Dead reckoning error compensation algorithm of AUV based on SVM

A precise submersing navigation position is always the target pursued by underwater technology. The conventional dead reckoning (DR) algorithm normally uses Kalman filter to estimate positional errors. In the present study, soft computing-based models, such as Artificial Neural Network (ANN), have been used to compensate the reckoning error. The problem with ANN is the inevitable local extremum phenomenon. The evaluated Support Vector Machine (SVM) which is a strong machine learning and data mining tool can be guaranteed to the global optimal value. This article proposes an algorithm to evaluate the error of DR by employing SVM. Then navigation system compensates the elevated DR error to DR system. Attitude angles, velocities and the relative time are given as input variables, while DR error is the output parameter. The longitude predictive error model and the latitude predictive error model are introduced. The algorithm is validated by Institute of Marine Equipment and Control of Harbin Engineering University in the lake trial of Erlongshan in China.