Abstract
This paper examines the problem of filtering or dividing a real table of observations into possible useful data and observation errors called noise. The research includes the development and justification of useful information models and noise models for each observation, along with selection of optimality criterion and a mathematical optimization model. The filtering problem is solved while modeling the linear process with its variables observed with interval errors. There are estimated coupling coefficients and tables of true values of process variables obtained when filtering. It is shown that the number of estimates is greater than the number of connections. Therefore, it is impossible to accurately separate useful information from noise. Following the problem description, the task of optimal filtering is to ensure that the specified separation is the maximum possible. Filtering algorithms are developed using systems of linear interval equations. Computer simulations of the filtration processes and multi-variant computational experiments are performed using the Microsoft Excel software product.