The Enhancement of the Integration of Cost-effective Satellite Inertial Navigation Systems Using the GMDH Neural Network

Nowadays, integrating data of cost-effective satellite inertial navigation systems has become conventional for the sake of increasing accuracy. Self-sufficiency, high rate of data specification, rotational data presentation and, of course, reduced accuracy over time in the inertial navigation systems on the one hand, and the low rate of data, the lack of presenting rotational data, and the obstruction or disruption of the receipt of the GNSS data on the other hand, has made this approach increasingly successful. Furthermore, the integrated approaches based on Kalman Filter have other limitations such as reliance on the model, the need to have prior knowledge, linearization, and most importantly, efficiency reduction at the obstruction of the GNSS. The purpose of this paper is to introduce an intelligent, highly reliable online method maintains its performance at the obstruction or disruption of the GNSS. Simulation results with GMDH neural network and comparison with the conventional Kalman Filter and MLP neural network methods reveal that GMDH, because of high speed and capability in approximation and correction of INS errors (due to simple structure and the elimination of inactive neurons through an effective method for training), can be used at online navigation operations and for the inevitable conditions of lacking the GNSS data.