Technologies for Tuning Neural Networks in Problems of Inertial Data Correction in Navigation Systems for Low-Speed Autonomous Objects

摘要

At present, when satellite navigation aids are widely available, inertial navigation systems still remain relevant where it is required to determine not only the coordinates of the object, but also its orientation in space; where the reception of satellite navigation signals is impossible, as well as in cases where the navigation solution should not depend on the operability of navigation satellites. However, due to the low accuracy of inertial sensors, provided by the presence of systematic and random errors, the characteristics of navigation systems based on them are clearly insufficient for continuous autonomous movement. An inertial navigation algorithm is beingdeveloped for low-speed autonomous objects. As an estimate of the coordinates of the objects, a Kalman recursive filter is used, in this case – an extended one. The computational cycle of the algorithm consists of two steps. At the first step, an estimate of the current errors of the navigation system in determining the coordinates, speeds and orientation angles is calculated. At the second step, the correction of the basic navigation solution is carried out on the basis of the amendments received. A neural network is used to filter primary inertial information. At the input of the neural network, acceleration and angular velocity vectors are supplied, and their filtered values are formed at the output. The neural network is configured for filtering by training it for a sequence of inertial data, corresponding to the passage along certain trajectories.