Improving DGPS Accuracy using Neural Network Modeling

Back-Propagation (BP), Extended Kalman Filter (EKF) and Particle Swarm Optimization (PSO) are three of the most widely used algorithms for training feed forward Neural Networks (NNs). This paper presents an accurate DGPS land vehicle navigation system using multi-layered NNs based on the BP, EKF and PSO learning algorithms. The network setup is developed based on mathematical models to avoid excessive training. The proposed methods use BP, EKF and PSO training rule, which achieves the optimal training criterion. The NNs predict the Differential GPS (DGPS) corrections for accurate positioning. The proposed methods are suitable for DGPS systems sampled at different rates. The experimental results on measured data demonstrate the suitability of these methods for accurate prediction of DGPS corrections. The experiments show that the prediction total RMS errors using NN based on PSO learning algorithm are 1.03 metre and 0.65 metre, before and after SA, respectively.