Leveraging LSTM and Reinforcement Learning for Adaptive Sensing in CIoT Nodes

Wireless sensor networks (WSNs) offer a broad range of applications in the Consumer Internet of Things (CIoT). The sensor nodes in a WSN are equipped with an array of sensors that often encounter limited energy availability. Thus, a joint Long Short-Term Memory (LSTM) and reinforcement-learning-based edge intelligence framework is proposed in this article for a multi-sensing node. This novel strategy aims to estimate an optimal set of active sensors during a measurement cycle by solving the trade-off between the cross-correlation between sensing signals and sensors’ energy consumption using a Q-learning-based optimization function at the edge node. A prediction model based on LSTM is employed to predict the sensing signals monitored by the inactive sensors from the cross-correlated sensing signals monitored by the active sensors. To assess the performance of the proposed framework in CIoT nodes, the algorithm is simulated on an air pollution monitoring dataset. The simulation results confirm the effectiveness and efficiency of the proposed framework. In comparison to the current state-of-the-art approach, the proposed algorithm shows a 13% improvement in error performance and up to 27% improvement in sensing energy consumption, while maintaining a lower bound of cross-correlation coefficient between the inactive and active sensor set.