Three-Dimensional Mobile Node Localization Method of WSNs Based on Improved LSSVR Algorithm

The traditional least squares support vector regression (LSSVR) node localization algorithm for wireless sensor networks (WSNs) uses the average hop distance to calculate the actual distance, which may result larger localization error in the obstacle conditions. An improved LSSVR WSNs three-dimensional mobile node localization method in an obstacle conditions was proposed in this paper. The average per hop distance of four anchor nodes closest was used to replace the average distance per hop of traditional LSSVR algorithm in the proposed method, and the new average per hop distance was used to calculate the measurement distance of each unknown node to anchor nodes. The LSSVR localization model was built through sampling of the grid and constructing the training sets. According to mean square deviation of predicted location of virtual nodes and their actual location, fitness function was constructed, and LSSVR kernel function and regularization parameters were optimized by the PSO algorithm. The simulation results show that, compared with the conventional LSSVR localization algorithm, the proposed localization algorithm has a higher localization accuracy, smaller localization errors and lower localization cost in the obstacle conditions.