Reliable and early detection of turbine unit faults is an important task to improve the reliability of power generation, especially today, when the demand for electricity is increasing. The predictive diagnostics system is proposed as a decision-making tool for identifying faults at an early stage and predicting their occurrence and progression. The study presents one of the system functions using a hybrid approach to anomaly detection and localization (FDD), based on statistical signal modeling, expert rules and probabilistic classifiers. This approach is based on the principles of ensemble learning of basic regression models, constructed using a fully connected feedforward neural network, generating diagnostic features and detecting faults through voting classifiers with weighted average probability calculations of input features, and forming a decision rule utilizing expert knowledge. The study elucidates the peculiarities and limitations in diagnostics of turbine units, approaches to the selection of predictor variables and formation of training dataset. Using this method enables for flexible and versatile tuning of algorithms and solving FDD problems under conditions of limited training dataset. The results of testing the algorithm for two instances of turbine unit journal bearing faults are presented. It is demonstrated that the proposed model enabled to identificate of the progression of an anomaly situation long well before the transgression of control threshold levels, whereas personnel and standard monitoring instruments overlooked these events.

The research introduces the application of an algorithm for optimal correction of the parameters of the control structure of a synchronous generator with a diode brushless exciter. The use of this algorithm allows you to solve the problem of maximum performance in real time. One of the difficulties of applying the optimal control algorithm is the high computational costs. In this paper, the application of the developed algorithm for mathematical models of a synchronous generator of the second and fourth order is investigated. The analysis of the algorithm for various models is carried out in terms of comparing the computational costs and the accuracy of the solution. The results of modeling the optimal control algorithm on a laboratory installation are presented. Recommendations are given on the choice of a control finding approach depending on the task formulation.

A problem of generator excitation control is considered. It is proposed to compare currently used algorithms with the algorithm of optimal correction of control structure parameters. A conclusion of the optimal control theory is provided that shows the feasibility of using the developed algorithm. For a mathematical model consisting of second-order differential equitation, the results of numerical modeling are given. The developed algorithm has been tested in a laboratory unit. Efficiency of using optimal control and high convergence of the proposed method for the considered problem have been shown.

An algorithm for correcting the control structure of a synchronous generator with a diode brushless exciter obtained using the maximum principle is considered. This algorithm is applied to a simplified mathematical model consisting of a system of differential equations of the second order. The obtained control is applied to the simulation mathematical model. The results of numerical simulation showing high convergence of the considered algorithm are presented.