Industries, especially in the energy sector, are using digital twins to improve work efficiency and optimize operating modes. Digital twins have based on models that accurately describe the geometry, physical properties, behavior, and rules that characterize an object. The article presents an example of functional decentralization of digital twin models as a decomposition of an NPP power unit (PU) with a VVER-1000 reactor as a control object into a set of technological subsystems of functional groups. Approaches to the digital twins’ creation of NPP PU and the main directions for using digital twins based on dynamic models of a PU in advanced control systems for NPP power units are presented. Within the framework of the intelligent operator support system, a "three-stage" approach to the problem of forecasting the state of the NPP PU, based on digital twins, and the principles of forming the PU optimal control are proposed. The proposed approaches to creating digital twins are used to create intelligent support systems for operators.