Filling the gaps in the input and output data using the algorithm of nonparametric identification

The task of identifying systems, that is, determining the structure and parameters of systems from observations, is one of the main tasks of a modern theory and technology of automatic control. The accuracy of solving the identifica- tion problem directly depends on the quality of the initial data (sample of observations). However, the data may contain various shortcomings, in particular, gaps. Gaps in the data are due to a variety of reasons, such as inability to observe, lack of necessary tools, and so on. The easiest method of working with such data is to exclude from the table an indicator (column) or an object (line) with a space. With a large number of gaps in the data, this approach leads to a reduction in the accuracy of the model due to a reduction in the sample size. It is important to note that in the described case the complexity of solving the identification problem increases, especially when the density of passes is high, their location is irregular, and the data is insufficient (very little). The aim of the paper is to improve the accuracy of solving the problem of identifying discrete-continuous multidi- mensional processes from samples of observations with gaps. To achieve this goal, methods of mathematical statistics, data analysis, and mathematical modelings were used. In the article the algorithm of a non-parametric estimation of the regression curve in a discrete-continuous process in the task of filling out the admissions of the observation matrix is described. Moreover, a model is built based on this algorithm. Two computational experiments were carried out. The first experiment was conducted in the presence of gaps in the output variable matrix of observations. The second experiment was conducted with gaps in the input variables. The experiments were conducted at different sample sizes. Based on the results of the algorithm under vari- ous conditions, conclusions are given. The results of the work can be useful in creating control systems for multidimensional discrete-continuous processes.