Synthesis of Centralized Adaptive Excitation Control System by Reinforcement Learning Approach

The paper is aimed at the study of existing approaches to creation of centralized PSS based on artificial neural networks, as well as the synthesis of such control systems and assessment of their effectiveness in four-machine Kundur scheme. Existing approaches applying adaptive and neural net the technologies in creating control systems for large power interconnections are analyzed. Test data sets for training a neural network using traditional excitation regulators (AVR+PSS) are created in a four-machine Kundur power system based on optimal parameters obtained for various operating conditions. Local and global adaptive control systems based on artificial neural networks and reinforcement learning are obtained and effectiveness of proposed approaches is evaluated by calculating transients with various disturbances in power system.