An Attention based Recurrent Neural Network Model for Link Quality based Path Selection in Wireless Sensor Networks

Path selection is important for enhcning the network survivability and the energy efficiency of Wireless Sensor Networks (WSNs). As the number of wireless nodes in WSNs continues to increase at a fast rate, the volume of data sent over the network grows in tandem, making routes selection crucial for conserving power in WSNs. However, these techniques assume that the communication link between the nodes is reliable. In practice, communication links can be unreliable and not only do they not have a constant delay but they also have a variable delay. This research proposes a deep learning model for path selection driven by the identification of reliable links. The proposed model called Link Quality Prediction guided Path Selection (LQP-PS) is implemented with a stacked Recurrent Neural Network (RNN) with an attention layer. The model's Area Under Curve (AUC) and precision for link prediction are the highest among 6 publicly available datasets, coming in at 0.9941, 0.9931, and 0.9938, respectively. It's worth noting that the Outage Probability (OP) is rather low for this scenario.