We used the approach based on the auto-associative neural networks of direct distribution. The purpose of training of the network is the most accurate reproduction of the input data in the output. If the input data contains gaps, an error of the training may be significant. Then we make changes of the value of missed elements and the training repeats. When the error does not exceed the threshold, the matrix of the input data without gaps can be saved as a part of the complete database.
training of the network based on the model of geometric transformations;
high speed with installing the linear or non-linear relations between the input data per iteration;
prompt implementation of the re-training when there is a new input data or changes in values of the missed elements.
reduction of impact of the absence, incomplete or irrelevance of measurements and observations data on the conclusions and decision-making in the engineering, natural and economic systems.
sale of the license, joint refinement of the development to the industrial level.