Velocity Model Identification For An AUV Navigation With Using NARX Neural Network Method

One of the challenges of the autonomous underwater vehicles (AUV) navigation is measuring their velocity. The usual method for measuring the velocity of AUV is to use a Doppler Velocity Logger (DVL), but it is not possible to use this sensor due to its placement in the category of expensive sensors, as well as the increase in time and even the lack of data collection due to high depth or sudden changes in depth in some cases. The aim of this research is to provide a cheap and economical method of speed identification based on an autoregressive exogenous (NARX) neural network with the least number of neural network inputs in 2-D floating motion. In the proposed algorithm, by removing the inputs of the neural network obtained from the output of low-cost sensors, the measurement error of the sensors is removed from the identification process and the accuracy of the velocity model output is improved. The proper performance of the proposed algorithm, compared to the output of the DVL and also the output obtained from the differential model identification method with the help of Least Square (LS) and Recursive Least Square (RLS) algorithms, confirms the advantage and efficiency of this method in identifying the velocity of AUV.