the 28 in EER-RL. It was observed that with increasing number of nodes, the proposed method performs better compared to the existing scheme, in terms of network life time, as well as the energy consumed.