Environmental problem arising from the global warming are motivating evolution of the power infrastructure by increasing the deployment of variable renewable energy technologies such as wind turbines and solar panels. This approach holds the promise of reducing CO2 emissions and facilitating a transition toward a sustainable future. On the other hand, the massive integration of variable renewable energy sources has led to a decline in the overall rotational inertia in the power grid, making it more susceptible and reliant on meteorological fluctuations. To counterbalance the lack of mechanical inertia, certain researchers have proposed emulating virtual inertia. The effective implementation of this concept necessitates development of a more robust and flexible control algorithms. This thesis proposes a hybrid solution that amalgamates the advantages of both paradigms – AI and classical PID controller. Specifically, it combines the adaptability inherent to neural networks with the robustness characteristic of classical controllers. Supported by deep reinforcement learning techniques, the proposed controller can acquire insights into system dynamics through a data-driven approach and can even be employed for coordinated or multi-node control strategies. Computational intelligence harbors considerable potential for emerging power systems in which renewable energy sources are dominant.

The thesis is published in the Digital Collection of TalTech Library.

Supervisors:

• prof Juri Belikov
• co-supervisor:  prof Eduard Petlenkov

Opponents:

• prof Gaber Magdy, Aswan University, Aswan, Egypt
• prof João Martins, Nova University of Lisbon, Caparica, Portugal

Follow public defense in Zoom

Meeting ID: 970 1118 1084
Passcode: 923503