A Localization Method Based on Detection of Beacon Node Drift in Wireless Sensor Networks

A distributed detection method for beacon node drift is proposed to address the node localization problem concerning beacon node drift. This method automatically identifies possible drift beacon nodes by calculating the degree of variation in RSSI similarity at different times. A localization algorithm combining Kernel Principal Component Analysis (KPCA) and PSO-BP neural network is also proposed to solve the high dimensionality problem concerning localization data. First, KPCA is used to eliminate data dependency, extract the principal component containing localization information and reduce dimensionality of sample space. Then the PSO-BP neural network is trained using the extracted eigenvectors of nonlinear principal component as input sample and position coordinate of grid vertex as output sample to create a localization model. The simulation result shows that this algorithm is highly practical and outperforms traditional method in terms of drift detection and localization error.