This article explores the development and application of methods for calculating technical and economic indicators (TEI) of nuclear power plant (NPP) units using stochastic approaches to system identification and predictive models based on associative search. The relevance of the study stems from the need to improve the efficiency and reliability of TEI estimation under the operating conditions of automated process control systems at the upper unit level, as well as to ensure correct calculations in the event of incomplete or unreliable initial data. The article presents a methodology that combines classical TEI estimation schemes with predictive models based on associative search methods, enabling the identification of hidden dependencies between the process parameters of an NPP unit and the stable calculation of indicators in the absence of individual measurement signals. The article examines the identifiability conditions for stochastic systems using consistent measures of random variable dependence, specifically the symmetric quadratic Tsallis mutual information. A method for selecting input variables for local models is proposed based on the normalized values of this measure and a criterion for the heterogeneity of local model contributions. The results of testing the proposed approach, illustrated by the example of calculating the feedwater temperature behind a high-pressure heater, are presented, demonstrating acceptable predictive accuracy with a limited set of key NPP unit signals. The prospects for the practical application of the developed methods for information support for NPP operating personnel, predictive maintenance, and ensuring safe operation are discussed.