Application of BP neural network to 3D localization in wireless sensor networks

To reduce the impact of the multi-hop distance estimation errors on the performance of localization in 3D wireless sensor networks, a BP neural network based localization (BPL) algorithm is proposed. The BPL algorithm constructs a BP neural network model to correct the multi-hop estimative distances of non-adjacent nodes and extracts the training samples of the BP network according to the location relationships between pairwise anchor nodes. After a training procedure, the BP neural network can reflect the main property of global space structure of 3D wireless sensor networks. Each unknown node uses the trained BP network to estimate its Euclidean distances to anchor nodes within the range of a certain hop count and then calculates its 3D coordinates. Numerous simulations show that the multi-hop estimative distances and the localization accuracy can be effectively improved by the BPL algorithm, especially in the sparse networks.