Feature Selection Framework for Multi-Source Energy Harvesting Wireless Sensor Networks

Abstract

Energy harvesting technologies are constantly evolving to help power sensor network nodes. Ranging from miniature power solar panels to micro wind turbines, nodes still express a deep need to harvest energies in order to keep both good performance level and energy autonomy. Recently, the simultaneous use of multiple sources has been proposed to tackle the time-varying characteristics of certain sources that can induce energy scarcity period and thus alter the node performance. In this context, this paper presents a methodology aimed at classifying the energy sources to choose the most efficient energy manager. As sensor nodes are embedded devices, it is necessary to ensure a balance between computational effort and classification accuracy. Feature extraction and selection phases can be processed and analyzed offline before deployment, and only a subset of features will be needed by the nodes to achieve efficient energy management. Simulations on real energy traces show that the proposed approach achieves classification accuracy higher than 95% through the computation of 4 features only.

Energy Conservation in Wireless Sensor Networks Using Partly-Informed Sparse Autoencoder

“…ML enables the selection of dominant set of features, which have the greater impact on inferring the network performance based on historical data patterns [18]. Over the last few decades, research efforts have focused on strategies that provides solutions to various challenges in WSN, such as routing [16], link-quality estimation [18], energy harvesting [19], link reliability prediction [9], just to mention a few.…”