Identification quality assessment of the complex object management in multicollinearity conditions

Abstract

The article presents the method of assessing the quality of the complex object management (COM) identification process in multicollinearity conditions. The main problem in the process of identifying complex object management is assessing its quality. When operating COM, multicollinearity often arises, which significantly complicates the task. Nowadays, a lot of evaluation methods have been developed and studied. Despite this, when solving practical tasks related to processing data from experiments about the management object, the accuracy of the MNS becomes insufficient. As a criterion for the optimality of the regression model, it is proposed to use the amount of forecast error in a given area. A method for finding the optimal regularization parameter for offset estimation of regression equation parameters is proposed. A scientific substantiation of the principles of assessing the quality of the COM identification process in multicollinearity conditions are resolved by building its predictive model using the method of offset assessment of regression equation parameters. In this case, a method is proposed for selecting the regularization parameter r based on the minimum root-mean-square error of the forecast obtained from this model. The simulation experiments using the technique, shown that the value of the optimal regularization parameter ropt obtained is close to the experimental one. This confirms the correctness of the proposed approach. Thus, the quality of the predictive model under conditions of multicollinearity has been improved, taking into account the uncertainty of the model structure and the method of biased estimation of model parameters.