The increasing capabilities of software and hardware complexes of control systems in the energy sector lead to a transition to predictive diagnostic systems that ensure the implementation of maintenance and repair of equipment based on its condition. When solving the problem of developing such systems, the technology of digital twins is widely used. This chapter proposes the use of power plants thermal circuits auxiliary equipment digital twins in assessing and predicting their technical condition based on ensembles of machine learning methods for use in predictive diagnostic systems. To solve this problem, two ways of using machine learning in the development of predictive diagnostic systems based on digital twins are considered, within which the specified algorithms are used to create an equipment model that simulates operation without defects, or to determine the current technical condition of equipment based on training sample obtained from the digital twin. A mathematical model of contamination of low-pressure heater tubes has been developed, on the basis of which a sample is created for training algorithms. A comparative analysis of various machine learning algorithms, such as artificial neural networks, Random Forest and Gradient Boosting, was carried out. The efficiency of the specified algorithms was calculated. The best results were shown by the Random Forest and Gradient Boosting algorithms, while the use of the Gradient Boosting algorithm is more preferable when solving problems of predicting the technical condition of heat exchange equipment of thermal power plants.