Ovcharova T. Improvement of models and methods of measurement information processing using artificial neural networks

The object - the process of measuring physical quantities by means of measurements with a nonlinear transformation function. The goal - improving measurement accuracy by improving the models of non-linear transducers based on the use of artificial neural networks. Theoretical and practical results - improvement of models and methods of measurement information processing using artificial neural networks, developed software complex neural network models and algorithms for processing measurement information. Methods - Mathematical and computer simulation, mathematical statistics, gradient optimization methods, the theory of artificial neural networks and methods of the theory of uncertainty. Scientific novelty - first proposed neural network model equation converter based three-layer perceptron, which allowed to obtain estimates of the combined standard and expanded uncertainty of measurement of the output signal of non-linear neural network model of measuring instruments; An improved method for compensating nonlinear static and dynamic transducers using inverse neural network models based on the three-layer perceptron that significantly improve metrological characteristics of measuring instruments; further developed neural network models of nonlinear static and dynamic measurement equipment on the basis of a three-layer perceptron, which is the advantage of the invariance of the form of the nonlinear function conversion object modeling and the ability to synthesize these models through training. The results are implemented: inLimited Liability Company "Promgidroprivid" (Kharkov), in the Limited Liability Company "Scientific and Production Association Аgrotechnics" (Kharkov), in the educational process in Kharkov National University of Radioelectronics.