DRL-Based Adaptive Routing for Energy-Efficient and Reliable Underwater Wireless Sensor Networks

Abstract The amount of energy lost in underwater wireless sensor networks is greatly influenced by the choice of relay: a node that is frequently chosen might end up with an empty battery, and a weak acoustic link can cause retransmissions to be needed. Routing has to take into account both the conservation of relay energy and the need to make progress, ensure timely forwarding, and keep a viable next node available. The present study focuses on developing an Adaptive Routing method based on Deep Reinforcement Learning for energy efficient and reliable Underwater Wireless Sensor Networks. Prior to carrying out each transmission, the current node deletes from consideration any of its neighbors, that have insufficient residual energy, poor link quality, or fail to make any positive contribution to depth advancement towards the surface sink. The DRL policy then assesses the remaining possible relays according to their residual energy, SNR, depth advancement, forwarding delay, and void risk. Feedback on delivery updates the action values, and both link failures and voids are penalized. Testing in NS-3, using Aqua-Sim included nine baseline protocols and five different deployment scenarios. In a number of repeated trials involving 100 nodes each, the proposed approach obtained a mean PDR of 93.26%, a mean residual energy of 87.732%, and a mean delay of 0.982 s. ADRP-DQL achieved a PDR of 85.70% and a delay of 1.162 s, while Dynamic DRL maintained 81.029% residual energy. The proposed method was able to keep the network running for 662.6 rounds, as opposed to 625 rounds for Dynamic DRL. When adaptive reward balancing was not used, the network lifetime dropped to 560 rounds.