Renner, Bernd-ChristianBernd-ChristianRennerTurau, VolkerVolkerTurau2022-07-082022-07-082012-03Sustainable Computing: Informatics and Systems 2 (1) : 43-56 (2012-03)http://hdl.handle.net/11420/13088Forecasting the expected energy harvest enables small-sized energy-harvesting sensor nodes to schedule tasks or adapt the radio duty cycle. This ability ensures depletion-safe and efficient operation. Most energy sources exhibit cyclic patterns of intensity, e.g., the Sun. These patterns show periods with unequal - low versus high and stable versus varying - energy production and heavily depend on a node's location as well as seasonal and environmental changes. Existing forecast algorithms do not exploit these patterns, but create and update forecasts at static and arbitrary points in time, the main knob being the number of updates per cycle. We present a method enabling sensor nodes to adapt to harvesting patterns at runtime. It is designed for seamlessly replacing the static scheme to improve the accuracy of a wide range of existing forecast algorithms. In our evaluation, we show that (i) the adaptive method traces the energy pattern in real-world deployments accurately, (ii) reacts to seasonal and environmental changes, (iii) increases forecast accuracy, and (iv) reduces the number of prediction updates. These achievements enhance depletion-safe operation and efficient task scheduling with fewer recalculations and adjustments of the duty cycle. They also facilitate the exchange of harvesting forecasts for collaborative node tasks, since less information has to be shared.en2210-5379201214356Energy harvestingEnergy-intake predictionSustainable sensor networksJournal Article10.1016/j.suscom.2012.02.001Journal Article