The paper considers the problems of identifying non-linear stochastic systems and its application to a specific task of the nuclear power plant control. The main attention is paid to the application of consistent (what means vanishing if and only if two random (in general) vectors are stochastically independent) measures of dependence. There are presented particular important cases concerned with applying Tsallis divergence as a basic tool to construct a corresponding measure of dependence leading, finally, to obtaining a total input/output system model by the use of local ones.