Al Rawashdeh L. Improvement of methods and instruments for measurement of satellite navigation systems parameters based on artificial neural networks

The dissertation is devoted to solving an urgent scientific problem to improve the accuracy of measurements of the spatial coordinates of moving transport objects using satellite navigation systems based on artificial neural networks. A method for measuring the parameters of a dynamic nonlinear system in conditions of an incomplete measured state vector based on modeling the investigated system using dynamic neural networks is developed. It is shown that the effective architecture of a neural network is a recurrent network with nonlinear activation functions and input delays. The dependence of the measurement accuracy on the number of directly measured components of the state vector and on the length of the input delay line is investigated. Experimental substantiation of the highest efficiency of the neural network learning algorithm based on the theory of Kalman filters for solving the problem of modeling the studied systems is proposed. The application of Kalman’s algorithm to the problem of measurements with an incomplete state vector showed the advantage of this algorithm over other learning algorithms for neural networks. A method for correcting learning parameters in a neural network learning algorithm based on the theory of Kalman filters is proposed, which allows increasing the convergence rate of the algorithm. A method for evaluating errors in measuring systems that use dynamic neural networks has been developed. A method for estimating the error in the output signal of a standard neuron, expressed in terms of the standard deviation, is proposed. The proposed method is extended to evaluate the errors of the neural network output signals with known errors of the input signals in real-time. The method can be applied to feedforward networks and recurrent networks that use standard neurons. The main problems arising in determining the location of mobile transport objects using satellite navigation systems are analyzed, and ways to improve the accuracy for this measurement problem are considered. An experimental study of a system for measuring the spatial coordinates of railway transport using satellite navigation systems, in which secondary information processing (correction of estimates of the navigation coordinates of a vehicle) was carried out by a neural network, was carried out. The experimental results showed the effectiveness of the proposed approach and reduction error in determining the spatial coordinates of moving objects after correction.