Semi-supervised manifold learning based on polynomial mapping for localization in wireless sensor networks

Localization of sensor nodes is becoming an interesting research area in wireless sensor networks(WSNs). In recent years, some localization algorithms are presented based on manifold learning which can achieve good performance. In this paper, based on the need of practical application and the development of manifold learning theory in WSNs, the localization problem is considered for fast determining the locations of unknown nodes in large scale WSNs. For this purpose, a new semi-supervised manifold learning algorithm is proposed based on polynomial mapping which is the combination of Gaussian kernel embedding and polynomial kernel embedding algorithms. This method is to compute a polynomial mapping function between the high dimensional location data space and the low dimensional physical space, which can obtain an explicit nonlinear feature mapping and has a high discriminative ability. At last, comparing with some related localization approaches, the performance of the proposed algorithm is evaluated and analyzed under different signal noise, anchor nodes, and communication range, respectively. In terms of the root-mean-square error and computational complexity, the experiment results demonstrate that the proposed algorithm has faster speed for large scale sensor nodes and higher accuracy with a small amount of known nodes. (C) 2020 Published by Elsevier B.V.