Estimate Information Fusion Weight of WSNs Nodes Based on Truth Discovery Optimization Method Among Conflicting Sources of Data

Abstract

In practical wireless sensor networks (WSNs)-based applications, often only a few nodes are active while most of the others are activated occasionally due to limited resource of the WSNs. Thus most sensors only provide a few observations and only a few sensors make many observations, which often cause long-tail issue to undermine information fusion performance. So we present a confidence-aware information fusion scheme named CAIF to solve such a problem. In particular, we first make a quantitative study on the long-tail data phenomenon and the relationship between node-target distance and node sensing capability, which can provide a guideline to improve node weight estimation error caused by long-tail data. Then, we propose a truth discovery-based method in WSNs via incorporating node-target distance into the truth discovery optimization solution framework to infer the sensor node’s fusion weight. In order to adapt to the distribution characteristic of the WSNs, we propose a distributed implementation to estimate the sensor nodes’ weights via confidence level and node-target distance to further improve the fusion performance. Besides, the iterative process shown in CAIF converges to a stationary point of the optimization problem and its time complexity is linear with respect to the total number of observations. Finally, we conduct extensive experiments on real data to validate and evaluate CAIF. The experimental results demonstrate the superior performance of our method over existing solutions in terms of root-mean-square error and accuracy.