Solar Harvest Prediction Supported by Cloud Cover Forecasts

Solar harvest prediction is used in energy-harvesting sensor networks to achieve perpetual node operation. Existing approaches only exploit local knowledge and thus fail in unforeseeable, changing weather conditions. We investigate the benefit of incorporating global knowledge in terms of fractional sky cloudiness, so-called cloud cover. We propose and evaluate two methods that combine local information of a node's harvest pattern with global cloud cover forecasts. We evaluate their performance with solar traces collected by three solar-harvesting sensor nodes and compare the results with existing prediction algorithms. We find that (i) harvest predictions using cloud cover forecasts improve overall prediction precision, (ii) prediction errors in changing weather conditions are considerably reduced, and (iii) coarse-grained cloud cover forecasts require low extra network traffic while sacrificing little prediction precision.