Machine Learning-Based Energy-Saving Framework for Environmental States-Adaptive Wireless Sensor Network

In this paper, we propose an energy-saving framework for Wireless Sensor Networks (WSN) using machine learning techniques and meta-heuristics according to environmental states. Unlike conventional topology-based energy-saving methods, we focus on the energy savings of the sensor node in the WSN itself. We attempt two-phase energy savings on the sensor nodes. First, network-level energy saving, called N1-energy saving, is achieved by finding the minimum sensor nodes needed to ensure the performance of the WSN. To find the minimum sensor nodes, we apply hybrid filter-wrapper feature selection, a typical machine learning method, to find the best feature subsets. Second, we achieve energy savings of the WSNs by manipulating the sampling rate and the transmission interval of the sensor nodes to achieve node-level energy saving, which is referred to as N2-energy saving. To do so, we propose an optimization method based on Simulated Annealing (SA), which is an efficient method that can find the approximate global optimum in datasets where it is difficult to collect precise values due to noise problems, such as sensor data. Some numerical examples are shown with respect to several control parameters. We conduct several experiments with real-world sensor data in a smart home to prove the superiority of the proposed method. Through these experiments, the sensor nodes are shown to be selected by a method performing N1-energy savings effectively while minimizing the loss of performance compared to the original WSN. In addition, we demonstrate that N2-energy savings can be achieved while maintaining the QoS of the WSN through an optimal sampling rate and transmission interval determined by the SA.