A Deep Learning-Based Intelligent Routing Protocol for Energy-Efficient Wireless Sensor Networks

One of the most important challenges in designing a Wireless Sensor Network (WSN) is to make a efficient routing protocol, given the limited battery power, topology, and unreliability of wireless links. The conventional routing protocols, like LEACH and PEGASIS, assume static clustering rules, which are heuristic and hence are unable to cope with varying traffic loads, mobility of the nodes, and the imbalance of energy availability. In this paper, an Adaptive Next-hop Selection & Cluster-head Selection (ANHSCS) protocol based on Deep Learning (DL) is proposed to optimize the selection of the next-hop and Cluster-head for the route in real-time and energy efficiently using a Long Short-Term Memory (LSTM) encoder and a Deep Q-Network (DQN) policy head. The LSTM component extracts any temporal energy consumption and traffic load patterns in the residual energy and traffic loads, and the DQN agent learns a routing policy that simultaneously reduces energy consumption, end-to-end delay and packet loss via an online reward mechanism. A simulation-based evaluation is performed that compares DL-RP with LEACH, PEGASIS and a shallow machine-learning routing baseline (SVM/Random Forest). The results show that the lifetime of the network is increased, cumulative energy dissipation is decreased, and network packet delivery ratio is maintained at a higher level during dynamic environment with using DL-RP. The paper also explores computational complexity, scalability and open research areas for using deep learning routing agents on resource-limited sensor nodes.