Persons

An experience of introducing the measuring and information technologies and simulation modeling into the bench tests of various space industry products has been presented. An example of using the neural network simulation modeling potential to predict the pressure in a turbo pump depending on the flow rate and the polymer additive concentration has been considered. The stages of the neural network model implementation as well as the user interface description have been presented. Using the model the optimal concentration of the polymer additive, permitting to increase the efficiency of the liquid-propellant fuel pump, has been refined.

• Counter: 1491 | Comments : 0

The article considers the algorithm of a representative samples formation for training the neural network of the multilayer perceptron. The purpose of the algorithm under consideration is the training set formation from the factor space and neuron characteristic calculation with application of the Lipschitz constant estimation. In the paper the application of the algorithm to the training set formation has been shown. The algorithm allows to form a training set with the entropy value more than when it is formed randomly, as well as to estimate the characteristics of the neuron from below and refine its form. The approach under consideration was concluded to have an influence on the increase in the entropy of the training set by 0,14 bit, decrease of mean square error by the value of 0,098 and (as a result) to lead to the quality improvement of training of multilayer perceptron with the small dimensionality of the factor space.

• Counter: 1976 | Comments : 0