To the problem of generation sample in solving the problem of non interial processes identification

The problem of improving the quality of the original data in the identification of H-processes is considered. There is stochastic dependence between input variables of such processes so this process precedes not at all the regulated area but only at some of its subdomain. This fact leads to some features that must be considered when identifying. If there is large number of a priori data when building models, you can use the methods of identification in the “broad” sense. However, if there is sufficient a priori information about the object being studied, it is necessary to apply the methods of identification in the “narrow” sense. These methods include nonparametric estimation of regression function from observations. The quality of solving the problem of identification depends on the quality of input data. It is advisable to conduct a preliminary analysis of data to identify and address all the deficiencies in the sample. Under the preliminary analysis of the data is taken to mean filling gaps in observations and eliminating emissions. However, the sample may have other flaws, that will be discussed below, which adversely affect the accuracy of estimation, and, in some cases, lead to the fact that the resulting model will be inadequate to the investigated process. If the point of the original sample in the field of process located patchy, there are low-pressure range and lack of observations, in the areas of reconstruction accuracy is low. Due to the properties of nonparametric estimation, which belongs to the class of local approximations, projections can not be given at the lack of observations subdomain. To resolve all these shortcomings we propose an algorithm to obtain a working sample by generating new points in regions where the density is low in comparison with other areas. After generating new working sample, the quality of estimation is significantly improved, as evidenced by the results of numerical experiments. This kind of tasks are relevant and can be used in solving the problem of recognition in various fields, where the classification accuracy is important, for example, in the diagnosis of space propulsion, electrical, etc.