Improved Input-Delayed Dynamic Neural Network for GPS/INS Integration

At present, real time navigation systems rely on kalman filtering for integrating global positioning system (GPS) and inertial navigation system (INS) in order to provide feasible solution for reliable navigation. But there exist some inadequacies related to the stochastic error models of inertial sensors, requirement of prior knowledge in fusing data and long design time. Moreover, recently artificial intelligence (AI) techniques based on static neural networks were suggested in place of kalman filter. But there also exist drawbacks like inadequate training during GPS absence, less reliability and more training time. To solve this problem, a new dynamic architecture based on Input-delayed model has been developed. This paper, therefore proposes a scheme based on Improved Input-delayed dynamic neural network (IIDDNN) to model the samples of INS values with respect to GPS values as target vectors (Normal mode). This model is capable of providing a reliable and accurate solution especially for GPS absence (prediction mode). The prediction mode output obtained using IIDDNN is compared with the performance of normal mode of IIDDNN. The proposed IIDDNN model is evaluated using test data of fixed trajectory path for an aircraft vehicle.