wireless sensor networks on artificial neural networks 

Localization of nodes in wireless sensor networks without the use of GPS is important for applications such as military surveillance, environmental monitoring, robotics, domotics, animal tracking, and many others. Low cost and energy efficient sensors require methods that compute their position using indirect information such as RSSI (Received Signal Strength Indicator). This work presents an artificial neural networks (ANNs) approach to localization in wireless sensor networks through the adjustment of the ANNs structures using Genetic Algorithms. A population of feedforward ANNs containing their structure in a genetic code is evolved during 20 generations. Each individual is evaluated through the training of the artificial neural network and further calculation of its root mean square error for all the testing set.

The RSSI measurements were used as the artificial neural networks inputs to localize the nodes. The approach was tested using the MATLAB-based Probabilistic Wireless Network Simulator (Prowler) to collect the artificial neural networks input data, under simulated static indoor network environment of 26×26 meters with 8 anchor nodes, i.e., nodes with awareness of their positions. The MATLAB’s genetic algorithms and artificial neural networks toolboxes were used. Results using the best artificial neural network structure found after optimization had a root mean square error of 0.41 meters, a maximum error of 1.07 meters and a minimum error of 0.014 meters.