GSCV-XGBoost based information reconstruction and fusion method for SINS/DVL integrated navigation system

Abstract

An underwater navigation system has specific requirements for reliability. In the frequently used strapdown inertial navigation system/Doppler velocity log (SINS/DVL) integrated navigation system, the process of the DVL measuring velocity is always disturbed in a complex underwater environment. Thus the velocity measurements on acoustic beam directions are prone to get lost, which seriously affects system reliability. This paper proposes a novel information reconstruction and integration algorithm to solve this problem, based on the tree boosting model. The algorithm is mainly divided into two consecutive stages: training and reconstruction. When the DVL velocity in all beam directions is available, the GSCV-XGBoost (extreme gradient boosting based on the grid-search and cross-validation theory) method is used to train the mapping model between DVL velocity and SINS/DVL integrated navigation system velocity. When the DVL velocity is incompletely available, the mapping model, that has been trained in advance, is used to reconstruct the missing DVL velocity measurements and resume normal SINS/DVL integration conditions. Simulation and underwater vehicle-mounted experiments are applied to verify that the proposed algorithm reconstructs more accurate velocity information than traditional XGBoost, which shows great prospects for application and reliability in the SINS/DVL underwater integrated navigation system.