DVL Modeling Using a RVM based on Artificial Bee Colony Algorithm in SINS/DVL

For underwater Strapdown Inertial Navigation System/Doppler Velocity Log(SINS/DVL) integrated navigation system, it is a crucial factor for the performance of integrated navigation system to estimate the scale factor of DVL and its installation relationship with SINS accurately. Firstly, the theoretical model of DVL influenced by scale factor error, installation error between SINS and DVL and dynamic attitude are established. Due to the effect of dynamic attitude angle, the velocity measurement model of DVL presents great complexity and strong nonlinearity. Hence, this paper proposes a novel DVL model training approach using relevance vector machine (RVM), in which all error sources are considered in this model without separated DVL velocity measurement models. The Artificial Bee Colony (ABC) algorithm is used to optimize the key parameters of RVM. Secondly, the constraint information from GNSS and depth sensor is applied to strengthen the accuracy and generalization of the model. Finally, the performance of this method is verified with experiment in the Yellow Sea. Compared with the traditional least square estimation method based on SVD (SVD-LS), the DVL measurement model shows higher accuracy, and the positioning error of SINS/DVL integrated navigation system has been reduced from 3.1‰ of the voyage to 1.5‰ of the voyage.