Abstract: Position information is nonlinear in the node localization of wireless sensor networks (WSN). Based on the robust ability of multivariate linear regression of partial least squares (PLS), and in combination with nonlinear data dimension reduction of manifold learning, a novel kernel matrix ISOMAP (Isometric Feature Mapping) algorithm is proposed. Geodesic distances between nodes are used as a measure of dissimilarity, and the contribution rate is then used to find and delete the "short circuit" edge. The matrix constructed by a double-centered transformation and the kernel transformation trick is mapped to a high dimensional feature space; finally, the relative position is obtained by PLS. Compared with the traditional ISOMAP algorithm and the multidimensional scale method (MDS), simulation results indicate that the proposed algorithm has good topology stability, generalization properties, robustness, positioning accuracy, and lower computational complexity.

Key words: wireless sensor network, node localization, kernel matrix, Isometric Feature Mapping (ISOMAP)