Efficient Data Collection and Event Boundary Detection in Wireless Sensor Networks Using Tiny Models
Using wireless geosensor networks (WGSN), sensor nodes often monitor a phenomenon that is both continuous in time and space. However, sensor nodes take discrete samples, and an analytical framework inside or outside the WSN is used to analyze the phenomenon. In both cases, expensive communication is used to stream a large number of data samples to other nodes and to the base station. In this work, we explore a novel alternative that utilizes predictive process knowledge of the observed phenomena to minimize upstream communication. Often, observed phenomena adhere to a process with predictable behavior over time. We present a strategy for developing and running so-called ’tiny models’ on individual sensor nodes that capture the predictable behavior of the phenomenon; nodes now only communicate when unexpected events are observed. Using multiple simulations, we demonstrate that a significant percentage of messages can be reduced during data collection.
KeywordsSensors wireless sensor network model continuous phenomenon tiny models process modeling prediction autonomous
Unable to display preview. Download preview PDF.