Resumen de Dynamic stability with artificial intelligence in smart grids
Environmental concerns are among the main drives of the energy transition in power systems. Smart grids are the natural evolution of power systems to become more efficient and sustainable. This modernization coincides with the vast and wide integration of energy generation and storage systems dependent on power electronics. At the same time, the low inertia power electronics, introduce new challenges in power system dynamics. In fact, the synchronisation capabilities of power systems are threatened by the emergence of new oscillations and the displacement of conventional solutions for ensuring the stability of power systems. This necessitates an equal modernization of the methods to maintain the rotor angle stability in the future smart grids. The applications of artificial intelligence in power systems are constantly increasing. The thesis reviews the most relevant works for monitoring, predicting, and controlling the rotor angle stability of power systems and presents a novel controller for power oscillation damping. Beside the type of the contingency, the main factors affecting the stability of the system are its operating conditions. However, conventional methods for identifying the characteristics of the system stability are either unsuitable for online applications or too slow for control purposes. In this thesis, the relationship between these factors and the oscillatory characteristics is modelled using artificial intelligence methods, such as ensembles and deep neural networks. The predicted oscillation frequency can then be accurately estimated as the system conditions vary. In addition, two architectures of univariate and multivariate regression are compared with respect to their errors in estimating the frequencies of multiple oscillations. From the analysis presented a trade-off between accuracy and speed appears that depending on the application one or the other approach can be used. The intelligent Power Oscillation Damper (iPOD) proposed in this thesis integrates the modelling power of artificial intelligence (superficially the Random Forest ensemble) and the ability of power converters to emulate the behaviour of a synchronous generator. The iPOD can attenuate an underdamped oscillation while adapting in the changing operating conditions. The multi-band iPOD (MiPOD) extends the controller to target more than one oscillation by emulating three rotors. In both cases, however, the main idea is that the oscillation frequencies are treated as a known parameter, hence making this implementation possible. The effectiveness and simplicity of the proposed controllers is demonstrated through a series of simulations for different types of contingencies. To demonstrate the adaptability of the iPOD, in each simulation the operating conditions vary randomly. Following the above rationale, this thesis presents the development of a deep neural network to monitor the electromechanical interactions of a gas-turbine power plant for a system in Europe for the FLEXITRANSTORE project. The trained neural network is embedded into a controller to adaptively update the parameters of the PSS device installed in the power plant. The neural network is trained using measurements within the area on influence of the power plant to predict the oscillation frequency to compute the parameters for the phase compensation. The applications of artificial intelligence are restricted to simulation platforms; therefore, this project is a unique opportunity to obtain invaluable results about the effectiveness, limitations, performance of the application of neural networks in real life conditions.
